Patch release: v0.21.1

* [Import] Add missing settings

* up

* up

* up

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -13,9 +13,8 @@ body:
              *Give your issue a fitting title. Assume that someone which very limited knowledge of diffusers can understand your issue. Add links to the source code, documentation other issues, pull requests etc...*
         - 2. If your issue is about something not working, **always** provide a reproducible code snippet. The reader should be able to reproduce your issue by **only copy-pasting your code snippet into a Python shell**.
              *The community cannot solve your issue if it cannot reproduce it. If your bug is related to training, add your training script and make everything needed to train public. Otherwise, just add a simple Python code snippet.*
-        - 3. Add the **minimum** amount of code / context that is needed to understand, reproduce your issue.
+        - 3. Add the **minimum amount of code / context that is needed to understand, reproduce your issue**.
              *Make the life of maintainers easy. `diffusers` is getting many issues every day. Make sure your issue is about one bug and one bug only. Make sure you add only the context, code needed to understand your issues - nothing more. Generally, every issue is a way of documenting this library, try to make it a good documentation entry.*
-        - 4. For issues related to community pipelines (i.e., the pipelines located in the `examples/community` folder), please tag the author of the pipeline in your issue thread as those pipelines are not maintained.
   - type: markdown
     attributes:
       value: |
@@ -61,46 +60,21 @@ body:
         All issues are read by one of the core maintainers, so if you don't know who to tag, just leave this blank and
         a core maintainer will ping the right person.
         
-        Please tag a maximum of 2 people.
+        Please tag fewer than 3 people.
+        
+        General library related questions: @patrickvonplaten and @sayakpaul
 
-        Questions on DiffusionPipeline (Saving, Loading, From pretrained, ...):
+        Questions on the training examples: @williamberman, @sayakpaul, @yiyixuxu
 
-        Questions on pipelines:
-        - Stable Diffusion @yiyixuxu @DN6 @patrickvonplaten @sayakpaul @patrickvonplaten
-        - Stable Diffusion XL @yiyixuxu @sayakpaul @DN6 @patrickvonplaten
-        - Kandinsky @yiyixuxu @patrickvonplaten
-        - ControlNet @sayakpaul @yiyixuxu @DN6 @patrickvonplaten
-        - T2I Adapter @sayakpaul @yiyixuxu @DN6 @patrickvonplaten
-        - IF @DN6 @patrickvonplaten
-        - Text-to-Video / Video-to-Video @DN6 @sayakpaul @patrickvonplaten
-        - Wuerstchen @DN6 @patrickvonplaten
-        - Other: @yiyixuxu @DN6
+        Questions on memory optimizations, LoRA, float16, etc.: @williamberman, @patrickvonplaten, and @sayakpaul
 
-        Questions on models:
-        - UNet @DN6 @yiyixuxu @sayakpaul @patrickvonplaten
-        - VAE @sayakpaul @DN6 @yiyixuxu @patrickvonplaten
-        - Transformers/Attention @DN6 @yiyixuxu @sayakpaul @DN6 @patrickvonplaten
+        Questions on schedulers: @patrickvonplaten and @williamberman
 
-        Questions on Schedulers: @yiyixuxu @patrickvonplaten
-
-        Questions on LoRA: @sayakpaul @patrickvonplaten
-
-        Questions on Textual Inversion: @sayakpaul @patrickvonplaten
-
-        Questions on Training: 
-        - DreamBooth @sayakpaul @patrickvonplaten
-        - Text-to-Image Fine-tuning @sayakpaul @patrickvonplaten
-        - Textual Inversion @sayakpaul @patrickvonplaten
-        - ControlNet @sayakpaul @patrickvonplaten
-
-        Questions on Tests: @DN6 @sayakpaul @yiyixuxu 
-
-        Questions on Documentation: @stevhliu
+        Questions on models and pipelines: @patrickvonplaten, @sayakpaul, and @williamberman
 
         Questions on JAX- and MPS-related things: @pcuenca
 
-        Questions on audio pipelines: @DN6 @patrickvonplaten
-        
-
+        Questions on audio pipelines: @patrickvonplaten, @kashif, and @sanchit-gandhi 
         
+        Documentation: @stevhliu and @yiyixuxu
       placeholder: "@Username ..."

.github/workflows/build_docker_images.yml

@@ -26,8 +26,6 @@ jobs:
         image-name:
           - diffusers-pytorch-cpu
           - diffusers-pytorch-cuda
-          - diffusers-pytorch-compile-cuda
-          - diffusers-pytorch-xformers-cuda
           - diffusers-flax-cpu
           - diffusers-flax-tpu
           - diffusers-onnxruntime-cpu

.github/workflows/pr_dependency_test.yml

@@ -20,7 +20,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: "3.8"
+          python-version: "3.7"
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip

.github/workflows/pr_quality.yml

@@ -20,7 +20,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: "3.8"
+          python-version: "3.7"
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
@@ -38,7 +38,7 @@ jobs:
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: "3.8"
+          python-version: "3.7"
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip

.github/workflows/pr_test_peft_backend.yml

@@ -1,67 +0,0 @@
-name: Fast tests for PRs - PEFT backend
-
-on:
-  pull_request:
-    branches:
-      - main
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  DIFFUSERS_IS_CI: yes
-  OMP_NUM_THREADS: 4
-  MKL_NUM_THREADS: 4
-  PYTEST_TIMEOUT: 60
-
-jobs:
-  run_fast_tests:
-    strategy:
-      fail-fast: false
-      matrix:
-        config:
-          - name: LoRA
-            framework: lora
-            runner: docker-cpu
-            image: diffusers/diffusers-pytorch-cpu
-            report: torch_cpu_lora
-
-
-    name: ${{ matrix.config.name }}
-
-    runs-on: ${{ matrix.config.runner }}
-
-    container:
-      image: ${{ matrix.config.image }}
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
-
-    defaults:
-      run:
-        shell: bash
-
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: Install dependencies
-      run: |
-        apt-get update && apt-get install libsndfile1-dev libgl1 -y
-        python -m pip install -e .[quality,test]
-        python -m pip install git+https://github.com/huggingface/accelerate.git
-        python -m pip install -U git+https://github.com/huggingface/transformers.git
-        python -m pip install -U git+https://github.com/huggingface/peft.git
-
-    - name: Environment
-      run: |
-        python utils/print_env.py
-
-    - name: Run fast PyTorch LoRA CPU tests with PEFT backend
-      if: ${{ matrix.config.framework == 'lora' }}
-      run: |
-        python -m pytest -n 2 --max-worker-restart=0 --dist=loadfile \
-          -s -v \
-          --make-reports=tests_${{ matrix.config.report }} \
-          tests/lora/test_lora_layers_peft.py

.github/workflows/pr_tests.yml

@@ -34,11 +34,6 @@ jobs:
             runner: docker-cpu
             image: diffusers/diffusers-pytorch-cpu
             report: torch_cpu_models_schedulers
-          - name: LoRA
-            framework: lora
-            runner: docker-cpu
-            image: diffusers/diffusers-pytorch-cpu
-            report: torch_cpu_lora
           - name: Fast Flax CPU tests
             framework: flax
             runner: docker-cpu
@@ -94,14 +89,6 @@ jobs:
           --make-reports=tests_${{ matrix.config.report }} \
           tests/models tests/schedulers tests/others
 
-    - name: Run fast PyTorch LoRA CPU tests
-      if: ${{ matrix.config.framework == 'lora' }}
-      run: |
-        python -m pytest -n 2 --max-worker-restart=0 --dist=loadfile \
-          -s -v -k "not Flax and not Onnx and not Dependency" \
-          --make-reports=tests_${{ matrix.config.report }} \
-          tests/lora
-
     - name: Run fast Flax TPU tests
       if: ${{ matrix.config.framework == 'flax' }}
       run: |
@@ -183,4 +170,4 @@ jobs:
       uses: actions/upload-artifact@v2
       with:
         name: pr_${{ matrix.config.report }}_test_reports
-        path: reports
+        path: reports

.github/workflows/push_tests.yml

@@ -1,11 +1,10 @@
-name: Slow Tests on main
+name: Slow tests on main
 
 on:
   push:
     branches:
       - main
 
-
 env:
   DIFFUSERS_IS_CI: yes
   HF_HOME: /mnt/cache
@@ -13,115 +12,53 @@ env:
   MKL_NUM_THREADS: 8
   PYTEST_TIMEOUT: 600
   RUN_SLOW: yes
-  PIPELINE_USAGE_CUTOFF: 50000
 
 jobs:
-  setup_torch_cuda_pipeline_matrix:
-    name: Setup Torch Pipelines CUDA Slow Tests Matrix
-    runs-on: docker-gpu
-    container:
-      image: diffusers/diffusers-pytorch-cpu # this is a CPU image, but we need it to fetch the matrix
-      options: --shm-size "16gb" --ipc host
-    outputs:
-      pipeline_test_matrix: ${{ steps.fetch_pipeline_matrix.outputs.pipeline_test_matrix }}
-    steps:
-      - name: Checkout diffusers
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 2
-      - name: Install dependencies
-        run: |
-          apt-get update && apt-get install libsndfile1-dev libgl1 -y
-          python -m pip install -e .[quality,test]
-          python -m pip install git+https://github.com/huggingface/accelerate.git
-
-      - name: Environment
-        run: |
-          python utils/print_env.py
-
-      - name: Fetch Pipeline Matrix
-        id: fetch_pipeline_matrix
-        run: |
-          matrix=$(python utils/fetch_torch_cuda_pipeline_test_matrix.py)
-          echo $matrix
-          echo "pipeline_test_matrix=$matrix" >> $GITHUB_OUTPUT
-
-      - name: Pipeline Tests Artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: test-pipelines.json
-          path: reports
-
-  torch_pipelines_cuda_tests:
-    name: Torch Pipelines CUDA Slow Tests
-    needs: setup_torch_cuda_pipeline_matrix
+  run_slow_tests:
     strategy:
       fail-fast: false
       max-parallel: 1
       matrix:
-        module: ${{ fromJson(needs.setup_torch_cuda_pipeline_matrix.outputs.pipeline_test_matrix) }}
-    runs-on: docker-gpu
-    container:
-      image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --gpus 0
-    steps:
-      - name: Checkout diffusers
-        uses: actions/checkout@v3
-        with:
-          fetch-depth: 2
-      - name: NVIDIA-SMI
-        run: |
-          nvidia-smi
-      - name: Install dependencies
-        run: |
-          apt-get update && apt-get install libsndfile1-dev libgl1 -y
-          python -m pip install -e .[quality,test]
-          python -m pip install git+https://github.com/huggingface/accelerate.git
-      - name: Environment
-        run: |
-          python utils/print_env.py
-      - name: Slow PyTorch CUDA checkpoint tests on Ubuntu
-        env:
-          HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
-          # https://pytorch.org/docs/stable/notes/randomness.html#avoiding-nondeterministic-algorithms
-          CUBLAS_WORKSPACE_CONFIG: :16:8
-        run: |
-          python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
-            -s -v -k "not Flax and not Onnx" \
-            --make-reports=tests_pipeline_${{ matrix.module }}_cuda \
-            tests/pipelines/${{ matrix.module }}
-      - name: Failure short reports
-        if: ${{ failure() }}
-        run: |
-          cat reports/tests_pipeline_${{ matrix.module }}_cuda_stats.txt
-          cat reports/tests_pipeline_${{ matrix.module }}_cuda_failures_short.txt
+        config:
+          - name: Slow PyTorch CUDA tests on Ubuntu
+            framework: pytorch
+            runner: docker-gpu
+            image: diffusers/diffusers-pytorch-cuda
+            report: torch_cuda
+          - name: Slow Flax TPU tests on Ubuntu
+            framework: flax
+            runner: docker-tpu
+            image: diffusers/diffusers-flax-tpu
+            report: flax_tpu
+          - name: Slow ONNXRuntime CUDA tests on Ubuntu
+            framework: onnxruntime
+            runner: docker-gpu
+            image: diffusers/diffusers-onnxruntime-cuda
+            report: onnx_cuda
 
-      - name: Test suite reports artifacts
-        if: ${{ always() }}
-        uses: actions/upload-artifact@v2
-        with:
-          name: pipeline_${{ matrix.module }}_test_reports
-          path: reports
+    name: ${{ matrix.config.name }}
+
+    runs-on: ${{ matrix.config.runner }}
 
-  torch_cuda_tests:
-    name: Torch CUDA Tests
-    runs-on: docker-gpu
     container:
-      image: diffusers/diffusers-pytorch-cuda
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --gpus 0
+      image: ${{ matrix.config.image }}
+      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ ${{ matrix.config.runner == 'docker-tpu' && '--privileged' || '--gpus 0'}}
+
     defaults:
       run:
         shell: bash
-    strategy:
-      matrix:
-        module: [models, schedulers, lora, others]
+
     steps:
     - name: Checkout diffusers
       uses: actions/checkout@v3
       with:
         fetch-depth: 2
 
+    - name: NVIDIA-SMI
+      if : ${{ matrix.config.runner == 'docker-gpu' }}
+      run: |
+        nvidia-smi
+
     - name: Install dependencies
       run: |
         apt-get update && apt-get install libsndfile1-dev libgl1 -y
@@ -133,201 +70,47 @@ jobs:
         python utils/print_env.py
 
     - name: Run slow PyTorch CUDA tests
+      if: ${{ matrix.config.framework == 'pytorch' }}
       env:
         HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
         # https://pytorch.org/docs/stable/notes/randomness.html#avoiding-nondeterministic-algorithms
-        CUBLAS_WORKSPACE_CONFIG: :16:8
+        CUBLAS_WORKSPACE_CONFIG: :16:8 
+
       run: |
         python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
           -s -v -k "not Flax and not Onnx" \
-          --make-reports=tests_torch_cuda \
-          tests/${{ matrix.module }}
-
-    - name: Failure short reports
-      if: ${{ failure() }}
-      run: |
-        cat reports/tests_torch_cuda_stats.txt
-        cat reports/tests_torch_cuda_failures_short.txt
-
-    - name: Test suite reports artifacts
-      if: ${{ always() }}
-      uses: actions/upload-artifact@v2
-      with:
-        name: torch_cuda_test_reports
-        path: reports
-
-  flax_tpu_tests:
-    name: Flax TPU Tests
-    runs-on: docker-tpu
-    container:
-      image: diffusers/diffusers-flax-tpu
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --privileged
-    defaults:
-      run:
-        shell: bash
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: Install dependencies
-      run: |
-        apt-get update && apt-get install libsndfile1-dev libgl1 -y
-        python -m pip install -e .[quality,test]
-        python -m pip install git+https://github.com/huggingface/accelerate.git
-
-    - name: Environment
-      run: |
-        python utils/print_env.py
+          --make-reports=tests_${{ matrix.config.report }} \
+          tests/
 
     - name: Run slow Flax TPU tests
+      if: ${{ matrix.config.framework == 'flax' }}
       env:
         HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       run: |
         python -m pytest -n 0 \
           -s -v -k "Flax" \
-          --make-reports=tests_flax_tpu \
+          --make-reports=tests_${{ matrix.config.report }} \
           tests/
 
-    - name: Failure short reports
-      if: ${{ failure() }}
-      run: |
-        cat reports/tests_flax_tpu_stats.txt
-        cat reports/tests_flax_tpu_failures_short.txt
-
-    - name: Test suite reports artifacts
-      if: ${{ always() }}
-      uses: actions/upload-artifact@v2
-      with:
-        name: flax_tpu_test_reports
-        path: reports
-
-  onnx_cuda_tests:
-    name: ONNX CUDA Tests
-    runs-on: docker-gpu
-    container:
-      image: diffusers/diffusers-onnxruntime-cuda
-      options: --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/ --gpus 0
-    defaults:
-      run:
-        shell: bash
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: Install dependencies
-      run: |
-        apt-get update && apt-get install libsndfile1-dev libgl1 -y
-        python -m pip install -e .[quality,test]
-        python -m pip install git+https://github.com/huggingface/accelerate.git
-
-    - name: Environment
-      run: |
-        python utils/print_env.py
-
     - name: Run slow ONNXRuntime CUDA tests
+      if: ${{ matrix.config.framework == 'onnxruntime' }}
       env:
         HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
       run: |
         python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile \
           -s -v -k "Onnx" \
-          --make-reports=tests_onnx_cuda \
+          --make-reports=tests_${{ matrix.config.report }} \
           tests/
 
     - name: Failure short reports
       if: ${{ failure() }}
-      run: |
-        cat reports/tests_onnx_cuda_stats.txt
-        cat reports/tests_onnx_cuda_failures_short.txt
+      run: cat reports/tests_${{ matrix.config.report }}_failures_short.txt
 
     - name: Test suite reports artifacts
       if: ${{ always() }}
       uses: actions/upload-artifact@v2
       with:
-        name: onnx_cuda_test_reports
-        path: reports
-
-  run_torch_compile_tests:
-    name: PyTorch Compile CUDA tests
-
-    runs-on: docker-gpu
-
-    container:
-      image: diffusers/diffusers-pytorch-compile-cuda
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
-
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: NVIDIA-SMI
-      run: |
-        nvidia-smi
-    - name: Install dependencies
-      run: |
-        python -m pip install -e .[quality,test,training]
-    - name: Environment
-      run: |
-        python utils/print_env.py
-    - name: Run example tests on GPU
-      env:
-        HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
-      run: |
-        python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile -s -v -k "compile" --make-reports=tests_torch_compile_cuda tests/
-    - name: Failure short reports
-      if: ${{ failure() }}
-      run: cat reports/tests_torch_compile_cuda_failures_short.txt
-
-    - name: Test suite reports artifacts
-      if: ${{ always() }}
-      uses: actions/upload-artifact@v2
-      with:
-        name: torch_compile_test_reports
-        path: reports
-
-  run_xformers_tests:
-    name: PyTorch xformers CUDA tests
-
-    runs-on: docker-gpu
-
-    container:
-      image: diffusers/diffusers-pytorch-xformers-cuda
-      options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/hf_cache:/mnt/cache/
-
-    steps:
-    - name: Checkout diffusers
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 2
-
-    - name: NVIDIA-SMI
-      run: |
-        nvidia-smi
-    - name: Install dependencies
-      run: |
-        python -m pip install -e .[quality,test,training]
-    - name: Environment
-      run: |
-        python utils/print_env.py
-    - name: Run example tests on GPU
-      env:
-        HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
-      run: |
-        python -m pytest -n 1 --max-worker-restart=0 --dist=loadfile -s -v -k "xformers" --make-reports=tests_torch_xformers_cuda tests/
-    - name: Failure short reports
-      if: ${{ failure() }}
-      run: cat reports/tests_torch_xformers_cuda_failures_short.txt
-
-    - name: Test suite reports artifacts
-      if: ${{ always() }}
-      uses: actions/upload-artifact@v2
-      with:
-        name: torch_xformers_test_reports
+        name: ${{ matrix.config.report }}_test_reports
         path: reports
 
   run_examples_tests:
@@ -365,13 +148,11 @@ jobs:
 
     - name: Failure short reports
       if: ${{ failure() }}
-      run: |
-        cat reports/examples_torch_cuda_stats.txt
-        cat reports/examples_torch_cuda_failures_short.txt
+      run: cat reports/examples_torch_cuda_failures_short.txt
 
     - name: Test suite reports artifacts
       if: ${{ always() }}
       uses: actions/upload-artifact@v2
       with:
         name: examples_test_reports
-        path: reports
+        path: reports

.github/workflows/stale.yml

@@ -17,7 +17,7 @@ jobs:
     - name: Setup Python
       uses: actions/setup-python@v1
       with:
-        python-version: 3.8
+        python-version: 3.7
 
     - name: Install requirements
       run: |

docker/diffusers-pytorch-compile-cuda/Dockerfile

@@ -1,46 +0,0 @@
-FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04
-LABEL maintainer="Hugging Face"
-LABEL repository="diffusers"
-
-ENV DEBIAN_FRONTEND=noninteractive
-
-RUN apt update && \
-    apt install -y bash \
-                   build-essential \
-                   git \
-                   git-lfs \
-                   curl \
-                   ca-certificates \
-                   libsndfile1-dev \
-                   libgl1 \
-                   python3.9 \
-                   python3.9-dev \
-                   python3-pip \
-                   python3.9-venv && \
-    rm -rf /var/lib/apt/lists
-
-# make sure to use venv
-RUN python3.9 -m venv /opt/venv
-ENV PATH="/opt/venv/bin:$PATH"
-
-# pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3.9 -m pip install --no-cache-dir --upgrade pip && \
-    python3.9 -m pip install --no-cache-dir \
-        torch \
-        torchvision \
-        torchaudio \
-        invisible_watermark && \
-    python3.9 -m pip install --no-cache-dir \
-        accelerate \
-        datasets \
-        hf-doc-builder \
-        huggingface-hub \
-        Jinja2 \
-        librosa \
-        numpy \
-        scipy \
-        tensorboard \
-        transformers \
-        omegaconf
-
-CMD ["/bin/bash"]

docker/diffusers-pytorch-cuda/Dockerfile

@@ -40,6 +40,8 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip && \
         scipy \
         tensorboard \
         transformers \
-        omegaconf
+        omegaconf \
+        pytorch-lightning \
+        xformers
 
 CMD ["/bin/bash"]

docker/diffusers-pytorch-xformers-cuda/Dockerfile

@@ -1,46 +0,0 @@
-FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu20.04
-LABEL maintainer="Hugging Face"
-LABEL repository="diffusers"
-
-ENV DEBIAN_FRONTEND=noninteractive
-
-RUN apt update && \
-    apt install -y bash \
-                   build-essential \
-                   git \
-                   git-lfs \
-                   curl \
-                   ca-certificates \
-                   libsndfile1-dev \
-                   libgl1 \
-                   python3.8 \
-                   python3-pip \
-                   python3.8-venv && \
-    rm -rf /var/lib/apt/lists
-
-# make sure to use venv
-RUN python3 -m venv /opt/venv
-ENV PATH="/opt/venv/bin:$PATH"
-
-# pre-install the heavy dependencies (these can later be overridden by the deps from setup.py)
-RUN python3 -m pip install --no-cache-dir --upgrade pip && \
-    python3 -m pip install --no-cache-dir \
-        torch==2.0.1 \
-        torchvision==0.15.2 \
-        torchaudio \
-        invisible_watermark && \
-    python3 -m pip install --no-cache-dir \
-        accelerate \
-        datasets \
-        hf-doc-builder \
-        huggingface-hub \
-        Jinja2 \
-        librosa \
-        numpy \
-        scipy \
-        tensorboard \
-        transformers \
-        omegaconf \
-        xformers
-
-CMD ["/bin/bash"]

docs/README.md

@@ -128,7 +128,7 @@ When adding a new pipeline:
     - Possible an end-to-end example of how to use it
 - Add all the pipeline classes that should be linked in the diffusion model. These classes should be added using our Markdown syntax. By default as follows:
 
-```py
+```
 ## XXXPipeline
 
 [[autodoc]] XXXPipeline
@@ -138,7 +138,7 @@ When adding a new pipeline:
 
 This will include every public method of the pipeline that is documented, as well as the  `__call__` method that is not documented by default. If you just want to add additional methods that are not documented, you can put the list of all methods to add in a list that contains `all`.
 
-```py
+```
 [[autodoc]] XXXPipeline
     - all
 	- __call__
@@ -172,7 +172,7 @@ Arguments should be defined with the `Args:` (or `Arguments:` or `Parameters:`)
 an indentation. The argument should be followed by its type, with its shape if it is a tensor, a colon, and its
 description:
 
-```py
+```
     Args:
         n_layers (`int`): The number of layers of the model.
 ```
@@ -182,7 +182,7 @@ after the argument.
 
 Here's an example showcasing everything so far:
 
-```py
+```
     Args:
         input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
             Indices of input sequence tokens in the vocabulary.
@@ -196,13 +196,13 @@ Here's an example showcasing everything so far:
 For optional arguments or arguments with defaults we follow the following syntax: imagine we have a function with the
 following signature:
 
-```py
+```
 def my_function(x: str = None, a: float = 1):
 ```
 
 then its documentation should look like this:
 
-```py
+```
     Args:
         x (`str`, *optional*):
             This argument controls ...
@@ -235,14 +235,14 @@ building the return.
 
 Here's an example of a single value return:
 
-```py
+```
     Returns:
         `List[int]`: A list of integers in the range [0, 1] --- 1 for a special token, 0 for a sequence token.
 ```
 
 Here's an example of a tuple return, comprising several objects:
 
-```py
+```
     Returns:
         `tuple(torch.FloatTensor)` comprising various elements depending on the configuration ([`BertConfig`]) and inputs:
         - ** loss** (*optional*, returned when `masked_lm_labels` is provided) `torch.FloatTensor` of shape `(1,)` --

docs/source/en/_toctree.yml

@@ -58,8 +58,6 @@
       title: Control image brightness
     - local: using-diffusers/weighted_prompts
       title: Prompt weighting
-    - local: using-diffusers/freeu
-      title: Improve generation quality with FreeU
     title: Techniques
   - sections:
     - local: using-diffusers/pipeline_overview
@@ -106,8 +104,6 @@
       title: Custom Diffusion
     - local: training/t2i_adapters
       title: T2I-Adapters
-    - local: training/ddpo
-      title: Reinforcement learning training with DDPO
     title: Training
   - sections:
     - local: using-diffusers/other-modalities
@@ -117,35 +113,27 @@
 - sections:
   - local: optimization/opt_overview
     title: Overview
-  - sections:
-    - local: optimization/fp16
-      title: Speed up inference
-    - local: optimization/memory
-      title: Reduce memory usage
-    - local: optimization/torch2.0
-      title: Torch 2.0
-    - local: optimization/xformers
-      title: xFormers
-    - local: optimization/tome
-      title: Token merging
-    title: General optimizations
-  - sections:
-    - local: using-diffusers/stable_diffusion_jax_how_to
-      title: JAX/Flax
-    - local: optimization/onnx
-      title: ONNX
-    - local: optimization/open_vino
-      title: OpenVINO
-    - local: optimization/coreml
-      title: Core ML
-    title: Optimized model types
-  - sections:
-    - local: optimization/mps
-      title: Metal Performance Shaders (MPS)
-    - local: optimization/habana
-      title: Habana Gaudi
-    title: Optimized hardware
-  title: Optimization
+  - local: optimization/fp16
+    title: Memory and Speed
+  - local: optimization/torch2.0
+    title: Torch2.0 support
+  - local: using-diffusers/stable_diffusion_jax_how_to
+    title: Stable Diffusion in JAX/Flax
+  - local: optimization/xformers
+    title: xFormers
+  - local: optimization/onnx
+    title: ONNX
+  - local: optimization/open_vino
+    title: OpenVINO
+  - local: optimization/coreml
+    title: Core ML
+  - local: optimization/mps
+    title: MPS
+  - local: optimization/habana
+    title: Habana Gaudi
+  - local: optimization/tome
+    title: Token Merging
+  title: Optimization/Special Hardware
 - sections:
   - local: conceptual/philosophy
     title: Philosophy
@@ -220,8 +208,6 @@
       title: AudioLDM 2
     - local: api/pipelines/auto_pipeline
       title: AutoPipeline
-    - local: api/pipelines/blip_diffusion
-      title: BLIP Diffusion
     - local: api/pipelines/consistency_models
       title: Consistency Models
     - local: api/pipelines/controlnet

docs/source/en/api/attnprocessor.md

@@ -17,9 +17,6 @@ An attention processor is a class for applying different types of attention mech
 ## CustomDiffusionAttnProcessor
 [[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor
 
-## CustomDiffusionAttnProcessor2_0
-[[autodoc]] models.attention_processor.CustomDiffusionAttnProcessor2_0
-
 ## AttnAddedKVProcessor
 [[autodoc]] models.attention_processor.AttnAddedKVProcessor
 
@@ -42,4 +39,4 @@ An attention processor is a class for applying different types of attention mech
 [[autodoc]] models.attention_processor.SlicedAttnProcessor
 
 ## SlicedAttnAddedKVProcessor
-[[autodoc]] models.attention_processor.SlicedAttnAddedKVProcessor
+[[autodoc]] models.attention_processor.SlicedAttnAddedKVProcessor

docs/source/en/api/loaders.md

@@ -28,10 +28,6 @@ Adapters (textual inversion, LoRA, hypernetworks) allow you to modify a diffusio
 
 [[autodoc]] loaders.TextualInversionLoaderMixin
 
-## StableDiffusionXLLoraLoaderMixin
-
-[[autodoc]] loaders.StableDiffusionXLLoraLoaderMixin
-
 ## LoraLoaderMixin
 
 [[autodoc]] loaders.LoraLoaderMixin

docs/source/en/api/logging.md

@@ -67,30 +67,30 @@ By default, `tqdm` progress bars are displayed during model download. [`logging.
 
 ## Base setters
 
-[[autodoc]] utils.logging.set_verbosity_error
+[[autodoc]] logging.set_verbosity_error
 
-[[autodoc]] utils.logging.set_verbosity_warning
+[[autodoc]] logging.set_verbosity_warning
 
-[[autodoc]] utils.logging.set_verbosity_info
+[[autodoc]] logging.set_verbosity_info
 
-[[autodoc]] utils.logging.set_verbosity_debug
+[[autodoc]] logging.set_verbosity_debug
 
 ## Other functions
 
-[[autodoc]] utils.logging.get_verbosity
+[[autodoc]] logging.get_verbosity
 
-[[autodoc]] utils.logging.set_verbosity
+[[autodoc]] logging.set_verbosity
 
-[[autodoc]] utils.logging.get_logger
+[[autodoc]] logging.get_logger
 
-[[autodoc]] utils.logging.enable_default_handler
+[[autodoc]] logging.enable_default_handler
 
-[[autodoc]] utils.logging.disable_default_handler
+[[autodoc]] logging.disable_default_handler
 
-[[autodoc]] utils.logging.enable_explicit_format
+[[autodoc]] logging.enable_explicit_format
 
-[[autodoc]] utils.logging.reset_format
+[[autodoc]] logging.reset_format
 
-[[autodoc]] utils.logging.enable_progress_bar
+[[autodoc]] logging.enable_progress_bar
 
-[[autodoc]] utils.logging.disable_progress_bar
+[[autodoc]] logging.disable_progress_bar

docs/source/en/api/pipelines/auto_pipeline.md

@@ -42,7 +42,7 @@ Check out the [AutoPipeline](/tutorials/autopipeline) tutorial to learn how to u
 `AutoPipeline` supports text-to-image, image-to-image, and inpainting for the following diffusion models:
 
 - [Stable Diffusion](./stable_diffusion)
-- [ControlNet](./controlnet)
+- [ControlNet](./api/pipelines/controlnet)
 - [Stable Diffusion XL (SDXL)](./stable_diffusion/stable_diffusion_xl)
 - [DeepFloyd IF](./if) 
 - [Kandinsky](./kandinsky)

docs/source/en/api/pipelines/blip_diffusion.md

@@ -1,29 +0,0 @@
-# Blip Diffusion
-
-Blip Diffusion was proposed in [BLIP-Diffusion: Pre-trained Subject Representation for Controllable Text-to-Image Generation and Editing](https://arxiv.org/abs/2305.14720). It enables zero-shot subject-driven generation and control-guided zero-shot generation. 
-
-
-The abstract from the paper is:
-
-*Subject-driven text-to-image generation models create novel renditions of an input subject based on text prompts. Existing models suffer from lengthy fine-tuning and difficulties preserving the subject fidelity. To overcome these limitations, we introduce BLIP-Diffusion, a new subject-driven image generation model that supports multimodal control which consumes inputs of subject images and text prompts. Unlike other subject-driven generation models, BLIP-Diffusion introduces a new multimodal encoder which is pre-trained to provide subject representation. We first pre-train the multimodal encoder following BLIP-2 to produce visual representation aligned with the text. Then we design a subject representation learning task which enables a diffusion model to leverage such visual representation and generates new subject renditions. Compared with previous methods such as DreamBooth, our model enables zero-shot subject-driven generation, and efficient fine-tuning for customized subject with up to 20x speedup. We also demonstrate that BLIP-Diffusion can be flexibly combined with existing techniques such as ControlNet and prompt-to-prompt to enable novel subject-driven generation and editing applications.*
-
-The original codebase can be found at [salesforce/LAVIS](https://github.com/salesforce/LAVIS/tree/main/projects/blip-diffusion). You can find the official BLIP Diffusion checkpoints under the [hf.co/SalesForce](https://hf.co/SalesForce) organization.
-
-`BlipDiffusionPipeline` and `BlipDiffusionControlNetPipeline` were contributed by [`ayushtues`](https://github.com/ayushtues/).
-
-<Tip>
-
-Make sure to check out the Schedulers [guide](/using-diffusers/schedulers) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](/using-diffusers/loading#reuse-components-across-pipelines) section to learn how to efficiently load the same components into multiple pipelines.
-
-</Tip>
-
-
-## BlipDiffusionPipeline
-[[autodoc]] BlipDiffusionPipeline
-    - all
-    - __call__
-
-## BlipDiffusionControlNetPipeline
-[[autodoc]] BlipDiffusionControlNetPipeline
-    - all
-    - __call__

docs/source/en/api/pipelines/pix2pix.md

@@ -34,7 +34,13 @@ Make sure to check out the Schedulers [guide](/using-diffusers/schedulers) to le
 	- load_lora_weights
 	- save_lora_weights
 
+## StableDiffusionPipelineOutput
+[[autodoc]] pipelines.stable_diffusion.StableDiffusionPipelineOutput
+
 ## StableDiffusionXLInstructPix2PixPipeline
 [[autodoc]] StableDiffusionXLInstructPix2PixPipeline
 	- __call__
 	- all
+
+## StableDiffusionXLPipelineOutput
+[[autodoc]] pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput

docs/source/en/api/pipelines/semantic_stable_diffusion.md

@@ -31,5 +31,5 @@ Make sure to check out the Schedulers [guide](/using-diffusers/schedulers) to le
 	- __call__
 
 ## StableDiffusionSafePipelineOutput
-[[autodoc]] pipelines.semantic_stable_diffusion.pipeline_output.SemanticStableDiffusionPipelineOutput
-	- all
+[[autodoc]] pipelines.semantic_stable_diffusion.SemanticStableDiffusionPipelineOutput
+	- all

docs/source/en/api/pipelines/stable_diffusion/adapter.md

@@ -28,8 +28,8 @@ This model was contributed by the community contributor [HimariO](https://github
 
 | Pipeline | Tasks | Demo
 |---|---|:---:|
-| [StableDiffusionAdapterPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py) | *Text-to-Image Generation with T2I-Adapter Conditioning* | -
-| [StableDiffusionXLAdapterPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py) | *Text-to-Image Generation with T2I-Adapter Conditioning on StableDiffusion-XL* | -
+| [StableDiffusionAdapterPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_adapter.py) | *Text-to-Image Generation with T2I-Adapter Conditioning* | -
+| [StableDiffusionXLAdapterPipeline](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_xl_adapter.py) | *Text-to-Image Generation with T2I-Adapter Conditioning on StableDiffusion-XL* | -
 
 ## Usage example with the base model of StableDiffusion-1.4/1.5
 

docs/source/en/api/pipelines/wuerstchen.md

@@ -2,18 +2,15 @@
 
 <img src="https://github.com/dome272/Wuerstchen/assets/61938694/0617c863-165a-43ee-9303-2a17299a0cf9">
 
-[Würstchen: Efficient Pretraining of Text-to-Image Models](https://huggingface.co/papers/2306.00637) is by Pablo Pernias, Dominic Rampas, Mats L. Richter and Christopher Pal and Marc Aubreville.
+[Würstchen: Efficient Pretraining of Text-to-Image Models](https://huggingface.co/papers/2306.00637) is by Pablo Pernias, Dominic Rampas, and Marc Aubreville.
 
 The abstract from the paper is:
 
 *We introduce Würstchen, a novel technique for text-to-image synthesis that unites competitive performance with unprecedented cost-effectiveness and ease of training on constrained hardware. Building on recent advancements in machine learning, our approach, which utilizes latent diffusion strategies at strong latent image compression rates, significantly reduces the computational burden, typically associated with state-of-the-art models, while preserving, if not enhancing, the quality of generated images. Wuerstchen achieves notable speed improvements at inference time, thereby rendering real-time applications more viable. One of the key advantages of our method lies in its modest training requirements of only 9,200 GPU hours, slashing the usual costs significantly without compromising the end performance. In a comparison against the state-of-the-art, we found the approach to yield strong competitiveness. This paper opens the door to a new line of research that prioritizes both performance and computational accessibility, hence democratizing the use of sophisticated AI technologies. Through Wuerstchen, we demonstrate a compelling stride forward in the realm of text-to-image synthesis, offering an innovative path to explore in future research.*
 
-## Würstchen Overview
-Würstchen is a diffusion model, whose text-conditional model works in a highly compressed latent space of images. Why is this important? Compressing data can reduce computational costs for both training and inference by magnitudes. Training on 1024x1024 images is way more expensive than training on 32x32. Usually, other works make use of a relatively small compression, in the range of 4x - 8x spatial compression. Würstchen takes this to an extreme. Through its novel design, we achieve a 42x spatial compression. This was unseen before because common methods fail to faithfully reconstruct detailed images after 16x spatial compression. Würstchen employs a two-stage compression, what we call Stage A and Stage B. Stage A is a VQGAN, and Stage B is a Diffusion Autoencoder (more details can be found in the [paper](https://huggingface.co/papers/2306.00637) ). A third model, Stage C, is learned in that highly compressed latent space. This training requires fractions of the compute used for current top-performing models, while also allowing cheaper and faster inference.
-
 ## Würstchen v2 comes to Diffusers
 
-After the initial paper release, we have improved numerous things in the architecture, training and sampling, making Würstchen competitive to current state-of-the-art models in many ways. We are excited to release this new version together with Diffusers. Here is a list of the improvements.
+After the initial paper release, we have improved numerous things in the architecture, training and sampling, making Würstchen competetive to current state-of-the-art models in many ways. We are excited to release this new version together with Diffusers. Here is a list of the improvements.
 
 - Higher resolution (1024x1024 up to 2048x2048)
 - Faster inference
@@ -25,16 +22,16 @@ We are releasing 3 checkpoints for the text-conditional image generation model (
 
 - v2-base
 - v2-aesthetic
-- **(default)** v2-interpolated (50% interpolation between v2-base and v2-aesthetic)
+- v2-interpolated (50% interpolation between v2-base and v2-aesthetic)
 
-We recommend using v2-interpolated, as it has a nice touch of both photorealism and aesthetics. Use v2-base for finetunings as it does not have a style bias and use v2-aesthetic for very artistic generations.
+We recommend to use v2-interpolated, as it has a nice touch of both photorealism and aesthetic. Use v2-base for finetunings as it does not have a style bias and use v2-aesthetic for very artistic generations.
 A comparison can be seen here:
 
 <img src="https://github.com/dome272/Wuerstchen/assets/61938694/2914830f-cbd3-461c-be64-d50734f4b49d" width=500>
 
 ## Text-to-Image Generation
 
-For the sake of usability, Würstchen can be used with a single pipeline. This pipeline can be used as follows:
+For the sake of usability Würstchen can be used with a single pipeline. This pipeline is called `WuerstchenCombinedPipeline` and can be used as follows:
 
 ```python
 import torch
@@ -88,6 +85,7 @@ decoder_output = decoder_pipeline(
     image_embeddings=prior_output.image_embeddings,
     prompt=caption,
     negative_prompt=negative_prompt,
+    num_images_per_prompt=num_images_per_prompt,
     guidance_scale=0.0,
     output_type="pil",
 ).images
@@ -97,8 +95,8 @@ decoder_output = decoder_pipeline(
 You can make use of `torch.compile` function and gain a speed-up of about 2-3x:
 
 ```python
-prior_pipeline.prior = torch.compile(prior_pipeline.prior, mode="reduce-overhead", fullgraph=True)
-decoder_pipeline.decoder = torch.compile(decoder_pipeline.decoder, mode="reduce-overhead", fullgraph=True)
+pipeline.prior = torch.compile(pipeline.prior, mode="reduce-overhead", fullgraph=True)
+pipeline.decoder = torch.compile(pipeline.decoder, mode="reduce-overhead", fullgraph=True)
 ```
 
 ## Limitations
@@ -113,7 +111,7 @@ after 1024x1024 is 1152x1152
 
 The original codebase, as well as experimental ideas, can be found at [dome272/Wuerstchen](https://github.com/dome272/Wuerstchen).
 
-## WuerstchenCombinedPipeline
+## WuerschenPipeline
 
 [[autodoc]] WuerstchenCombinedPipeline
 	- all
@@ -121,7 +119,8 @@ The original codebase, as well as experimental ideas, can be found at [dome272/W
 
 ## WuerstchenPriorPipeline
 
-[[autodoc]] WuerstchenPriorPipeline
+[[autodoc]] WuerstchenDecoderPipeline
+
 	- all
 	- __call__
 
@@ -134,16 +133,3 @@ The original codebase, as well as experimental ideas, can be found at [dome272/W
 [[autodoc]] WuerstchenDecoderPipeline
 	- all
 	- __call__
-
-## Citation
-
-```bibtex
-      @misc{pernias2023wuerstchen,
-            title={Wuerstchen: Efficient Pretraining of Text-to-Image Models}, 
-            author={Pablo Pernias and Dominic Rampas and Mats L. Richter and Christopher Pal and Marc Aubreville},
-            year={2023},
-            eprint={2306.00637},
-            archivePrefix={arXiv},
-            primaryClass={cs.CV}
-      }
-```

docs/source/en/installation.md

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 Install 🤗 Diffusers for whichever deep learning library you're working with.
 
-🤗 Diffusers is tested on Python 3.8+, PyTorch 1.7.0+ and Flax. Follow the installation instructions below for the deep learning library you are using:
+🤗 Diffusers is tested on Python 3.7+, PyTorch 1.7.0+ and Flax. Follow the installation instructions below for the deep learning library you are using:
 
 - [PyTorch](https://pytorch.org/get-started/locally/) installation instructions.
 - [Flax](https://flax.readthedocs.io/en/latest/) installation instructions.
@@ -106,7 +106,7 @@ pip install -e ".[flax]"
 
 These commands will link the folder you cloned the repository to and your Python library paths.
 Python will now look inside the folder you cloned to in addition to the normal library paths.
-For example, if your Python packages are typically installed in `~/anaconda3/envs/main/lib/python3.8/site-packages/`, Python will also search the `~/diffusers/` folder you cloned to.
+For example, if your Python packages are typically installed in `~/anaconda3/envs/main/lib/python3.7/site-packages/`, Python will also search the `~/diffusers/` folder you cloned to.
 
 <Tip warning={true}>
 

docs/source/en/optimization/fp16.md

@@ -10,19 +10,13 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Speed up inference
+# Memory and speed
 
-There are several ways to optimize 🤗 Diffusers for inference speed. As a general rule of thumb, we recommend using either [xFormers](xformers) or `torch.nn.functional.scaled_dot_product_attention` in PyTorch 2.0 for their memory-efficient attention. 
+We present some techniques and ideas to optimize 🤗 Diffusers _inference_ for memory or speed. As a general rule, we recommend the use of [xFormers](https://github.com/facebookresearch/xformers) for memory efficient attention, please see the recommended [installation instructions](xformers).
 
-<Tip>
+We'll discuss how the following settings impact performance and memory.
 
-In many cases, optimizing for speed or memory leads to improved performance in the other, so you should try to optimize for both whenever you can. This guide focuses on inference speed, but you can learn more about preserving memory in the [Reduce memory usage](memory) guide.
-
-</Tip>
-
-The results below are obtained from generating a single 512x512 image from the prompt `a photo of an astronaut riding a horse on mars` with 50 DDIM steps on a Nvidia Titan RTX, demonstrating the speed-up you can expect.
-
-|                  | latency | speed-up |
+|                  | Latency | Speedup |
 | ---------------- | ------- | ------- |
 | original         | 9.50s   | x1      |
 | fp16             | 3.61s   | x2.63   |
@@ -30,9 +24,15 @@ The results below are obtained from generating a single 512x512 image from the p
 | traced UNet      | 3.21s   | x2.96   |
 | memory efficient attention  | 2.63s  | x3.61   |
 
-## Use TensorFloat-32
+<em>
+  obtained on NVIDIA TITAN RTX by generating a single image of size 512x512 from
+  the prompt "a photo of an astronaut riding a horse on mars" with 50 DDIM
+  steps.
+</em>
 
-On Ampere and later CUDA devices, matrix multiplications and convolutions can use the [TensorFloat-32 (TF32)](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/) mode for faster, but slightly less accurate computations. By default, PyTorch enables TF32 mode for convolutions but not matrix multiplications. Unless your network requires full float32 precision, we recommend enabling TF32 for matrix multiplications. It can significantly speeds up computations with typically negligible loss in numerical accuracy.
+### Use tf32 instead of fp32 (on Ampere and later CUDA devices)
+
+On Ampere and later CUDA devices matrix multiplications and convolutions can use the TensorFloat32 (TF32) mode for faster but slightly less accurate computations. By default PyTorch enables TF32 mode for convolutions but not matrix multiplications, and unless a network requires full float32 precision we recommend enabling this setting for matrix multiplications, too. It can significantly speed up computations with typically negligible loss of numerical accuracy. You can read more about it [here](https://huggingface.co/docs/transformers/v4.18.0/en/performance#tf32). All you need to do is to add this before your inference:
 
 ```python
 import torch
@@ -40,11 +40,9 @@ import torch
 torch.backends.cuda.matmul.allow_tf32 = True
 ```
 
-You can learn more about TF32 in the [Mixed precision training](https://huggingface.co/docs/transformers/en/perf_train_gpu_one#tf32) guide.
+## Half precision weights
 
-## Half-precision weights
-
-To save GPU memory and get more speed, try loading and running the model weights directly in half-precision or float16:
+To save more GPU memory and get more speed, you can load and run the model weights directly in half precision. This involves loading the float16 version of the weights, which was saved to a branch named `fp16`, and telling PyTorch to use the `float16` type when loading them:
 
 ```Python
 import torch
@@ -63,6 +61,351 @@ image = pipe(prompt).images[0]
 
 <Tip warning={true}>
 
-Don't use [`torch.autocast`](https://pytorch.org/docs/stable/amp.html#torch.autocast) in any of the pipelines as it can lead to black images and is always slower than pure float16 precision.
+  It is strongly discouraged to make use of [`torch.autocast`](https://pytorch.org/docs/stable/amp.html#torch.autocast) in any of the pipelines as it can lead to black images and is always slower than using pure 
+  float16 precision.
   
-</Tip>
+</Tip>
+
+## Sliced VAE decode for larger batches
+
+To decode large batches of images with limited VRAM, or to enable batches with 32 images or more, you can use sliced VAE decode that decodes the batch latents one image at a time.
+
+You likely want to couple this with [`~StableDiffusionPipeline.enable_xformers_memory_efficient_attention`] to further minimize memory use.
+
+To perform the VAE decode one image at a time, invoke [`~StableDiffusionPipeline.enable_vae_slicing`] in your pipeline before inference. For example:
+
+```Python
+import torch
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+pipe = pipe.to("cuda")
+
+prompt = "a photo of an astronaut riding a horse on mars"
+pipe.enable_vae_slicing()
+images = pipe([prompt] * 32).images
+```
+
+You may see a small performance boost in VAE decode on multi-image batches. There should be no performance impact on single-image batches.
+
+
+## Tiled VAE decode and encode for large images
+
+Tiled VAE processing makes it possible to work with large images on limited VRAM. For example, generating 4k images in 8GB of VRAM. Tiled VAE decoder splits the image into overlapping tiles, decodes the tiles, and blends the outputs to make the final image.
+
+You want to couple this with [`~StableDiffusionPipeline.enable_xformers_memory_efficient_attention`] to further minimize memory use.
+
+To use tiled VAE processing, invoke [`~StableDiffusionPipeline.enable_vae_tiling`] in your pipeline before inference. For example:
+
+```python
+import torch
+from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+pipe = pipe.to("cuda")
+prompt = "a beautiful landscape photograph"
+pipe.enable_vae_tiling()
+pipe.enable_xformers_memory_efficient_attention()
+
+image = pipe([prompt], width=3840, height=2224, num_inference_steps=20).images[0]
+```
+
+The output image will have some tile-to-tile tone variation from the tiles having separate decoders, but you shouldn't see sharp seams between the tiles. The tiling is turned off for images that are 512x512 or smaller.
+
+
+<a name="sequential_offloading"></a>
+## Offloading to CPU with accelerate for memory savings
+
+For additional memory savings, you can offload the weights to CPU and only load them to GPU when performing the forward pass.
+
+To perform CPU offloading, all you have to do is invoke [`~StableDiffusionPipeline.enable_sequential_cpu_offload`]:
+
+```Python
+import torch
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+
+prompt = "a photo of an astronaut riding a horse on mars"
+pipe.enable_sequential_cpu_offload()
+image = pipe(prompt).images[0]
+```
+
+And you can get the memory consumption to < 3GB.
+
+Note that this method works at the submodule level, not on whole models. This is the best way to minimize memory consumption, but inference is much slower due to the iterative nature of the process. The UNet component of the pipeline runs several times (as many as `num_inference_steps`); each time, the different submodules of the UNet are sequentially onloaded and then offloaded as they are needed, so the number of memory transfers is large.
+
+<Tip>
+Consider using <a href="#model_offloading">model offloading</a> as another point in the optimization space: it will be much faster, but memory savings won't be as large.
+</Tip>
+
+It is also possible to chain offloading with attention slicing for minimal memory consumption (< 2GB).
+
+```Python
+import torch
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+
+prompt = "a photo of an astronaut riding a horse on mars"
+pipe.enable_sequential_cpu_offload()
+
+image = pipe(prompt).images[0]
+```
+
+**Note**: When using `enable_sequential_cpu_offload()`, it is important to **not** move the pipeline to CUDA beforehand or else the gain in memory consumption will only be minimal. See [this issue](https://github.com/huggingface/diffusers/issues/1934) for more information.
+
+**Note**: `enable_sequential_cpu_offload()` is a stateful operation that installs hooks on the models.
+
+
+<a name="model_offloading"></a>
+## Model offloading for fast inference and memory savings
+
+[Sequential CPU offloading](#sequential_offloading), as discussed in the previous section, preserves a lot of memory but makes inference slower, because submodules are moved to GPU as needed, and immediately returned to CPU when a new module runs.
+
+Full-model offloading is an alternative that moves whole models to the GPU, instead of handling each model's constituent _modules_. This results in a negligible impact on inference time (compared with moving the pipeline to `cuda`), while still providing some memory savings.
+
+In this scenario, only one of the main components of the pipeline (typically: text encoder, unet and vae)
+will be in the GPU while the others wait in the CPU. Components like the UNet that run for multiple iterations will stay on GPU until they are no longer needed.
+
+This feature can be enabled by invoking `enable_model_cpu_offload()` on the pipeline, as shown below.
+
+```Python
+import torch
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",  
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+
+prompt = "a photo of an astronaut riding a horse on mars"
+pipe.enable_model_cpu_offload()
+image = pipe(prompt).images[0]
+```
+
+This is also compatible with attention slicing for additional memory savings.
+
+```Python
+import torch
+from diffusers import StableDiffusionPipeline
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+)
+
+prompt = "a photo of an astronaut riding a horse on mars"
+pipe.enable_model_cpu_offload()
+
+image = pipe(prompt).images[0]
+```
+
+<Tip>
+This feature requires `accelerate` version 0.17.0 or larger.
+</Tip>
+
+**Note**: `enable_model_cpu_offload()` is a stateful operation that installs hooks on the models and state on the pipeline. In order to properly offload
+models after they are called, it is required that the entire pipeline is run and models are called in the order the pipeline expects them to be. Exercise caution
+if models are re-used outside the context of the pipeline after hooks have been installed. See [accelerate](https://huggingface.co/docs/accelerate/v0.18.0/en/package_reference/big_modeling#accelerate.hooks.remove_hook_from_module)
+for further docs on removing hooks.
+
+## Using Channels Last memory format
+
+Channels last memory format is an alternative way of ordering NCHW tensors in memory preserving dimensions ordering. Channels last tensors ordered in such a way that channels become the densest dimension (aka storing images pixel-per-pixel). Since not all operators currently support channels last format it may result in a worst performance, so it's better to try it and see if it works for your model.
+
+For example, in order to set the UNet model in our pipeline to use channels last format, we can use the following:
+
+```python
+print(pipe.unet.conv_out.state_dict()["weight"].stride())  # (2880, 9, 3, 1)
+pipe.unet.to(memory_format=torch.channels_last)  # in-place operation
+print(
+    pipe.unet.conv_out.state_dict()["weight"].stride()
+)  # (2880, 1, 960, 320) having a stride of 1 for the 2nd dimension proves that it works
+```
+
+## Tracing
+
+Tracing runs an example input tensor through your model, and captures the operations that are invoked as that input makes its way through the model's layers so that an executable or `ScriptFunction` is returned that will be optimized using just-in-time compilation.
+
+To trace our UNet model, we can use the following:
+
+```python
+import time
+import torch
+from diffusers import StableDiffusionPipeline
+import functools
+
+# torch disable grad
+torch.set_grad_enabled(False)
+
+# set variables
+n_experiments = 2
+unet_runs_per_experiment = 50
+
+
+# load inputs
+def generate_inputs():
+    sample = torch.randn(2, 4, 64, 64).half().cuda()
+    timestep = torch.rand(1).half().cuda() * 999
+    encoder_hidden_states = torch.randn(2, 77, 768).half().cuda()
+    return sample, timestep, encoder_hidden_states
+
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+).to("cuda")
+unet = pipe.unet
+unet.eval()
+unet.to(memory_format=torch.channels_last)  # use channels_last memory format
+unet.forward = functools.partial(unet.forward, return_dict=False)  # set return_dict=False as default
+
+# warmup
+for _ in range(3):
+    with torch.inference_mode():
+        inputs = generate_inputs()
+        orig_output = unet(*inputs)
+
+# trace
+print("tracing..")
+unet_traced = torch.jit.trace(unet, inputs)
+unet_traced.eval()
+print("done tracing")
+
+
+# warmup and optimize graph
+for _ in range(5):
+    with torch.inference_mode():
+        inputs = generate_inputs()
+        orig_output = unet_traced(*inputs)
+
+
+# benchmarking
+with torch.inference_mode():
+    for _ in range(n_experiments):
+        torch.cuda.synchronize()
+        start_time = time.time()
+        for _ in range(unet_runs_per_experiment):
+            orig_output = unet_traced(*inputs)
+        torch.cuda.synchronize()
+        print(f"unet traced inference took {time.time() - start_time:.2f} seconds")
+    for _ in range(n_experiments):
+        torch.cuda.synchronize()
+        start_time = time.time()
+        for _ in range(unet_runs_per_experiment):
+            orig_output = unet(*inputs)
+        torch.cuda.synchronize()
+        print(f"unet inference took {time.time() - start_time:.2f} seconds")
+
+# save the model
+unet_traced.save("unet_traced.pt")
+```
+
+Then we can replace the `unet` attribute of the pipeline with the traced model like the following
+
+```python
+from diffusers import StableDiffusionPipeline
+import torch
+from dataclasses import dataclass
+
+
+@dataclass
+class UNet2DConditionOutput:
+    sample: torch.FloatTensor
+
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+).to("cuda")
+
+# use jitted unet
+unet_traced = torch.jit.load("unet_traced.pt")
+
+
+# del pipe.unet
+class TracedUNet(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.in_channels = pipe.unet.in_channels
+        self.device = pipe.unet.device
+
+    def forward(self, latent_model_input, t, encoder_hidden_states):
+        sample = unet_traced(latent_model_input, t, encoder_hidden_states)[0]
+        return UNet2DConditionOutput(sample=sample)
+
+
+pipe.unet = TracedUNet()
+
+with torch.inference_mode():
+    image = pipe([prompt] * 1, num_inference_steps=50).images[0]
+```
+
+
+## Memory Efficient Attention
+
+Recent work on optimizing the bandwitdh in the attention block has generated huge speed ups and gains in GPU memory usage. The most recent being Flash Attention from @tridao: [code](https://github.com/HazyResearch/flash-attention), [paper](https://arxiv.org/pdf/2205.14135.pdf).
+
+Here are the speedups we obtain on a few Nvidia GPUs when running the inference at 512x512 with a batch size of 1 (one prompt):
+
+| GPU              	| Base Attention FP16 	| Memory Efficient Attention FP16 	|
+|------------------	|---------------------	|---------------------------------	|
+| NVIDIA Tesla T4  	| 3.5it/s             	| 5.5it/s                         	|
+| NVIDIA 3060 RTX  	| 4.6it/s             	| 7.8it/s                         	|
+| NVIDIA A10G      	| 8.88it/s            	| 15.6it/s                        	|
+| NVIDIA RTX A6000 	| 11.7it/s            	| 21.09it/s                       	|
+| NVIDIA TITAN RTX  | 12.51it/s         	| 18.22it/s                       	|
+| A100-SXM4-40GB    	| 18.6it/s            	| 29.it/s                        	|
+| A100-SXM-80GB    	| 18.7it/s            	| 29.5it/s                        	|
+
+To leverage it just make sure you have:
+
+<Tip warning={true}>
+
+If you have PyTorch 2.0 installed, you shouldn't use xFormers!
+
+</Tip>
+
+ - PyTorch > 1.12
+ - Cuda available
+ - [Installed the xformers library](xformers).
+```python
+from diffusers import DiffusionPipeline
+import torch
+
+pipe = DiffusionPipeline.from_pretrained(
+    "runwayml/stable-diffusion-v1-5",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+).to("cuda")
+
+pipe.enable_xformers_memory_efficient_attention()
+
+with torch.inference_mode():
+    sample = pipe("a small cat")
+
+# optional: You can disable it via
+# pipe.disable_xformers_memory_efficient_attention()
+```

docs/source/en/optimization/habana.md

@@ -10,22 +10,25 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Habana Gaudi
+# How to use Stable Diffusion on Habana Gaudi
 
-🤗 Diffusers is compatible with Habana Gaudi through 🤗 [Optimum](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion). Follow the [installation](https://docs.habana.ai/en/latest/Installation_Guide/index.html) guide to install the SynapseAI and Gaudi drivers, and then install Optimum Habana:
+🤗 Diffusers is compatible with Habana Gaudi through 🤗 [Optimum Habana](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion).
 
-```bash
-python -m pip install --upgrade-strategy eager optimum[habana]
-```
+## Requirements
+
+- Optimum Habana 1.6 or later, [here](https://huggingface.co/docs/optimum/habana/installation) is how to install it.
+- SynapseAI 1.10.
+
+
+## Inference Pipeline
 
 To generate images with Stable Diffusion 1 and 2 on Gaudi, you need to instantiate two instances:
+- A pipeline with [`GaudiStableDiffusionPipeline`](https://huggingface.co/docs/optimum/habana/package_reference/stable_diffusion_pipeline). This pipeline supports *text-to-image generation*.
+- A scheduler with [`GaudiDDIMScheduler`](https://huggingface.co/docs/optimum/habana/package_reference/stable_diffusion_pipeline#optimum.habana.diffusers.GaudiDDIMScheduler). This scheduler has been optimized for Habana Gaudi.
 
-- [`~optimum.habana.diffusers.GaudiStableDiffusionPipeline`], a pipeline for text-to-image generation.
-- [`~optimum.habana.diffusers.GaudiDDIMScheduler`], a Gaudi-optimized scheduler.
-
-When you initialize the pipeline, you have to specify `use_habana=True` to deploy it on HPUs and to get the fastest possible generation, you should enable **HPU graphs** with `use_hpu_graphs=True`.
-
-Finally, specify a [`~optimum.habana.GaudiConfig`] which can be downloaded from the [Habana](https://huggingface.co/Habana) organization on the Hub.
+When initializing the pipeline, you have to specify `use_habana=True` to deploy it on HPUs.
+Furthermore, in order to get the fastest possible generations you should enable **HPU graphs** with `use_hpu_graphs=True`.
+Finally, you will need to specify a [Gaudi configuration](https://huggingface.co/docs/optimum/habana/package_reference/gaudi_config) which can be downloaded from the [Hugging Face Hub](https://huggingface.co/Habana).
 
 ```python
 from optimum.habana import GaudiConfig
@@ -42,8 +45,7 @@ pipeline = GaudiStableDiffusionPipeline.from_pretrained(
 )
 ```
 
-Now you can call the pipeline to generate images by batches from one or several prompts:
-
+You can then call the pipeline to generate images by batches from one or several prompts:
 ```python
 outputs = pipeline(
     prompt=[
@@ -55,21 +57,21 @@ outputs = pipeline(
 )
 ```
 
-For more information, check out 🤗 Optimum Habana's [documentation](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion) and the [example](https://github.com/huggingface/optimum-habana/tree/main/examples/stable-diffusion) provided in the official Github repository.
+For more information, check out Optimum Habana's [documentation](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion) and the [example](https://github.com/huggingface/optimum-habana/tree/main/examples/stable-diffusion) provided in the official Github repository.
 
 
 ## Benchmark
 
-We benchmarked Habana's first-generation Gaudi and Gaudi2 with the [Habana/stable-diffusion](https://huggingface.co/Habana/stable-diffusion) and [Habana/stable-diffusion-2](https://huggingface.co/Habana/stable-diffusion-2) Gaudi configurations (mixed precision bf16/fp32) to demonstrate their performance.
+Here are the latencies for Habana first-generation Gaudi and Gaudi2 with the [Habana/stable-diffusion](https://huggingface.co/Habana/stable-diffusion) and [Habana/stable-diffusion-2](https://huggingface.co/Habana/stable-diffusion-2) Gaudi configurations (mixed precision bf16/fp32):
 
-For [Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5) on 512x512 images:
+- [Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5) (512x512 resolution):
 
-|                        | Latency (batch size = 1) | Throughput  |
+|                        | Latency (batch size = 1) | Throughput (batch size = 8) |
 | ---------------------- |:------------------------:|:---------------------------:|
-| first-generation Gaudi | 3.80s                    | 0.308 images/s (batch size = 8)             |
-| Gaudi2                 | 1.33s                    | 1.081 images/s (batch size = 8)             |
+| first-generation Gaudi | 3.80s                    | 0.308 images/s              |
+| Gaudi2                 | 1.33s                    | 1.081 images/s              |
 
-For [Stable Diffusion v2.1](https://huggingface.co/stabilityai/stable-diffusion-2-1) on 768x768 images:
+- [Stable Diffusion v2.1](https://huggingface.co/stabilityai/stable-diffusion-2-1) (768x768 resolution):
 
 |                        | Latency (batch size = 1) | Throughput                      |
 | ---------------------- |:------------------------:|:-------------------------------:|

docs/source/en/optimization/memory.md

@@ -1,357 +0,0 @@
-# Reduce memory usage
-
-A barrier to using diffusion models is the large amount of memory required. To overcome this challenge, there are several memory-reducing techniques you can use to run even some of the largest models on free-tier or consumer GPUs. Some of these techniques can even be combined to further reduce memory usage.
-
-<Tip>
-
-In many cases, optimizing for memory or speed leads to improved performance in the other, so you should try to optimize for both whenever you can. This guide focuses on minimizing memory usage, but you can also learn more about how to [Speed up inference](fp16).
-
-</Tip>
-
-The results below are obtained from generating a single 512x512 image from the prompt a photo of an astronaut riding a horse on mars with 50 DDIM steps on a Nvidia Titan RTX, demonstrating the speed-up you can expect as a result of reduced memory consumption.
-
-|                  | latency | speed-up |
-| ---------------- | ------- | ------- |
-| original         | 9.50s   | x1      |
-| fp16             | 3.61s   | x2.63   |
-| channels last    | 3.30s   | x2.88   |
-| traced UNet      | 3.21s   | x2.96   |
-| memory-efficient attention  | 2.63s  | x3.61   |
-
-
-## Sliced VAE
-
-Sliced VAE enables decoding large batches of images with limited VRAM or batches with 32 images or more by decoding the batches of latents one image at a time. You'll likely want to couple this with [`~ModelMixin.enable_xformers_memory_efficient_attention`] to further reduce memory use.
-
-To use sliced VAE, call [`~StableDiffusionPipeline.enable_vae_slicing`] on your pipeline before inference:
-
-```python
-import torch
-from diffusers import StableDiffusionPipeline
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-pipe = pipe.to("cuda")
-
-prompt = "a photo of an astronaut riding a horse on mars"
-pipe.enable_vae_slicing()
-images = pipe([prompt] * 32).images
-```
-
-You may see a small performance boost in VAE decoding on multi-image batches, and there should be no performance impact on single-image batches.
-
-## Tiled VAE
-
-Tiled VAE processing also enables working with large images on limited VRAM (for example, generating 4k images on 8GB of VRAM) by splitting the image into overlapping tiles, decoding the tiles, and then blending the outputs together to compose the final image. You should also used tiled VAE with [`~ModelMixin.enable_xformers_memory_efficient_attention`] to further reduce memory use.
-
-To use tiled VAE processing, call [`~StableDiffusionPipeline.enable_vae_tiling`] on your pipeline before inference:
-
-```python
-import torch
-from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
-pipe = pipe.to("cuda")
-prompt = "a beautiful landscape photograph"
-pipe.enable_vae_tiling()
-pipe.enable_xformers_memory_efficient_attention()
-
-image = pipe([prompt], width=3840, height=2224, num_inference_steps=20).images[0]
-```
-
-The output image has some tile-to-tile tone variation because the tiles are decoded separately, but you shouldn't see any sharp and obvious seams between the tiles. Tiling is turned off for images that are 512x512 or smaller.
-
-## CPU offloading
-
-Offloading the weights to the CPU and only loading them on the GPU when performing the forward pass can also save memory. Often, this technique can reduce memory consumption to less than 3GB.
-
-To perform CPU offloading, call [`~StableDiffusionPipeline.enable_sequential_cpu_offload`]:
-
-```Python
-import torch
-from diffusers import StableDiffusionPipeline
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-
-prompt = "a photo of an astronaut riding a horse on mars"
-pipe.enable_sequential_cpu_offload()
-image = pipe(prompt).images[0]
-```
-
-CPU offloading works on submodules rather than whole models. This is the best way to minimize memory consumption, but inference is much slower due to the iterative nature of the diffusion process. The UNet component of the pipeline runs several times (as many as `num_inference_steps`); each time, the different UNet submodules are sequentially onloaded and offloaded as needed, resulting in a large number of memory transfers.
-
-<Tip>
-
-Consider using [model offloading](#model-offloading) if you want to optimize for speed because it is much faster. The tradeoff is your memory savings won't be as large.
-
-</Tip>
-
-CPU offloading can also be chained with attention slicing to reduce memory consumption to less than 2GB.
-
-```Python
-import torch
-from diffusers import StableDiffusionPipeline
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-
-prompt = "a photo of an astronaut riding a horse on mars"
-pipe.enable_sequential_cpu_offload()
-
-image = pipe(prompt).images[0]
-```
-
-<Tip warning={true}>
-
-When using [`~StableDiffusionPipeline.enable_sequential_cpu_offload`], don't move the pipeline to CUDA beforehand or else the gain in memory consumption will only be minimal (see this [issue](https://github.com/huggingface/diffusers/issues/1934) for more information).
-
-[`~StableDiffusionPipeline.enable_sequential_cpu_offload`] is a stateful operation that installs hooks on the models.
-
-</Tip>
-
-## Model offloading
-
-<Tip>
-
-Model offloading requires 🤗 Accelerate version 0.17.0 or higher.
-
-</Tip>
-
-[Sequential CPU offloading](#cpu-offloading) preserves a lot of memory but it makes inference slower because submodules are moved to GPU as needed, and they're immediately returned to the CPU when a new module runs.
-
-Full-model offloading is an alternative that moves whole models to the GPU, instead of handling each model's constituent *submodules*. There is a negligible impact on inference time (compared with moving the pipeline to `cuda`), and it still provides some memory savings.
-
-During model offloading, only one of the main components of the pipeline (typically the text encoder, UNet and VAE)
-is placed on the GPU while the others wait on the CPU. Components like the UNet that run for multiple iterations stay on the GPU until they're no longer needed.
-
-Enable model offloading by calling [`~StableDiffusionPipeline.enable_model_cpu_offload`] on the pipeline:
-
-```Python
-import torch
-from diffusers import StableDiffusionPipeline
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",  
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-
-prompt = "a photo of an astronaut riding a horse on mars"
-pipe.enable_model_cpu_offload()
-image = pipe(prompt).images[0]
-```
-
-Model offloading can also be combined with attention slicing for additional memory savings.
-
-```Python
-import torch
-from diffusers import StableDiffusionPipeline
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-)
-
-prompt = "a photo of an astronaut riding a horse on mars"
-pipe.enable_model_cpu_offload()
-
-image = pipe(prompt).images[0]
-```
-
-<Tip warning={true}>
-
-In order to properly offload models after they're called, it is required to run the entire pipeline and models are called in the pipeline's expected order. Exercise caution if models are reused outside the context of the pipeline after hooks have been installed. See [Removing Hooks](https://huggingface.co/docs/accelerate/en/package_reference/big_modeling#accelerate.hooks.remove_hook_from_module)
-for more information.
-
-[`~StableDiffusionPipeline.enable_model_cpu_offload`] is a stateful operation that installs hooks on the models and state on the pipeline.
-
-</Tip>
-
-## Channels-last memory format
-
-The channels-last memory format is an alternative way of ordering NCHW tensors in memory to preserve dimension ordering. Channels-last tensors are ordered in such a way that the channels become the densest dimension (storing images pixel-per-pixel). Since not all operators currently support the channels-last format, it may result in worst performance but you should still try and see if it works for your model.
-
-For example, to set the pipeline's UNet to use the channels-last format:
-
-```python
-print(pipe.unet.conv_out.state_dict()["weight"].stride())  # (2880, 9, 3, 1)
-pipe.unet.to(memory_format=torch.channels_last)  # in-place operation
-print(
-    pipe.unet.conv_out.state_dict()["weight"].stride()
-)  # (2880, 1, 960, 320) having a stride of 1 for the 2nd dimension proves that it works
-```
-
-## Tracing
-
-Tracing runs an example input tensor through the model and captures the operations that are performed on it as that input makes its way through the model's layers. The executable or `ScriptFunction` that is returned is optimized with just-in-time compilation.
-
-To trace a UNet:
-
-```python
-import time
-import torch
-from diffusers import StableDiffusionPipeline
-import functools
-
-# torch disable grad
-torch.set_grad_enabled(False)
-
-# set variables
-n_experiments = 2
-unet_runs_per_experiment = 50
-
-
-# load inputs
-def generate_inputs():
-    sample = torch.randn(2, 4, 64, 64).half().cuda()
-    timestep = torch.rand(1).half().cuda() * 999
-    encoder_hidden_states = torch.randn(2, 77, 768).half().cuda()
-    return sample, timestep, encoder_hidden_states
-
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-).to("cuda")
-unet = pipe.unet
-unet.eval()
-unet.to(memory_format=torch.channels_last)  # use channels_last memory format
-unet.forward = functools.partial(unet.forward, return_dict=False)  # set return_dict=False as default
-
-# warmup
-for _ in range(3):
-    with torch.inference_mode():
-        inputs = generate_inputs()
-        orig_output = unet(*inputs)
-
-# trace
-print("tracing..")
-unet_traced = torch.jit.trace(unet, inputs)
-unet_traced.eval()
-print("done tracing")
-
-
-# warmup and optimize graph
-for _ in range(5):
-    with torch.inference_mode():
-        inputs = generate_inputs()
-        orig_output = unet_traced(*inputs)
-
-
-# benchmarking
-with torch.inference_mode():
-    for _ in range(n_experiments):
-        torch.cuda.synchronize()
-        start_time = time.time()
-        for _ in range(unet_runs_per_experiment):
-            orig_output = unet_traced(*inputs)
-        torch.cuda.synchronize()
-        print(f"unet traced inference took {time.time() - start_time:.2f} seconds")
-    for _ in range(n_experiments):
-        torch.cuda.synchronize()
-        start_time = time.time()
-        for _ in range(unet_runs_per_experiment):
-            orig_output = unet(*inputs)
-        torch.cuda.synchronize()
-        print(f"unet inference took {time.time() - start_time:.2f} seconds")
-
-# save the model
-unet_traced.save("unet_traced.pt")
-```
-
-Replace the `unet` attribute of the pipeline with the traced model:
-
-```python
-from diffusers import StableDiffusionPipeline
-import torch
-from dataclasses import dataclass
-
-
-@dataclass
-class UNet2DConditionOutput:
-    sample: torch.FloatTensor
-
-
-pipe = StableDiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-).to("cuda")
-
-# use jitted unet
-unet_traced = torch.jit.load("unet_traced.pt")
-
-
-# del pipe.unet
-class TracedUNet(torch.nn.Module):
-    def __init__(self):
-        super().__init__()
-        self.in_channels = pipe.unet.in_channels
-        self.device = pipe.unet.device
-
-    def forward(self, latent_model_input, t, encoder_hidden_states):
-        sample = unet_traced(latent_model_input, t, encoder_hidden_states)[0]
-        return UNet2DConditionOutput(sample=sample)
-
-
-pipe.unet = TracedUNet()
-
-with torch.inference_mode():
-    image = pipe([prompt] * 1, num_inference_steps=50).images[0]
-```
-
-## Memory-efficient attention
-
-Recent work on optimizing bandwidth in the attention block has generated huge speed-ups and reductions in GPU memory usage. The most recent type of memory-efficient attention is [Flash Attention](https://arxiv.org/pdf/2205.14135.pdf) (you can check out the original code at [HazyResearch/flash-attention](https://github.com/HazyResearch/flash-attention)).
-
-<Tip>
-
-If you have PyTorch >= 2.0 installed, you should not expect a speed-up for inference when enabling `xformers`.
-
-</Tip>
-
-To use Flash Attention, install the following:
-
-- PyTorch > 1.12
-- CUDA available
-- [xFormers](xformers)
-
-Then call [`~ModelMixin.enable_xformers_memory_efficient_attention`] on the pipeline:
-
-```python
-from diffusers import DiffusionPipeline
-import torch
-
-pipe = DiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5",
-    torch_dtype=torch.float16,
-    use_safetensors=True,
-).to("cuda")
-
-pipe.enable_xformers_memory_efficient_attention()
-
-with torch.inference_mode():
-    sample = pipe("a small cat")
-
-# optional: You can disable it via
-# pipe.disable_xformers_memory_efficient_attention()
-```
-
-The iteration speed when using `xformers` should match the iteration speed of Torch 2.0 as described [here](torch2.0).

docs/source/en/optimization/mps.md

@@ -10,16 +10,29 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Metal Performance Shaders (MPS)
+# How to use Stable Diffusion in Apple Silicon (M1/M2)
 
-🤗 Diffusers is compatible with Apple silicon (M1/M2 chips) using the PyTorch [`mps`](https://pytorch.org/docs/stable/notes/mps.html) device, which uses the Metal framework to leverage the GPU on MacOS devices. You'll need to have:
+🤗 Diffusers is compatible with Apple silicon for Stable Diffusion inference, using the PyTorch `mps` device. These are the steps you need to follow to use your M1 or M2 computer with Stable Diffusion.
 
-- macOS computer with Apple silicon (M1/M2) hardware
-- macOS 12.6 or later (13.0 or later recommended)
-- arm64 version of Python
-- [PyTorch 2.0](https://pytorch.org/get-started/locally/) (recommended) or 1.13 (minimum version supported for `mps`)
+## Requirements
 
-The `mps` backend uses PyTorch's `.to()` interface to move the Stable Diffusion pipeline on to your M1 or M2 device:
+- Mac computer with Apple silicon (M1/M2) hardware.
+- macOS 12.6 or later (13.0 or later recommended).
+- arm64 version of Python.
+- PyTorch 2.0 (recommended) or 1.13 (minimum version supported for `mps`). You can install it with `pip` or `conda` using the instructions in https://pytorch.org/get-started/locally/.
+
+
+## Inference Pipeline
+
+The snippet below demonstrates how to use the `mps` backend using the familiar `to()` interface to move the Stable Diffusion pipeline to your M1 or M2 device.
+
+<Tip warning={true}>
+
+**If you are using PyTorch 1.13** you need to "prime" the pipeline using an additional one-time pass through it. This is a temporary workaround for a weird issue we detected: the first inference pass produces slightly different results than subsequent ones. You only need to do this pass once, and it's ok to use just one inference step and discard the result.
+
+</Tip>
+
+We strongly recommend you use PyTorch 2 or better, as it solves a number of problems like the one described in the previous tip.
 
 ```python
 from diffusers import DiffusionPipeline
@@ -31,41 +44,24 @@ pipe = pipe.to("mps")
 pipe.enable_attention_slicing()
 
 prompt = "a photo of an astronaut riding a horse on mars"
-```
 
-<Tip warning={true}>
-
-Generating multiple prompts in a batch can [crash](https://github.com/huggingface/diffusers/issues/363) or fail to work reliably. We believe this is related to the [`mps`](https://github.com/pytorch/pytorch/issues/84039) backend in PyTorch. While this is being investigated, you should iterate instead of batching.
-
-</Tip>
-
-If you're using **PyTorch 1.13**, you need to "prime" the pipeline with an additional one-time pass through it. This is a temporary workaround for an issue where the first inference pass produces slightly different results than subsequent ones. You only need to do this pass once, and after just one inference step you can discard the result.
-
-```diff
-  from diffusers import DiffusionPipeline
-
-  pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5").to("mps")
-  pipe.enable_attention_slicing()
-
-  prompt = "a photo of an astronaut riding a horse on mars"
-# First-time "warmup" pass if PyTorch version is 1.13
-+ _ = pipe(prompt, num_inference_steps=1)
+# First-time "warmup" pass if PyTorch version is 1.13 (see explanation above)
+_ = pipe(prompt, num_inference_steps=1)
 
 # Results match those from the CPU device after the warmup pass.
-  image = pipe(prompt).images[0]
+image = pipe(prompt).images[0]
 ```
 
-## Troubleshoot
+## Performance Recommendations
 
-M1/M2 performance is very sensitive to memory pressure. When this occurs, the system automatically swaps if it needs to which significantly degrades performance.
+M1/M2 performance is very sensitive to memory pressure. The system will automatically swap if it needs to, but performance will degrade significantly when it does.
 
-To prevent this from happening, we recommend *attention slicing* to reduce memory pressure during inference and prevent swapping. This is especially relevant if your computer has less than 64GB of system RAM, or if you generate images at non-standard resolutions larger than 512×512 pixels. Call the [`~DiffusionPipeline.enable_attention_slicing`] function on your pipeline:
+We recommend you use _attention slicing_ to reduce memory pressure during inference and prevent swapping, particularly if your computer has less than 64 GB of system RAM, or if you generate images at non-standard resolutions larger than 512 × 512 pixels. Attention slicing performs the costly attention operation in multiple steps instead of all at once. It usually has a performance impact of ~20% in computers without universal memory, but we have observed _better performance_ in most Apple Silicon computers, unless you have 64 GB or more.
 
-```py
-from diffusers import DiffusionPipeline
-
-pipeline = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True).to("mps")
+```python
 pipeline.enable_attention_slicing()
 ```
 
-Attention slicing performs the costly attention operation in multiple steps instead of all at once. It usually improves performance by ~20% in computers without universal memory, but we've observed *better performance* in most Apple silicon computers unless you have 64GB of RAM or more.
+## Known Issues
+
+- Generating multiple prompts in a batch [crashes or doesn't work reliably](https://github.com/huggingface/diffusers/issues/363). We believe this is related to the [`mps` backend in PyTorch](https://github.com/pytorch/pytorch/issues/84039). This is being resolved, but for now we recommend to iterate instead of batching.

docs/source/en/optimization/onnx.md

@@ -11,19 +11,23 @@ specific language governing permissions and limitations under the License.
 -->
 
 
-# ONNX Runtime
+# How to use ONNX Runtime for inference
 
-🤗 [Optimum](https://github.com/huggingface/optimum) provides a Stable Diffusion pipeline compatible with ONNX Runtime. You'll need to install 🤗 Optimum with the following command for ONNX Runtime support:
+🤗 [Optimum](https://github.com/huggingface/optimum) provides a Stable Diffusion pipeline compatible with ONNX Runtime. 
 
-```bash
+## Installation
+
+Install 🤗 Optimum with the following command for ONNX Runtime support:
+
+```
 pip install optimum["onnxruntime"]
 ```
 
-This guide will show you how to use the Stable Diffusion and Stable Diffusion XL (SDXL) pipelines with ONNX Runtime.
-
 ## Stable Diffusion
 
-To load and run inference, use the [`~optimum.onnxruntime.ORTStableDiffusionPipeline`]. If you want to load a PyTorch model and convert it to the ONNX format on-the-fly, set `export=True`:
+### Inference
+
+To load an ONNX model and run inference with ONNX Runtime, you need to replace [`StableDiffusionPipeline`] with `ORTStableDiffusionPipeline`. In case you want to load a PyTorch model and convert it to the ONNX format on-the-fly, you can set `export=True`.
 
 ```python
 from optimum.onnxruntime import ORTStableDiffusionPipeline
@@ -35,20 +39,14 @@ image = pipeline(prompt).images[0]
 pipeline.save_pretrained("./onnx-stable-diffusion-v1-5")
 ```
 
-<Tip warning={true}>
-
-Generating multiple prompts in a batch seems to take too much memory. While we look into it, you may need to iterate instead of batching.
-
-</Tip>
-
-To export the pipeline in the ONNX format offline and use it later for inference,
-use the [`optimum-cli export`](https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#exporting-a-model-to-onnx-using-the-cli) command:
+If you want to export the pipeline in the ONNX format offline and later use it for inference,
+you can use the [`optimum-cli export`](https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#exporting-a-model-to-onnx-using-the-cli) command: 
 
 ```bash
 optimum-cli export onnx --model runwayml/stable-diffusion-v1-5 sd_v15_onnx/
 ```
 
-Then to perform inference (you don't have to specify `export=True` again):
+Then perform inference:
 
 ```python 
 from optimum.onnxruntime import ORTStableDiffusionPipeline
@@ -59,15 +57,36 @@ prompt = "sailing ship in storm by Leonardo da Vinci"
 image = pipeline(prompt).images[0]
 ```
 
+Notice that we didn't have to specify `export=True` above.
+
 <div class="flex justify-center">
     <img src="https://huggingface.co/datasets/optimum/documentation-images/resolve/main/onnxruntime/stable_diffusion_v1_5_ort_sail_boat.png">
 </div>
 
-You can find more examples in 🤗 Optimum [documentation](https://huggingface.co/docs/optimum/), and Stable Diffusion is supported for text-to-image, image-to-image, and inpainting.
+You can find more examples in [optimum documentation](https://huggingface.co/docs/optimum/).
+
+
+### Supported tasks
+
+| Task                                 | Loading Class                        |
+|--------------------------------------|--------------------------------------|
+| `text-to-image`                      | `ORTStableDiffusionPipeline`         |
+| `image-to-image`                     | `ORTStableDiffusionImg2ImgPipeline`  |
+| `inpaint`                            | `ORTStableDiffusionInpaintPipeline`  |
 
 ## Stable Diffusion XL
 
-To load and run inference with SDXL, use the [`~optimum.onnxruntime.ORTStableDiffusionXLPipeline`]:
+### Export
+
+To export your model to ONNX, you can use the [Optimum CLI](https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#exporting-a-model-to-onnx-using-the-cli) as follows :
+
+```bash
+optimum-cli export onnx --model stabilityai/stable-diffusion-xl-base-1.0 --task stable-diffusion-xl sd_xl_onnx/
+```
+
+### Inference
+
+Here is an example of how you can load a SDXL ONNX model from [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) and run inference with ONNX Runtime :
 
 ```python
 from optimum.onnxruntime import ORTStableDiffusionXLPipeline
@@ -78,10 +97,13 @@ prompt = "sailing ship in storm by Leonardo da Vinci"
 image = pipeline(prompt).images[0]
 ```
 
-To export the pipeline in the ONNX format and use it later for inference, use the [`optimum-cli export`](https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#exporting-a-model-to-onnx-using-the-cli) command:
+### Supported tasks
 
-```bash
-optimum-cli export onnx --model stabilityai/stable-diffusion-xl-base-1.0 --task stable-diffusion-xl sd_xl_onnx/
-```
+| Task                                 | Loading Class                        |
+|--------------------------------------|--------------------------------------|
+| `text-to-image`                      | `ORTStableDiffusionXLPipeline`       |
+| `image-to-image`                     | `ORTStableDiffusionXLImg2ImgPipeline`|
 
-SDXL in the ONNX format is supported for text-to-image and image-to-image.
+## Known Issues
+
+- Generating multiple prompts in a batch seems to take too much memory. While we look into it, you may need to iterate instead of batching.

docs/source/en/optimization/open_vino.md

@@ -11,21 +11,26 @@ specific language governing permissions and limitations under the License.
 -->
 
 
-# OpenVINO
+# How to use OpenVINO for inference
 
-🤗 [Optimum](https://github.com/huggingface/optimum-intel) provides Stable Diffusion pipelines compatible with OpenVINO to perform inference on a variety of Intel processors (see the [full list]((https://docs.openvino.ai/latest/openvino_docs_OV_UG_supported_plugins_Supported_Devices.html)) of supported devices).
+🤗 [Optimum](https://github.com/huggingface/optimum-intel) provides Stable Diffusion pipelines compatible with OpenVINO. You can now easily perform inference with OpenVINO Runtime on a variety of Intel processors ([see](https://docs.openvino.ai/latest/openvino_docs_OV_UG_supported_plugins_Supported_Devices.html) the full list of supported devices).
 
-You'll need to install 🤗 Optimum Intel with the `--upgrade-strategy eager` option to ensure [`optimum-intel`](https://github.com/huggingface/optimum-intel) is using the latest version:
+## Installation
+
+Install 🤗 Optimum Intel with the following command:
 
 ```
 pip install --upgrade-strategy eager optimum["openvino"]
 ```
 
-This guide will show you how to use the Stable Diffusion and Stable Diffusion XL (SDXL) pipelines with OpenVINO.
+The `--upgrade-strategy eager` option is needed to ensure [`optimum-intel`](https://github.com/huggingface/optimum-intel) is upgraded to its latest version.
+
 
 ## Stable Diffusion
 
-To load and run inference, use the [`~optimum.intel.OVStableDiffusionPipeline`]. If you want to load a PyTorch model and convert it to the OpenVINO format on-the-fly, set `export=True`:
+### Inference
+
+To load an OpenVINO model and run inference with OpenVINO Runtime, you need to replace `StableDiffusionPipeline` with `OVStableDiffusionPipeline`. In case you want to load a PyTorch model and convert it to the OpenVINO format on-the-fly, you can set `export=True`.
 
 ```python
 from optimum.intel import OVStableDiffusionPipeline
@@ -39,7 +44,7 @@ image = pipeline(prompt).images[0]
 pipeline.save_pretrained("openvino-sd-v1-5")
 ```
 
-To further speed-up inference, statically reshape the model. If you change any parameters such as the outputs height or width, you’ll need to statically reshape your model again.
+To further speed up inference, the model can be statically reshaped :
 
 ```python
 # Define the shapes related to the inputs and desired outputs
@@ -57,15 +62,30 @@ image = pipeline(
     num_images_per_prompt=num_images,
 ).images[0]
 ```
+
+In case you want to change any parameters such as the outputs height or width, you’ll need to statically reshape your model once again.
+
 <div class="flex justify-center">
     <img src="https://huggingface.co/datasets/optimum/documentation-images/resolve/main/intel/openvino/stable_diffusion_v1_5_sail_boat_rembrandt.png">
 </div>
 
-You can find more examples in the 🤗 Optimum [documentation](https://huggingface.co/docs/optimum/intel/inference#stable-diffusion), and Stable Diffusion is supported for text-to-image, image-to-image, and inpainting.
+
+### Supported tasks
+
+| Task                                 | Loading Class                        |
+|--------------------------------------|--------------------------------------|
+| `text-to-image`                      | `OVStableDiffusionPipeline`          |
+| `image-to-image`                     | `OVStableDiffusionImg2ImgPipeline`   |
+| `inpaint`                            | `OVStableDiffusionInpaintPipeline`   |
+
+You can find more examples in the optimum [documentation](https://huggingface.co/docs/optimum/intel/inference#stable-diffusion).
+
 
 ## Stable Diffusion XL
 
-To load and run inference with SDXL, use the [`~optimum.intel.OVStableDiffusionXLPipeline`]:
+### Inference
+
+Here is an example of how you can load a SDXL OpenVINO model from [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0) and run inference with OpenVINO Runtime :
 
 ```python
 from optimum.intel import OVStableDiffusionXLPipeline
@@ -76,6 +96,15 @@ prompt = "sailing ship in storm by Rembrandt"
 image = pipeline(prompt).images[0]
 ```
 
-To further speed-up inference, [statically reshape](#stable-diffusion) the model as shown in the Stable Diffusion section.
+To further speed up inference, the model can be statically reshaped as showed above.
+You can find more examples in the optimum [documentation](https://huggingface.co/docs/optimum/intel/inference#stable-diffusion-xl).
+
+### Supported tasks
+
+| Task                                 | Loading Class                        |
+|--------------------------------------|--------------------------------------|
+| `text-to-image`                      | `OVStableDiffusionXLPipeline`        |
+| `image-to-image`                     | `OVStableDiffusionXLImg2ImgPipeline` |
+
+
 
-You can find more examples in the 🤗 Optimum [documentation](https://huggingface.co/docs/optimum/intel/inference#stable-diffusion-xl), and running SDXL in OpenVINO is supported for text-to-image and image-to-image.

docs/source/en/optimization/opt_overview.md

@@ -12,6 +12,6 @@ specific language governing permissions and limitations under the License.
 
 # Overview
 
-Generating high-quality outputs is computationally intensive, especially during each iterative step where you go from a noisy output to a less noisy output. One of 🤗 Diffuser's goal is to make this technology widely accessible to everyone, which includes enabling fast inference on consumer and specialized hardware.
+Generating high-quality outputs is computationally intensive, especially during each iterative step where you go from a noisy output to a less noisy output. One of 🧨 Diffuser's goal is to make this technology widely accessible to everyone, which includes enabling fast inference on consumer and specialized hardware. 
 
-This section will cover tips and tricks - like half-precision weights and sliced attention - for optimizing inference speed and reducing memory-consumption. You'll also learn how to speed up your PyTorch code with [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) or [ONNX Runtime](https://onnxruntime.ai/docs/), and enable memory-efficient attention with [xFormers](https://facebookresearch.github.io/xformers/). There are also guides for running inference on specific hardware like Apple Silicon, and Intel or Habana processors.
+This section will cover tips and tricks - like half-precision weights and sliced attention - for optimizing inference speed and reducing memory-consumption. You can also learn how to speed up your PyTorch code with [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) or [ONNX Runtime](https://onnxruntime.ai/docs/), and enable memory-efficient attention with [xFormers](https://facebookresearch.github.io/xformers/). There are also guides for running inference on specific hardware like Apple Silicon, and Intel or Habana processors.

docs/source/en/optimization/tome.md

@@ -10,39 +10,35 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Token merging
+# Token Merging
 
-[Token merging](https://huggingface.co/papers/2303.17604) (ToMe) merges redundant tokens/patches progressively in the forward pass of a Transformer-based network which can speed-up the inference latency of [`StableDiffusionPipeline`].
+Token Merging (introduced in [Token Merging: Your ViT But Faster](https://arxiv.org/abs/2210.09461)) works by merging the redundant tokens / patches progressively in the forward pass of a Transformer-based network. It can speed up the inference latency of the underlying network.
 
-You can use ToMe from the [`tomesd`](https://github.com/dbolya/tomesd) library with the [`apply_patch`](https://github.com/dbolya/tomesd?tab=readme-ov-file#usage) function:
+After Token Merging (ToMe) was released, the authors released [Token Merging for Fast Stable Diffusion](https://arxiv.org/abs/2303.17604), which introduced a version of ToMe which is more compatible with Stable Diffusion. We can use ToMe to gracefully speed up the inference latency of a [`DiffusionPipeline`]. This doc discusses how to apply ToMe to the [`StableDiffusionPipeline`], the expected speedups, and the qualitative aspects of using ToMe on the [`StableDiffusionPipeline`]. 
+
+## Using ToMe
+
+The authors of ToMe released a convenient Python library called [`tomesd`](https://github.com/dbolya/tomesd) that lets us apply ToMe to a [`DiffusionPipeline`] like so:
 
 ```diff
 from diffusers import StableDiffusionPipeline
 import tomesd
 
 pipeline = StableDiffusionPipeline.from_pretrained(
-      "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True,
+      "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
 ).to("cuda")
 + tomesd.apply_patch(pipeline, ratio=0.5)
 
 image = pipeline("a photo of an astronaut riding a horse on mars").images[0]
 ```
 
-The `apply_patch` function exposes a number of [arguments](https://github.com/dbolya/tomesd#usage) to help strike a balance between pipeline inference speed and the quality of the generated tokens. The most important argument is `ratio` which controls the number of tokens that are merged during the forward pass.
+And that’s it! 
 
-As reported in the [paper](https://huggingface.co/papers/2303.17604), ToMe can greatly preserve the quality of the generated images while boosting inference speed. By increasing the `ratio`, you can speed-up inference even further, but at the cost of some degraded image quality.
+`tomesd.apply_patch()` exposes [a number of arguments](https://github.com/dbolya/tomesd#usage) to let us strike a balance between the pipeline inference speed and the quality of the generated tokens. Amongst those arguments, the most important one is `ratio`. `ratio` controls the number of tokens that will be merged during the forward pass. For more details on `tomesd`, please refer to the original repository https://github.com/dbolya/tomesd and [the paper](https://arxiv.org/abs/2303.17604). 
 
-To test the quality of the generated images, we sampled a few prompts from [Parti Prompts](https://parti.research.google/) and performed inference with the [`StableDiffusionPipeline`] with the following settings:
+## Benchmarking `tomesd` with `StableDiffusionPipeline`
 
-<div class="flex justify-center">
-      <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/tome/tome_samples.png">
-</div>
-
-We didn’t notice any significant decrease in the quality of the generated samples, and you can check out the generated samples in this [WandB report](https://wandb.ai/sayakpaul/tomesd-results/runs/23j4bj3i?workspace=). If you're interested in reproducing this experiment, use this [script](https://gist.github.com/sayakpaul/8cac98d7f22399085a060992f411ecbd).
-
-## Benchmarks
-
-We also benchmarked the impact of `tomesd` on the [`StableDiffusionPipeline`] with [xFormers](https://huggingface.co/docs/diffusers/optimization/xformers) enabled across several image resolutions. The results are obtained from A100 and V100 GPUs in the following development environment:
+We benchmarked the impact of using `tomesd` on [`StableDiffusionPipeline`] along with [xformers](https://huggingface.co/docs/diffusers/optimization/xformers) across different image resolutions. We used A100 and V100 as our test GPU devices with the following development environment (with Python 3.8.5):
 
 ```bash
 - `diffusers` version: 0.15.1
@@ -55,35 +51,66 @@ We also benchmarked the impact of `tomesd` on the [`StableDiffusionPipeline`] wi
 - tomesd version: 0.1.2
 ```
 
-To reproduce this benchmark, feel free to use this [script](https://gist.github.com/sayakpaul/27aec6bca7eb7b0e0aa4112205850335). The results are reported in seconds, and where applicable we report the speed-up percentage over the vanilla pipeline when using ToMe and ToMe + xFormers.
+We used this script for benchmarking: [https://gist.github.com/sayakpaul/27aec6bca7eb7b0e0aa4112205850335](https://gist.github.com/sayakpaul/27aec6bca7eb7b0e0aa4112205850335). Following are our findings: 
 
-| **GPU**  | **Resolution** | **Batch size** | **Vanilla** | **ToMe**       | **ToMe + xFormers** |
-|----------|----------------|----------------|-------------|----------------|---------------------|
-| **A100** |            512 |             10 |        6.88 | 5.26 (+23.55%) |      4.69 (+31.83%) |
-|          |            768 |             10 |         OOM |          14.71 |                  11 |
-|          |                |              8 |         OOM |          11.56 |                8.84 |
-|          |                |              4 |         OOM |           5.98 |                4.66 |
-|          |                |              2 |        4.99 | 3.24 (+35.07%) |       2.1 (+37.88%) |
-|          |                |              1 |        3.29 | 2.24 (+31.91%) |       2.03 (+38.3%) |
-|          |           1024 |             10 |         OOM |            OOM |                 OOM |
-|          |                |              8 |         OOM |            OOM |                 OOM |
-|          |                |              4 |         OOM |          12.51 |                9.09 |
-|          |                |              2 |         OOM |           6.52 |                4.96 |
-|          |                |              1 |         6.4 | 3.61 (+43.59%) |      2.81 (+56.09%) |
-| **V100** |            512 |             10 |         OOM |          10.03 |                9.29 |
-|          |                |              8 |         OOM |           8.05 |                7.47 |
-|          |                |              4 |         5.7 |  4.3 (+24.56%) |      3.98 (+30.18%) |
-|          |                |              2 |        3.14 | 2.43 (+22.61%) |      2.27 (+27.71%) |
-|          |                |              1 |        1.88 | 1.57 (+16.49%) |      1.57 (+16.49%) |
-|          |            768 |             10 |         OOM |            OOM |               23.67 |
-|          |                |              8 |         OOM |            OOM |               18.81 |
-|          |                |              4 |         OOM |          11.81 |                 9.7 |
-|          |                |              2 |         OOM |           6.27 |                 5.2 |
-|          |                |              1 |        5.43 | 3.38 (+37.75%) |      2.82 (+48.07%) |
-|          |           1024 |             10 |         OOM |            OOM |                 OOM |
-|          |                |              8 |         OOM |            OOM |                 OOM |
-|          |                |              4 |         OOM |            OOM |               19.35 |
-|          |                |              2 |         OOM |             13 |               10.78 |
-|          |                |              1 |         OOM |           6.66 |                5.54 |
+### A100
 
-As seen in the tables above, the speed-up from `tomesd` becomes more pronounced for larger image resolutions. It is also interesting to note that with `tomesd`, it is possible to run the pipeline on a higher resolution like 1024x1024. You may be able to speed-up inference even more with [`torch.compile`](torch2.0).
+| Resolution | Batch size | Vanilla | ToMe | ToMe + xFormers | ToMe speedup (%) | ToMe + xFormers speedup (%) |
+| --- | --- | --- | --- | --- | --- | --- |
+| 512 | 10 | 6.88 | 5.26 | 4.69 | 23.54651163 | 31.83139535 |
+|  |  |  |  |  |  |  |
+| 768 | 10 | OOM | 14.71 | 11 |  |  |
+|  | 8 | OOM | 11.56 | 8.84 |  |  |
+|  | 4 | OOM | 5.98 | 4.66 |  |  |
+|  | 2 | 4.99 | 3.24 | 3.1 | 35.07014028 | 37.8757515 |
+|  | 1 | 3.29 | 2.24 | 2.03 | 31.91489362 | 38.29787234 |
+|  |  |  |  |  |  |  |
+| 1024 | 10 | OOM | OOM | OOM |  |  |
+|  | 8 | OOM | OOM | OOM |  |  |
+|  | 4 | OOM | 12.51 | 9.09 |  |  |
+|  | 2 | OOM | 6.52 | 4.96 |  |  |
+|  | 1 | 6.4 | 3.61 | 2.81 | 43.59375 | 56.09375 |
+
+***The timings reported here are in seconds. Speedups are calculated over the `Vanilla` timings.*** 
+
+### V100
+
+| Resolution | Batch size | Vanilla | ToMe | ToMe + xFormers | ToMe speedup (%) | ToMe + xFormers speedup (%) |
+| --- | --- | --- | --- | --- | --- | --- |
+| 512 | 10 | OOM | 10.03 | 9.29 |  |  |
+|  | 8 | OOM | 8.05 | 7.47 |  |  |
+|  | 4 | 5.7 | 4.3 | 3.98 | 24.56140351 | 30.1754386 |
+|  | 2 | 3.14 | 2.43 | 2.27 | 22.61146497 | 27.70700637 |
+|  | 1 | 1.88 | 1.57 | 1.57 | 16.4893617 | 16.4893617 |
+|  |  |  |  |  |  |  |
+| 768 | 10 | OOM | OOM | 23.67 |  |  |
+|  | 8 | OOM | OOM | 18.81 |  |  |
+|  | 4 | OOM | 11.81 | 9.7 |  |  |
+|  | 2 | OOM | 6.27 | 5.2 |  |  |
+|  | 1 | 5.43 | 3.38 | 2.82 | 37.75322284 | 48.06629834 |
+|  |  |  |  |  |  |  |
+| 1024 | 10 | OOM | OOM | OOM |  |  |
+|  | 8 | OOM | OOM | OOM |  |  |
+|  | 4 | OOM | OOM | 19.35 |  |  |
+|  | 2 | OOM | 13 | 10.78 |  |  |
+|  | 1 | OOM | 6.66 | 5.54 |  |  |
+
+As seen in the tables above, the speedup with `tomesd` becomes more pronounced for larger image resolutions. It is also interesting to note that with `tomesd`, it becomes possible to run the pipeline on a higher resolution, like 1024x1024. 
+
+It might be possible to speed up inference even further with [`torch.compile()`](https://huggingface.co/docs/diffusers/optimization/torch2.0). 
+
+## Quality
+
+As reported in [the paper](https://arxiv.org/abs/2303.17604), ToMe can preserve the quality of the generated images to a great extent while speeding up inference. By increasing the `ratio`, it is possible to further speed up inference, but that might come at the cost of a deterioration in the image quality. 
+
+To test the quality of the generated samples using our setup, we sampled a few prompts from the “Parti Prompts” (introduced in [Parti](https://parti.research.google/)) and performed inference with the [`StableDiffusionPipeline`] in the following settings:
+
+- Vanilla [`StableDiffusionPipeline`]
+- [`StableDiffusionPipeline`] + ToMe
+- [`StableDiffusionPipeline`] + ToMe + xformers
+
+We didn’t notice any significant decrease in the quality of the generated samples. Here are samples: 
+
+![tome-samples](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/tome/tome_samples.png)
+
+You can check out the generated samples [here](https://wandb.ai/sayakpaul/tomesd-results/runs/23j4bj3i?workspace=). We used [this script](https://gist.github.com/sayakpaul/8cac98d7f22399085a060992f411ecbd) for conducting this experiment.

docs/source/en/optimization/torch2.0.md

@@ -10,83 +10,96 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Torch 2.0
+# Accelerated PyTorch 2.0 support in Diffusers
 
-🤗 Diffusers supports the latest optimizations from [PyTorch 2.0](https://pytorch.org/get-started/pytorch-2.0/) which include:
+Starting from version `0.13.0`, Diffusers supports the latest optimization from [PyTorch 2.0](https://pytorch.org/get-started/pytorch-2.0/). These include:
+1. Support for accelerated transformers implementation with memory-efficient attention – no extra dependencies (such as `xformers`) required.
+2. [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) support for extra performance boost when individual models are compiled.
 
-1. A memory-efficient attention implementation, scaled dot product attention, without requiring any extra dependencies such as xFormers.
-2. [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html), a just-in-time (JIT) compiler to provide an extra performance boost when individual models are compiled.
 
-Both of these optimizations require PyTorch 2.0 or later and 🤗 Diffusers > 0.13.0.
+## Installation
+
+To benefit from the accelerated attention implementation and `torch.compile()`, you just need to install the latest versions of PyTorch 2.0 from pip, and make sure you are on diffusers 0.13.0 or later. As explained below, diffusers automatically uses the optimized attention processor ([`AttnProcessor2_0`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L798)) (but not `torch.compile()`)
+when PyTorch 2.0 is available.
 
 ```bash
 pip install --upgrade torch diffusers
 ```
 
-## Scaled dot product attention
+## Using accelerated transformers and `torch.compile`.
 
-[`torch.nn.functional.scaled_dot_product_attention`](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) (SDPA) is an optimized and memory-efficient attention (similar to xFormers) that automatically enables several other optimizations depending on the model inputs and GPU type. SDPA is enabled by default if you're using PyTorch 2.0 and the latest version of 🤗 Diffusers, so you don't need to add anything to your code.
 
-However, if you want to explicitly enable it, you can set a [`DiffusionPipeline`] to use [`~models.attention_processor.AttnProcessor2_0`]:
+1. **Accelerated Transformers implementation**
 
-```diff
-  import torch
-  from diffusers import DiffusionPipeline
-+ from diffusers.models.attention_processor import AttnProcessor2_0
+   PyTorch 2.0 includes an optimized and memory-efficient attention implementation through the [`torch.nn.functional.scaled_dot_product_attention`](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) function, which automatically enables several optimizations depending on the inputs and the GPU type. This is similar to the `memory_efficient_attention` from [xFormers](https://github.com/facebookresearch/xformers), but built natively into PyTorch. 
 
-  pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
-+ pipe.unet.set_attn_processor(AttnProcessor2_0())
+   These optimizations will be enabled by default in Diffusers if PyTorch 2.0 is installed and if `torch.nn.functional.scaled_dot_product_attention` is available. To use it, just install `torch 2.0` as suggested above and simply use the pipeline. For example:
 
-  prompt = "a photo of an astronaut riding a horse on mars"
-  image = pipe(prompt).images[0]
-```
+    ```Python
+    import torch
+    from diffusers import DiffusionPipeline
 
-SDPA should be as fast and memory efficient as `xFormers`; check the [benchmark](#benchmark) for more details.
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True)
+    pipe = pipe.to("cuda")
 
-In some cases - such as making the pipeline more deterministic or converting it to other formats - it may be helpful to use the vanilla attention processor, [`~models.attention_processor.AttnProcessor`]. To revert to [`~models.attention_processor.AttnProcessor`], call the [`~UNet2DConditionModel.set_default_attn_processor`] function on the pipeline:
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
 
-```diff
-  import torch
-  from diffusers import DiffusionPipeline
-  from diffusers.models.attention_processor import AttnProcessor
+    If you want to enable it explicitly (which is not required), you can do so as shown below.
 
-  pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
-+ pipe.unet.set_default_attn_processor()
+    ```diff
+    import torch
+    from diffusers import DiffusionPipeline
+    + from diffusers.models.attention_processor import AttnProcessor2_0
 
-  prompt = "a photo of an astronaut riding a horse on mars"
-  image = pipe(prompt).images[0]
-```
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
+    + pipe.unet.set_attn_processor(AttnProcessor2_0())
 
-## torch.compile
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
 
-The `torch.compile` function can often provide an additional speed-up to your PyTorch code. In 🤗 Diffusers, it is usually best to wrap the UNet with `torch.compile` because it does most of the heavy lifting in the pipeline.
+    This should be as fast and memory efficient as `xFormers`. More details [in our benchmark](#benchmark).
 
-```python
-from diffusers import DiffusionPipeline
-import torch
+    It is possible to revert to the vanilla attention processor ([`AttnProcessor`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L402)), which can be helpful to make the pipeline more deterministic, or if you need to convert a fine-tuned model to other formats such as [Core ML](https://huggingface.co/docs/diffusers/v0.16.0/en/optimization/coreml#how-to-run-stable-diffusion-with-core-ml). To use the normal attention processor you can use the [`~diffusers.UNet2DConditionModel.set_default_attn_processor`] function:
 
-pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
-pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
-images = pipe(prompt, num_inference_steps=steps, num_images_per_prompt=batch_size).images[0]
-```
+    ```Python
+    import torch
+    from diffusers import DiffusionPipeline
+    from diffusers.models.attention_processor import AttnProcessor
 
-Depending on GPU type, `torch.compile` can provide an *addtional speed-up* of **5-300x** on top of SDPA! If you're using more recent GPU architectures such as Ampere (A100, 3090), Ada (4090), and Hopper (H100), `torch.compile` is able to squeeze even more performance out of these GPUs.
+    pipe = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
+    pipe.unet.set_default_attn_processor()
 
-Compilation requires some time to complete, so it is best suited for situations where you prepare your pipeline once and then perform the same type of inference operations multiple times. For example, calling the compiled pipeline on a different image size triggers compilation again which can be expensive.
+    prompt = "a photo of an astronaut riding a horse on mars"
+    image = pipe(prompt).images[0]
+    ```
+
+2. **torch.compile**
+
+    To get an additional speedup, we can use the new `torch.compile` feature. Since the UNet of the pipeline is usually the most computationally expensive, we wrap the `unet` with `torch.compile` leaving rest of the sub-models (text encoder and VAE) as is. For more information and different options, refer to the 
+    [torch compile docs](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html).
+
+    ```python
+    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+    images = pipe(prompt, num_inference_steps=steps, num_images_per_prompt=batch_size).images
+    ```
+
+    Depending on the type of GPU, `compile()` can yield between **5% - 300%** of _additional speed-up_ over the accelerated transformer optimizations. Note, however, that compilation is able to squeeze more performance improvements in more recent GPU architectures such as Ampere (A100, 3090), Ada (4090) and Hopper (H100).
+    
+    Compilation takes some time to complete, so it is best suited for situations where you need to prepare your pipeline once and then perform the same type of inference operations multiple times. Calling the compiled pipeline on a different image size will re-trigger compilation which can be expensive.
 
-For more information and different options about `torch.compile`, refer to the [`torch_compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) tutorial.
 
 ## Benchmark
 
-We conducted a comprehensive benchmark with PyTorch 2.0's efficient attention implementation and `torch.compile` across different GPUs and batch sizes for five of our most used pipelines. The code is benchmarked on 🤗 Diffusers v0.17.0.dev0 to optimize `torch.compile` usage (see [here](https://github.com/huggingface/diffusers/pull/3313) for more details).
+We conducted a comprehensive benchmark with PyTorch 2.0's efficient attention implementation and `torch.compile` across different GPUs and batch sizes for five of our most used pipelines. We used `diffusers 0.17.0.dev0`, which [makes sure `torch.compile()` is leveraged optimally](https://github.com/huggingface/diffusers/pull/3313).
 
-Expand the dropdown below to find the code used to benchmark each pipeline:
+### Benchmarking code 
 
-<details>
+#### Stable Diffusion text-to-image 
 
-### Stable Diffusion text-to-image
-
-```python
+```python 
 from diffusers import DiffusionPipeline
 import torch
 
@@ -108,7 +121,7 @@ for _ in range(3):
     images = pipe(prompt=prompt).images
 ```
 
-### Stable Diffusion image-to-image
+#### Stable Diffusion image-to-image 
 
 ```python 
 from diffusers import StableDiffusionImg2ImgPipeline
@@ -141,7 +154,7 @@ for _ in range(3):
     image = pipe(prompt=prompt, image=init_image).images[0]
 ```
 
-### Stable Diffusion inpainting
+#### Stable Diffusion - inpainting
 
 ```python 
 from diffusers import StableDiffusionInpaintPipeline
@@ -181,7 +194,7 @@ for _ in range(3):
     image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
 ```
 
-### ControlNet
+#### ControlNet 
 
 ```python 
 from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
@@ -219,7 +232,7 @@ for _ in range(3):
     image = pipe(prompt=prompt, image=init_image).images[0]
 ```
 
-### DeepFloyd IF text-to-image + upscaling
+#### IF text-to-image + upscaling
 
 ```python 
 from diffusers import DiffusionPipeline
@@ -254,18 +267,24 @@ for _ in range(3):
     image_2 = pipe_2(image=image, prompt_embeds=prompt_embeds, negative_prompt_embeds=neg_prompt_embeds, output_type="pt").images
     image_3 = pipe_3(prompt=prompt, image=image, noise_level=100).images
 ```
-</details>
 
-The graph below highlights the relative speed-ups for the [`StableDiffusionPipeline`] across five GPU families with PyTorch 2.0 and `torch.compile` enabled. The benchmarks for the following graphs are measured in *number of iterations/second*.
+To give you a pictorial overview of the possible speed-ups that can be obtained with PyTorch 2.0 and `torch.compile()`,
+here is a plot that shows relative speed-ups for the [Stable Diffusion text-to-image pipeline](StableDiffusionPipeline) across five
+different GPU families (with a batch size of 4):
 
 ![t2i_speedup](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/t2i_speedup.png)
 
-To give you an even better idea of how this speed-up holds for the other pipelines, consider the following
-graph for an A100 with PyTorch 2.0 and `torch.compile`:
+To give you an even better idea of how this speed-up holds for the other pipelines presented above, consider the following 
+plot that shows the benchmarking numbers from an A100 across three different batch sizes
+(with PyTorch 2.0 nightly and `torch.compile()`):
 
 ![a100_numbers](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/a100_numbers.png)
 
-In the following tables, we report our findings in terms of the *number of iterations/second*.
+_(Our benchmarking metric for the plots above is **number of iterations/second**)_
+
+But we reveal all the benchmarking numbers in the interest of transparency! 
+
+In the following tables, we report our findings in terms of the number of **_iterations processed per second_**. 
 
 ### A100 (batch size: 1)
 
@@ -276,7 +295,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 22.24 | 23.23 | 43.76 | 49.25 |
 | SD - controlnet | 15.02 | 15.82 | 32.13 | 36.08 |
 | IF | 20.21 / <br>13.84 / <br>24.00 | 20.12 / <br>13.70 / <br>24.03 | ❌ | 97.34 / <br>27.23 / <br>111.66 |
-| SDXL - txt2img | 8.64 | 9.9 | - | - |
 
 ### A100 (batch size: 4)
 
@@ -287,7 +305,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 11.67 | 13.31 | 14.88 | 17.48 |
 | SD - controlnet | 8.28 | 9.38 | 10.51 | 12.41 |
 | IF | 25.02 | 18.04 | ❌ | 48.47 |
-| SDXL - txt2img | 2.44 | 2.74 | - | - |
 
 ### A100 (batch size: 16)
 
@@ -298,7 +315,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 3.04 | 3.66 | 3.9 | 4.76 |
 | SD - controlnet | 2.15 | 2.58 | 2.74 | 3.35 |
 | IF | 8.78 | 9.82 | ❌ | 16.77 |
-| SDXL - txt2img | 0.64 | 0.72 | - | - |
 
 ### V100 (batch size: 1)
 
@@ -339,7 +355,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 6.91 | 6.7 | 7.01 | 7.37 |
 | SD - controlnet | 4.89 | 4.86 | 5.35 | 5.48 |
 | IF | 17.42 / <br>2.47 / <br>18.52 | 16.96 / <br>2.45 / <br>18.69 | ❌ | 24.63 / <br>2.47 / <br>23.39 |
-| SDXL - txt2img | 1.15 | 1.16 | - | - |
 
 ### T4 (batch size: 4)
 
@@ -350,7 +365,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 1.81 | 1.82 | 2.09 | 2.09 |
 | SD - controlnet | 1.34 | 1.27 | 1.47 | 1.46 |
 | IF | 5.79 |  5.61 | ❌ | 7.39 |
-| SDXL - txt2img | 0.288 | 0.289 | - | - |
 
 ### T4 (batch size: 16)
 
@@ -361,7 +375,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 2.30s | 2.26s | OOM after 2nd iteration | 1.95s |
 | SD - controlnet | OOM after 2nd iteration | OOM after 2nd iteration | OOM after warmup | OOM after warmup |
 | IF * | 1.44 | 1.44 | ❌ | 1.94 |
-| SDXL - txt2img | OOM | OOM | - | - |
 
 ### RTX 3090 (batch size: 1)
 
@@ -402,7 +415,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 40.51 | 41.88 | 44.58 | 49.72 |
 | SD - controlnet | 29.27 | 30.29 | 32.26 | 36.03 |
 | IF | 69.71 / <br>18.78 / <br>85.49 | 69.13 / <br>18.80 / <br>85.56 | ❌ | 124.60 / <br>26.37 / <br>138.79 |
-| SDXL - txt2img | 6.8 | 8.18 | - | - |
 
 ### RTX 4090 (batch size: 4)
 
@@ -413,7 +425,6 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 12.65 | 12.81 | 15.3 | 15.58 |
 | SD - controlnet | 9.1 | 9.25 | 11.03 | 11.22 |
 | IF | 31.88 | 31.14 | ❌ | 43.92 |
-| SDXL - txt2img | 2.19 | 2.35 | - | - |
 
 ### RTX 4090 (batch size: 16)
 
@@ -424,11 +435,10 @@ In the following tables, we report our findings in terms of the *number of itera
 | SD - inpaint | 3.17 | 3.2 | 3.85 | 3.85 |
 | SD - controlnet | 2.23 | 2.3 | 2.7 | 2.75 |
 | IF | 9.26 | 9.2 | ❌ | 13.31 |
-| SDXL - txt2img | 0.52 | 0.53 | - | - |
 
 ## Notes 
 
-* Follow this [PR](https://github.com/huggingface/diffusers/pull/3313) for more details on the environment used for conducting the benchmarks. 
-* For the DeepFloyd IF pipeline where batch sizes > 1, we only used a batch size of > 1 in the first IF pipeline for text-to-image generation and NOT for upscaling. That means the two upscaling pipelines received a batch size of 1.
+* Follow [this PR](https://github.com/huggingface/diffusers/pull/3313) for more details on the environment used for conducting the benchmarks. 
+* For the IF pipeline and batch sizes > 1, we only used a batch size of >1 in the first IF pipeline for text-to-image generation and NOT for upscaling. So, that means the two upscaling pipelines received a batch size of 1. 
 
-*Thanks to [Horace He](https://github.com/Chillee) from the PyTorch team for their support in improving our support of `torch.compile()` in Diffusers.*
+*Thanks to [Horace He](https://github.com/Chillee) from the PyTorch team for their support in improving our support of `torch.compile()` in Diffusers.*

docs/source/en/optimization/xformers.md

@@ -10,11 +10,11 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# xFormers
+# Installing xFormers
 
-We recommend [xFormers](https://github.com/facebookresearch/xformers) for both inference and training. In our tests, the optimizations performed in the attention blocks allow for both faster speed and reduced memory consumption.
+We recommend the use of [xFormers](https://github.com/facebookresearch/xformers) for both inference and training. In our tests, the optimizations performed in the attention blocks allow for both faster speed and reduced memory consumption.
 
-Install xFormers from `pip`:
+Starting from version `0.0.16` of xFormers, released on January 2023, installation can be easily performed using pre-built pip wheels:
 
 ```bash
 pip install xformers
@@ -22,14 +22,14 @@ pip install xformers
 
 <Tip>
 
-The xFormers `pip` package requires the latest version of PyTorch. If you need to use a previous version of PyTorch, then we recommend [installing xFormers from the source](https://github.com/facebookresearch/xformers#installing-xformers).
+The xFormers PIP package requires the latest version of PyTorch (1.13.1 as of xFormers 0.0.16). If you need to use a previous version of PyTorch, then we recommend you install xFormers from source using [the project instructions](https://github.com/facebookresearch/xformers#installing-xformers).
 
 </Tip>
 
-After xFormers is installed, you can use `enable_xformers_memory_efficient_attention()` for faster inference and reduced memory consumption as shown in this [section](memory#memory-efficient-attention).
+After xFormers is installed, you can use `enable_xformers_memory_efficient_attention()` for faster inference and reduced memory consumption, as discussed [here](fp16#memory-efficient-attention).
 
 <Tip warning={true}>
 
-According to this [issue](https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212), xFormers `v0.0.16` cannot be used for training (fine-tune or DreamBooth) in some GPUs. If you observe this problem, please install a development version as indicated in the issue comments.
+According to [this issue](https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212), xFormers `v0.0.16` cannot be used for training (fine-tune or Dreambooth) in some GPUs. If you observe that problem, please install a development version as indicated in that comment.
 
 </Tip>

docs/source/en/training/ddpo.md

@@ -1,17 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-
-# Reinforcement learning training with DDPO
-
-You can fine-tune Stable Diffusion on a reward function via reinforcement learning with the 🤗 TRL library and 🤗 Diffusers. This is done with the Denoising Diffusion Policy Optimization (DDPO) algorithm introduced by Black et al. in [Training Diffusion Models with Reinforcement Learning](https://arxiv.org/abs/2305.13301), which is implemented in 🤗 TRL with the [`~trl.DDPOTrainer`].
-
-For more information, check out the [`~trl.DDPOTrainer`] API reference and the [Finetune Stable Diffusion Models with DDPO via TRL](https://huggingface.co/blog/trl-ddpo) blog post.

docs/source/en/training/lora.md

@@ -34,7 +34,7 @@ the attention layers of a language model is sufficient to obtain good downstream
 
 [cloneofsimo](https://github.com/cloneofsimo) was the first to try out LoRA training for Stable Diffusion in the popular [lora](https://github.com/cloneofsimo/lora) GitHub repository. 🧨 Diffusers now supports finetuning with LoRA for [text-to-image generation](https://github.com/huggingface/diffusers/tree/main/examples/text_to_image#training-with-lora) and [DreamBooth](https://github.com/huggingface/diffusers/tree/main/examples/dreambooth#training-with-low-rank-adaptation-of-large-language-models-lora). This guide will show you how to do both.
 
-If you'd like to store or share your model with the community, login to your Hugging Face account (create [one](https://hf.co/join) if you don't have one already):
+If you'd like to store or share your model with the community, login to your Hugging Face account (create [one](hf.co/join) if you don't have one already):
 
 ```bash
 huggingface-cli login
@@ -321,7 +321,7 @@ pipe.fuse_lora()
 
 generator = torch.manual_seed(0)
 images_fusion = pipe(
-    "masterpiece, best quality, mountain", generator=generator, num_inference_steps=2
+    "masterpiece, best quality, mountain", output_type="np", generator=generator, num_inference_steps=2
 ).images
 
 # To work with a different `lora_scale`, first reverse the effects of `fuse_lora()`.
@@ -333,101 +333,10 @@ pipe.fuse_lora(lora_scale=0.5)
 
 generator = torch.manual_seed(0)
 images_fusion = pipe(
-    "masterpiece, best quality, mountain", generator=generator, num_inference_steps=2
+    "masterpiece, best quality, mountain", output_type="np", generator=generator, num_inference_steps=2
 ).images
 ```
 
-## Serializing pipelines with fused LoRA parameters
-
-Let's say you want to load the pipeline above that has its UNet fused with the LoRA parameters. You can easily do so by simply calling the `save_pretrained()` method on `pipe`. 
-
-After loading the LoRA parameters into a pipeline, if you want to serialize the pipeline such that the affected model components are already fused with the LoRA parameters, you should:
-
-* call `fuse_lora()` on the pipeline with the desired `lora_scale`, given you've already loaded the LoRA parameters into it.
-* call `save_pretrained()` on the pipeline. 
-
-Here is a complete example:
-
-```python
-from diffusers import DiffusionPipeline
-import torch 
-
-pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
-lora_model_id = "hf-internal-testing/sdxl-1.0-lora"
-lora_filename = "sd_xl_offset_example-lora_1.0.safetensors"
-pipe.load_lora_weights(lora_model_id, weight_name=lora_filename)
-
-# First, fuse the LoRA parameters.
-pipe.fuse_lora()
-
-# Then save.
-pipe.save_pretrained("my-pipeline-with-fused-lora")
-```
-
-Now, you can load the pipeline and directly perform inference without having to load the LoRA parameters again:
-
-```python
-from diffusers import DiffusionPipeline
-import torch 
-
-pipe = DiffusionPipeline.from_pretrained("my-pipeline-with-fused-lora", torch_dtype=torch.float16).to("cuda")
-
-generator = torch.manual_seed(0)
-images_fusion = pipe(
-    "masterpiece, best quality, mountain", generator=generator, num_inference_steps=2
-).images
-```
-
-## Working with multiple LoRA checkpoints
-
-With the `fuse_lora()` method as described above, it's possible to load multiple LoRA checkpoints. Let's work through a complete example. First we load the base pipeline:
-
-```python
-from diffusers import StableDiffusionXLPipeline, AutoencoderKL
-import torch
-
-vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
-pipe = StableDiffusionXLPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-base-1.0",
-    vae=vae,
-    torch_dtype=torch.float16,
-)
-pipe.to("cuda")
-```
-
-Then let's two LoRA checkpoints and fuse them with specific `lora_scale` values:
-
-```python
-# LoRA one.
-pipe.load_lora_weights("goofyai/cyborg_style_xl")
-pipe.fuse_lora(lora_scale=0.7)
-
-# LoRA two.
-pipe.load_lora_weights("TheLastBen/Pikachu_SDXL")
-pipe.fuse_lora(lora_scale=0.7)
-```
-
-<Tip>
-
-Play with the `lora_scale` parameter when working with multiple LoRAs to control the amount of their influence on the final outputs.
-
-</Tip>
-
-Let's see them in action:
-
-```python
-prompt = "cyborg style pikachu"
-image = pipe(prompt, num_inference_steps=30, guidance_scale=7.5).images[0]
-```
-
-![cyborg_pikachu](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/cyborg_pikachu.png)
-
-<Tip warning={true}>
-
-Currently, unfusing multiple LoRA checkpoints is not possible. 
-
-</Tip>
-
 ## Supporting different LoRA checkpoints from Diffusers
 
 🤗 Diffusers supports loading checkpoints from popular LoRA trainers such as [Kohya](https://github.com/kohya-ss/sd-scripts/) and [TheLastBen](https://github.com/TheLastBen/fast-stable-diffusion). In this section, we outline the current API's details and limitations. 

docs/source/en/training/text2image.md

@@ -281,8 +281,3 @@ image.save("yoda-pokemon.png")
 
 * We support fine-tuning the UNet shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) via the `train_text_to_image_sdxl.py` script. Please refer to the docs [here](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/README_sdxl.md). 
 * We also support fine-tuning of the UNet and Text Encoder shipped in [Stable Diffusion XL](https://huggingface.co/papers/2307.01952) with LoRA via the `train_text_to_image_lora_sdxl.py` script. Please refer to the docs [here](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/README_sdxl.md). 
-
-
-## Kandinsky 2.2
-
-* We support fine-tuning both the decoder and prior in Kandinsky2.2 with the `train_text_to_image_prior.py` and `train_text_to_image_decoder.py` scripts. LoRA support is also included. Please refer to the docs [here](https://github.com/huggingface/diffusers/blob/main/examples/kandinsky2_2/text_to_image/README_sdxl.md).

docs/source/en/using-diffusers/conditional_image_generation.md

@@ -10,297 +10,51 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Text-to-image
+# Conditional image generation
 
 [[open-in-colab]]
 
-When you think of diffusion models, text-to-image is usually one of the first things that come to mind. Text-to-image generates an image from a text description (for example, "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k") which is also known as a *prompt*.
+Conditional image generation allows you to generate images from a text prompt. The text is converted into embeddings which are used to condition the model to generate an image from noise.
 
-From a very high level, a diffusion model takes a prompt and some random initial noise, and iteratively removes the noise to construct an image. The *denoising* process is guided by the prompt, and once the denoising process ends after a predetermined number of time steps, the image representation is decoded into an image.
+The [`DiffusionPipeline`] is the easiest way to use a pre-trained diffusion system for inference.
 
-<Tip>
+Start by creating an instance of [`DiffusionPipeline`] and specify which pipeline [checkpoint](https://huggingface.co/models?library=diffusers&sort=downloads) you would like to download.
 
-Read the [How does Stable Diffusion work?](https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work) blog post to learn more about how a latent diffusion model works.
+In this guide, you'll use [`DiffusionPipeline`] for text-to-image generation with [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5):
 
-</Tip>
+```python
+>>> from diffusers import DiffusionPipeline
 
-You can generate images from a prompt in 🤗 Diffusers in two steps:
-
-1. Load a checkpoint into the [`AutoPipelineForText2Image`] class, which automatically detects the appropriate pipeline class to use based on the checkpoint:
-
-```py
-from diffusers import AutoPipelineForText2Image
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
+>>> generator = DiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", use_safetensors=True)
 ```
 
-2. Pass a prompt to the pipeline to generate an image:
+The [`DiffusionPipeline`] downloads and caches all modeling, tokenization, and scheduling components. 
+Because the model consists of roughly 1.4 billion parameters, we strongly recommend running it on a GPU.
+You can move the generator object to a GPU, just like you would in PyTorch:
 
-```py
-image = pipeline(
-	"stained glass of darth vader, backlight, centered composition, masterpiece, photorealistic, 8k"
-).images[0]
+```python
+>>> generator.to("cuda")
 ```
 
-<div class="flex justify-center">
-	<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-vader.png"/>
-</div>
+Now you can use the `generator` on your text prompt:
 
-## Popular models
-
-The most common text-to-image models are [Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5), [Stable Diffusion XL (SDXL)](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0), and [Kandinsky 2.2](https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder). There are also ControlNet models or adapters that can be used with text-to-image models for more direct control in generating images. The results from each model are slightly different because of their architecture and training process, but no matter which model you choose, their usage is more or less the same. Let's use the same prompt for each model and compare their results.
-
-### Stable Diffusion v1.5
-
-[Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5) is a latent diffusion model initialized from [Stable Diffusion v1-4](https://huggingface.co/CompVis/stable-diffusion-v1-4), and finetuned for 595K steps on 512x512 images from the LAION-Aesthetics V2 dataset. You can use this model like:
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-generator = torch.Generator("cuda").manual_seed(31)
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", generator=generator).images[0]
+```python
+>>> image = generator("An image of a squirrel in Picasso style").images[0]
 ```
 
-### Stable Diffusion XL
+The output is by default wrapped into a [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) object.
 
-SDXL is a much larger version of the previous Stable Diffusion models, and involves a two-stage model process that adds even more details to an image. It also includes some additional *micro-conditionings* to generate high-quality images centered subjects. Take a look at the more comprehensive [SDXL](sdxl) guide to learn more about how to use it. In general, you can use SDXL like:
+You can save the image by calling:
 
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-generator = torch.Generator("cuda").manual_seed(31)
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", generator=generator).images[0]
+```python
+>>> image.save("image_of_squirrel_painting.png")
 ```
 
-### Kandinsky 2.2
+Try out the Spaces below, and feel free to play around with the guidance scale parameter to see how it affects the image quality!
 
-The Kandinsky model is a bit different from the Stable Diffusion models because it also uses an image prior model to create embeddings that are used to better align text and images in the diffusion model.
-
-The easiest way to use Kandinsky 2.2 is:
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-generator = torch.Generator("cuda").manual_seed(31)
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", generator=generator).images[0]
-```
-
-### ControlNet
-
-ControlNet are auxiliary models or adapters that are finetuned on top of text-to-image models, such as [Stable Diffusion V1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5). Using ControlNet models in combination with text-to-image models offers diverse options for more explicit control over how to generate an image. With ControlNet's, you add an additional conditioning input image to the model. For example, if you provide an image of a human pose (usually represented as multiple keypoints that are connected into a skeleton) as a conditioning input, the model generates an image that follows the pose of the image. Check out the more in-depth [ControlNet](controlnet) guide to learn more about other conditioning inputs and how to use them.
-
-In this example, let's condition the ControlNet with a human pose estimation image. Load the ControlNet model pretrained on human pose estimations:
-
-```py
-from diffusers import ControlNetModel, AutoPipelineForText2Image
-from diffusers.utils import load_image
-import torch
-
-controlnet = ControlNetModel.from_pretrained(
-	"lllyasviel/control_v11p_sd15_openpose", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pose_image = load_image("https://huggingface.co/lllyasviel/control_v11p_sd15_openpose/resolve/main/images/control.png")
-```
-
-Pass the `controlnet` to the [`AutoPipelineForText2Image`], and provide the prompt and pose estimation image:
-
-```py
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-generator = torch.Generator("cuda").manual_seed(31)
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", image=pose_image, generator=generator).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-1.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion v1.5</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-text2img.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion XL</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-2.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Kandinsky 2.2</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-3.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">ControlNet (pose conditioning)</figcaption>
-  </div>
-</div>
-
-## Configure pipeline parameters
-
-There are a number of parameters that can be configured in the pipeline that affect how an image is generated. You can change the image's output size, specify a negative prompt to improve image quality, and more. This section dives deeper into how to use these parameters.
-
-### Height and width
-
-The `height` and `width` parameters control the height and width (in pixels) of the generated image. By default, the Stable Diffusion v1.5 model outputs 512x512 images, but you can change this to any size that is a multiple of 8. For example, to create a rectangular image:
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-image = pipeline(
-	"Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", height=768, width=512
-).images[0]
-```
-
-<div class="flex justify-center">
-	<img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-hw.png"/>
-</div>
-
-<Tip warning={true}>
-
-Other models may have different default image sizes depending on the image size's in the training dataset. For example, SDXL's default image size is 1024x1024 and using lower `height` and `width` values may result in lower quality images. Make sure you check the model's API reference first!
-
-</Tip>
-
-### Guidance scale
-
-The `guidance_scale` parameter affects how much the prompt influences image generation. A lower value gives the model "creativity" to generate images that are more loosely related to the prompt. Higher `guidance_scale` values push the model to follow the prompt more closely, and if this value is too high, you may observe some artifacts in the generated image.
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
-).to("cuda")
-image = pipeline(
-	"Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", guidance_scale=3.5
-).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-guidance-scale-2.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 2.5</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-guidance-scale-7.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 7.5</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-guidance-scale-10.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 10.5</figcaption>
-  </div>
-</div>
-
-### Negative prompt
-
-Just like how a prompt guides generation, a *negative prompt* steers the model away from things you don't want the model to generate. This is commonly used to improve overall image quality by removing poor or bad image features such as "low resolution" or "bad details". You can also use a negative prompt to remove or modify the content and style of an image.
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
-).to("cuda")
-image = pipeline(
-	prompt="Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", 
-	negative_prompt="ugly, deformed, disfigured, poor details, bad anatomy",
-).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-neg-prompt-1.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">negative prompt = "ugly, deformed, disfigured, poor details, bad anatomy"</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/text2img-neg-prompt-2.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">negative prompt = "astronaut"</figcaption>
-  </div>
-</div>
-
-### Generator
-
-A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html#generator) object enables reproducibility in a pipeline by setting a manual seed. You can use a `Generator` to generate batches of images and iteratively improve on an image generated from a seed as detailed in the [Improve image quality with deterministic generation](reusing_seeds) guide.
-
-You can set a seed and `Generator` as shown below. Creating an image with a `Generator` should return the same result each time instead of randomly generating a new image.
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
-).to("cuda")
-generator = torch.Generator(device="cuda").manual_seed(30)
-image = pipeline(
-	"Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", 
-	generator=generator,
-).images[0]
-```
-
-## Control image generation
-
-There are several ways to exert more control over how an image is generated outside of configuring a pipeline's parameters, such as prompt weighting and ControlNet models.
-
-### Prompt weighting
-
-Prompt weighting is a technique for increasing or decreasing the importance of concepts in a prompt to emphasize or minimize certain features in an image. We recommend using the [Compel](https://github.com/damian0815/compel) library to help you generate the weighted prompt embeddings.
-
-<Tip>
-
-Learn how to create the prompt embeddings in the [Prompt weighting](weighted_prompts) guide. This example focuses on how to use the prompt embeddings in the pipeline.
-
-</Tip>
-
-Once you've created the embeddings, you can pass them to the `prompt_embeds` (and `negative_prompt_embeds` if you're using a negative prompt) parameter in the pipeline.
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-	"runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16
-).to("cuda")
-image = pipeline(
-	prompt_emebds=prompt_embeds, # generated from Compel
-	negative_prompt_embeds=negative_prompt_embeds, # generated from Compel
-).images[0]
-```
-
-### ControlNet
-
-As you saw in the [ControlNet](#controlnet) section, these models offer a more flexible and accurate way to generate images by incorporating an additional conditioning image input. Each ControlNet model is pretrained on a particular type of conditioning image to generate new images that resemble it. For example, if you take a ControlNet pretrained on depth maps, you can give the model a depth map as a conditioning input and it'll generate an image that preserves the spatial information in it. This is quicker and easier than specifying the depth information in a prompt. You can even combine multiple conditioning inputs with a [MultiControlNet](controlnet#multicontrolnet)!
-
-There are many types of conditioning inputs you can use, and 🤗 Diffusers supports ControlNet for Stable Diffusion and SDXL models. Take a look at the more comprehensive [ControlNet](controlnet) guide to learn how you can use these models.
-
-## Optimize
-
-Diffusion models are large, and the iterative nature of denoising an image is computationally expensive and intensive. But this doesn't mean you need access to powerful - or even many - GPUs to use them. There are many optimization techniques for running diffusion models on consumer and free-tier resources. For example, you can load model weights in half-precision to save GPU memory and increase speed or offload the entire model to the GPU to save even more memory.
-
-PyTorch 2.0 also supports a more memory-efficient attention mechanism called [*scaled dot product attention*](../optimization/torch2.0#scaled-dot-product-attention) that is automatically enabled if you're using PyTorch 2.0. You can combine this with [`torch.compile`](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) to speed your code up even more:
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained("runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16").to("cuda")
-pipeline.unet = torch.compile(pipeline.unet, mode="reduce-overheard", fullgraph=True)
-```
-
-For more tips on how to optimize your code to save memory and speed up inference, read the [Memory and speed](../optimization/fp16) and [Torch 2.0](../optimization/torch2.0) guides.
+<iframe
+	src="https://stabilityai-stable-diffusion.hf.space"
+	frameborder="0"
+	width="850"
+	height="500"
+></iframe>

docs/source/en/using-diffusers/freeu.md

@@ -1,123 +0,0 @@
-# Improve generation quality with FreeU
-
-[[open-in-colab]]
-
-The UNet is responsible for denoising during the reverse diffusion process, and there are two distinct features in its architecture: 
-
-1. Backbone features primarily contribute to the denoising process
-2. Skip features mainly introduce high-frequency features into the decoder module and can make the network overlook the semantics in the backbone features
-
-However, the skip connection can sometimes introduce unnatural image details. [FreeU](https://hf.co/papers/2309.11497) is a technique for improving image quality by rebalancing the contributions from the UNet’s skip connections and backbone feature maps. 
-
-FreeU is applied during inference and it does not require any additional training. The technique works for different tasks such as text-to-image, image-to-image, and text-to-video.
-
-In this guide, you will apply FreeU to the [`StableDiffusionPipeline`], [`StableDiffusionXLPipeline`], and [`TextToVideoSDPipeline`].
-
-## StableDiffusionPipeline
-
-Load the pipeline: 
-
-```py
-from diffusers import DiffusionPipeline
-import torch 
-
-pipeline = DiffusionPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, safety_checker=None
-).to("cuda")
-```
-
-Then enable the FreeU mechanism with the FreeU-specific hyperparameters. These values are scaling factors for the backbone and skip features.
-
-```py
-pipeline.enable_freeu(s1=0.9, s2=0.2, b1=1.2, b2=1.4)
-```
-
-The values above are from the official FreeU [code repository](https://github.com/ChenyangSi/FreeU) where you can also find [reference hyperparameters](https://github.com/ChenyangSi/FreeU#range-for-more-parameters) for different models.
-
-<Tip>
-
-Disable the FreeU mechanism by calling `disable_freeu()` on a pipeline.
-
-</Tip>
-
-And then run inference:
-
-```py
-prompt = "A squirrel eating a burger"
-seed = 2023
-image = pipeline(prompt, generator=torch.manual_seed(seed)).images[0]
-```
-
-The figure below compares non-FreeU and FreeU results respectively for the same hyperparameters used above (`prompt` and `seed`):
-
-![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/freeu/sdv1_5_freeu.jpg)
-
-
-Let's see how Stable Diffusion 2 results are impacted:
-
-```py
-from diffusers import DiffusionPipeline
-import torch 
-
-pipeline = DiffusionPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16, safety_checker=None
-).to("cuda")
-
-prompt = "A squirrel eating a burger"
-seed = 2023
-
-pipeline.enable_freeu(s1=0.9, s2=0.2, b1=1.1, b2=1.2)
-image = pipeline(prompt, generator=torch.manual_seed(seed)).images[0]
-```
-
-
-![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/freeu/sdv2_1_freeu.jpg)
-
-## Stable Diffusion XL
-
-Finally, let's take a look at how FreeU affects Stable Diffusion XL results:
-
-```py
-from diffusers import DiffusionPipeline
-import torch 
-
-pipeline = DiffusionPipeline.from_pretrained(
-    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16,
-).to("cuda")
-
-prompt = "A squirrel eating a burger"
-seed = 2023
-
-# Comes from 
-# https://wandb.ai/nasirk24/UNET-FreeU-SDXL/reports/FreeU-SDXL-Optimal-Parameters--Vmlldzo1NDg4NTUw
-pipeline.enable_freeu(s1=0.6, s2=0.4, b1=1.1, b2=1.2)
-image = pipeline(prompt, generator=torch.manual_seed(seed)).images[0]
-```
-
-
-![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/freeu/sdxl_freeu.jpg)
-
-## Text-to-video generation
-
-FreeU can also be used to improve video quality:
-
-```python
-from diffusers import DiffusionPipeline
-from diffusers.utils import export_to_video
-import torch
-
-model_id = "cerspense/zeroscope_v2_576w"
-pipe = DiffusionPipeline.from_pretrained("cerspense/zeroscope_v2_576w", torch_dtype=torch.float16).to("cuda")
-pipe = pipe.to("cuda")
-
-prompt = "an astronaut riding a horse on mars"
-seed = 2023
-
-# The values come from
-# https://github.com/lyn-rgb/FreeU_Diffusers#video-pipelines
-pipe.enable_freeu(b1=1.2, b2=1.4, s1=0.9, s2=0.2)
-video_frames = pipe(prompt, height=320, width=576, num_frames=30, generator=torch.manual_seed(seed)).frames
-export_to_video(video_frames, "astronaut_rides_horse.mp4")
-```
-
-Thanks to [kadirnar](https://github.com/kadirnar/) for helping to integrate the feature, and to [justindujardin](https://github.com/justindujardin) for the helpful discussions.

docs/source/en/using-diffusers/img2img.md

@@ -10,597 +10,91 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Image-to-image
+# Text-guided image-to-image generation
 
 [[open-in-colab]]
 
-Image-to-image is similar to [text-to-image](conditional_image_generation), but in addition to a prompt, you can also pass an initial image as a starting point for the diffusion process. The initial image is encoded to latent space and noise is added to it. Then the latent diffusion model takes a prompt and the noisy latent image, predicts the added noise, and removes the predicted noise from the initial latent image to get the new latent image. Lastly, a decoder decodes the new latent image back into an image.
+The [`StableDiffusionImg2ImgPipeline`] lets you pass a text prompt and an initial image to condition the generation of new images.
 
-With 🤗 Diffusers, this is as easy as 1-2-3:
-
-1. Load a checkpoint into the [`AutoPipelineForImage2Image`] class; this pipeline automatically handles loading the correct pipeline class  based on the checkpoint:
+Before you begin, make sure you have all the necessary libraries installed:
 
 ```py
-from diffusers import AutoPipelineForImage2Image
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
+# uncomment to install the necessary libraries in Colab
+#!pip install diffusers transformers ftfy accelerate
 ```
 
+Get started by creating a [`StableDiffusionImg2ImgPipeline`] with a pretrained Stable Diffusion model like [`nitrosocke/Ghibli-Diffusion`](https://huggingface.co/nitrosocke/Ghibli-Diffusion).
+
+```python
+import torch
+import requests
+from PIL import Image
+from io import BytesIO
+from diffusers import StableDiffusionImg2ImgPipeline
+
+device = "cuda"
+pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
+    "nitrosocke/Ghibli-Diffusion", torch_dtype=torch.float16, use_safetensors=True
+).to(device)
+```
+
+Download and preprocess an initial image so you can pass it to the pipeline:
+
+```python
+url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+
+response = requests.get(url)
+init_image = Image.open(BytesIO(response.content)).convert("RGB")
+init_image.thumbnail((768, 768))
+init_image
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/image_2_image_using_diffusers_cell_8_output_0.jpeg"/>
+</div>
+
 <Tip>
 
-You'll notice throughout the guide, we use [`~DiffusionPipeline.enable_model_cpu_offload`] and [`~DiffusionPipeline.enable_xformers_memory_efficient_attention`], to save memory and increase inference speed. If you're using PyTorch 2.0, then you don't need to call [`~DiffusionPipeline.enable_xformers_memory_efficient_attention`] on your pipeline because it'll already be using PyTorch 2.0's native [scaled-dot product attention](../optimization/torch2.0#scaled-dot-product-attention).
+💡 `strength` is a value between 0.0 and 1.0 that controls the amount of noise added to the input image. Values that approach 1.0 allow for lots of variations but will also produce images that are not semantically consistent with the input.
 
 </Tip>
 
-2. Load an image to pass to the pipeline:
+Define the prompt (for this checkpoint finetuned on Ghibli-style art, you need to prefix the prompt with the `ghibli style` tokens) and run the pipeline:
 
-```py
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png")
-```
-
-3. Pass a prompt and image to the pipeline to generate an image:
-
-```py
-prompt = "cat wizard, gandalf, lord of the rings, detailed, fantasy, cute, adorable, Pixar, Disney, 8k"
-image = pipeline(prompt, image=init_image).images[0]
-image
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
-  </div>
-</div>
-
-## Popular models
-
-The most popular image-to-image models are [Stable Diffusion v1.5](https://huggingface.co/runwayml/stable-diffusion-v1-5), [Stable Diffusion XL (SDXL)](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0), and [Kandinsky 2.2](https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder). The results from the Stable Diffusion and Kandinsky models vary due to their architecture differences and training process; you can generally expect SDXL to produce higher quality images than Stable Diffusion v1.5. Let's take a quick look at how to use each of these models and compare their results.
-
-### Stable Diffusion v1.5
-
-Stable Diffusion v1.5 is a latent diffusion model intialized from an earlier checkpoint, and further finetuned for 595K steps on 512x512 images. To use this pipeline for image-to-image, you'll need to prepare an initial image to pass to the pipeline. Then you can pass a prompt and the image to the pipeline to generate a new image:
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image).images[0]
-image
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-sdv1.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
-  </div>
-</div>
-
-### Stable Diffusion XL (SDXL)
-
-SDXL is a more powerful version of the Stable Diffusion model. It uses a larger base model, and an additional refiner model to increase the quality of the base model's output. Read the [SDXL](sdxl) guide for a more detailed walkthrough of how to use this model, and other techniques it uses to produce high quality images.
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-sdxl-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image, strength=0.5).images[0]
-image
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-sdxl-init.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-sdxl.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
-  </div>
-</div>
-
-### Kandinsky 2.2
-
-The Kandinsky model is different from the Stable Diffusion models because it uses an image prior model to create image embeddings. The embeddings help create a better alignment between text and images, allowing the latent diffusion model to generate better images.
-
-The simplest way to use Kandinsky 2.2 is:
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image).images[0]
-image
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-kandinsky.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
-  </div>
-</div>
-
-## Configure pipeline parameters
-
-There are several important parameters you can configure in the pipeline that'll affect the image generation process and image quality. Let's take a closer look at what these parameters do and how changing them affects the output.
-
-### Strength
-
-`strength` is one of the most important parameters to consider and it'll have a huge impact on your generated image. It determines how much the generated image resembles the initial image. In other words:
-
-- 📈 a higher `strength` value gives the model more "creativity" to generate an image that's different from the initial image; a `strength` value of 1.0 means the initial image is more or less ignored
-- 📉 a lower `strength` value means the generated image is more similar to the initial image
-
-The `strength` and `num_inference_steps` parameter are related because `strength` determines the number of noise steps to add. For example, if the `num_inference_steps` is 50 and `strength` is 0.8, then this means adding 40 (50 * 0.8) steps of noise to the initial image and then denoising for 40 steps to get the newly generated image.
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-image = init_image
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image, strength=0.8).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-strength-0.4.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.4</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-strength-0.6.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.6</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-strength-1.0.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 1.0</figcaption>
-  </div>
-</div>
-
-### Guidance scale
-
-The `guidance_scale` parameter is used to control how closely aligned the generated image and text prompt are. A higher `guidance_scale` value means your generated image is more aligned with the prompt, while a lower `guidance_scale` value means your generated image has more space to deviate from the prompt.
-
-You can combine `guidance_scale` with `strength` for even more precise control over how expressive the model is. For example, combine a high `strength + guidance_scale` for maximum creativity or use a combination of low `strength` and low `guidance_scale` to generate an image that resembles the initial image but is not as strictly bound to the prompt.
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image, guidance_scale=8.0).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-guidance-0.1.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 0.1</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-guidance-3.0.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 5.0</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-guidance-7.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 10.0</figcaption>
-  </div>
-</div>
-
-### Negative prompt
-
-A negative prompt conditions the model to *not* include things in an image, and it can be used to improve image quality or modify an image. For example, you can improve image quality by including negative prompts like "poor details" or "blurry" to encourage the model to generate a higher quality image. Or you can modify an image by specifying things to exclude from an image.
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-negative_prompt = "ugly, deformed, disfigured, poor details, bad anatomy"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, negative_prompt=negative_prompt, image=init_image).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-negative-1.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">negative prompt = "ugly, deformed, disfigured, poor details, bad anatomy"</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-negative-2.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">negative prompt = "jungle"</figcaption>
-  </div>
-</div>
-
-## Chained image-to-image pipelines
-
-There are some other interesting ways you can use an image-to-image pipeline aside from just generating an image (although that is pretty cool too). You can take it a step further and chain it with other pipelines.
-
-### Text-to-image-to-image
-
-Chaining a text-to-image and image-to-image pipeline allows you to generate an image from text and use the generated image as the initial image for the image-to-image pipeline. This is useful if you want to generate an image entirely from scratch. For example, let's chain a Stable Diffusion and a Kandinsky model.
-
-Start by generating an image with the text-to-image pipeline:
-
-```py
-from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k").images[0]
-```
-
-Now you can pass this generated image to the image-to-image pipeline:
-
-```py
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline("Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", image=image).images[0]
-image
-```
-
-### Image-to-image-to-image
-
-You can also chain multiple image-to-image pipelines together to create more interesting images. This can be useful for iteratively performing style transfer on an image, generate short GIFs, restore color to an image, or restore missing areas of an image.
-
-Start by generating an image:
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image = pipeline(prompt, image=init_image, output_type="latent").images[0]
-```
-
-<Tip>
-
-It is important to specify `output_type="latent"` in the pipeline to keep all the outputs in latent space to avoid an unnecessary decode-encode step. This only works if the chained pipelines are using the same VAE.
-
-</Tip>
-
-Pass the latent output from this pipeline to the next pipeline to generate an image in a [comic book art style](https://huggingface.co/ogkalu/Comic-Diffusion):
-
-```py
-pipelne = AutoPipelineForImage2Image.from_pretrained(
-    "ogkalu/Comic-Diffusion", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# need to include the token "charliebo artstyle" in the prompt to use this checkpoint
-image = pipeline("Astronaut in a jungle, charliebo artstyle", image=image, output_type="latent").images[0]
-```
-
-Repeat one more time to generate the final image in a [pixel art style](https://huggingface.co/kohbanye/pixel-art-style):
-
-```py
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "kohbanye/pixel-art-style", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# need to include the token "pixelartstyle" in the prompt to use this checkpoint
-image = pipeline("Astronaut in a jungle, pixelartstyle", image=image).images[0]
-image
-```
-
-### Image-to-upscaler-to-super-resolution
-
-Another way you can chain your image-to-image pipeline is with an upscaler and super-resolution pipeline to really increase the level of details in an image.
-
-Start with an image-to-image pipeline:
-
-```py
-import torch
-import requests
-from PIL import Image
-from io import BytesIO
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-
-# pass prompt and image to pipeline
-image_1 = pipeline(prompt, image=init_image, output_type="latent").images[0]
-```
-
-<Tip>
-
-It is important to specify `output_type="latent"` in the pipeline to keep all the outputs in *latent* space to avoid an unnecessary decode-encode step. This only works if the chained pipelines are using the same VAE.
-
-</Tip>
-
-Chain it to an upscaler pipeline to increase the image resolution:
-
-```py
-upscaler = AutoPipelineForImage2Image.from_pretrained(
-    "stabilityai/sd-x2-latent-upscaler", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-upscaler.enable_model_cpu_offload()
-upscaler.enable_xformers_memory_efficient_attention()
-
-image_2 = upscaler(prompt, image=image_1, output_type="latent").images[0]
-```
-
-Finally, chain it to a super-resolution pipeline to further enhance the resolution:
-
-```py
-super_res = AutoPipelineForImage2Image.from_pretrained(
-    "stabilityai/stable-diffusion-x4-upscaler", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-super_res.enable_model_cpu_offload()
-super_res.enable_xformers_memory_efficient_attention()
-
-image_3 = upscaler(prompt, image=image_2).images[0]
-image_3
-```
-
-## Control image generation
-
-Trying to generate an image that looks exactly the way you want can be difficult, which is why controlled generation techniques and models are so useful. While you can use the `negative_prompt` to partially control image generation, there are more robust methods like prompt weighting and ControlNets.
-
-### Prompt weighting
-
-Prompt weighting allows you to scale the representation of each concept in a prompt. For example, in a prompt like "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k", you can choose to increase or decrease the embeddings of "astronaut" and "jungle". The [Compel](https://github.com/damian0815/compel) library provides a simple syntax for adjusting prompt weights and generating the embeddings. You can learn how to create the embeddings in the [Prompt weighting](weighted_prompts) guide.
-
-[`AutoPipelineForImage2Image`] has a `prompt_embeds` (and `negative_prompt_embeds` if you're using a negative prompt) parameter where you can pass the embeddings which replaces the `prompt` parameter.
-
-```py
-from diffusers import AutoPipelineForImage2Image
-import torch
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline(prompt_emebds=prompt_embeds, # generated from Compel
-    negative_prompt_embeds, # generated from Compel
-    image=init_image,
-).images[0]
-```
-
-### ControlNet
-
-ControlNets provide a more flexible and accurate way to control image generation because you can use an additional conditioning image. The conditioning image can be a canny image, depth map, image segmentation, and even scribbles! Whatever type of conditioning image you choose, the ControlNet generates an image that preserves the information in it.
-
-For example, let's condition an image with a depth map to keep the spatial information in the image.
-
-```py
-# prepare image
-url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"
-response = requests.get(url)
-init_image = Image.open(BytesIO(response.content)).convert("RGB")
-init_image = init_image.resize((958, 960)) # resize to depth image dimensions
-depth_image = load_image("https://huggingface.co/lllyasviel/control_v11f1p_sd15_depth/resolve/main/images/control.png")
-```
-
-Load a ControlNet model conditioned on depth maps and the [`AutoPipelineForImage2Image`]:
-
-```py
-from diffusers import ControlNetModel, AutoPipelineForImage2Image
-from diffusers.utils import load_image
-import torch
-
-controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11f1p_sd15_depth", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-```
-
-Now generate a new image conditioned on the depth map, initial image, and prompt:
-
-```py
-prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
-image = pipeline(prompt, image=init_image, control_image=depth_image).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-init.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/lllyasviel/control_v11f1p_sd15_depth/resolve/main/images/control.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">depth image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-controlnet.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">ControlNet image</figcaption>
-  </div>
-</div>
-
-Let's apply a new [style](https://huggingface.co/nitrosocke/elden-ring-diffusion) to the image generated from the ControlNet by chaining it with an image-to-image pipeline:
-
-```py
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "nitrosocke/elden-ring-diffusion", torch_dtype=torch.float16,
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-prompt = "elden ring style astronaut in a jungle" # include the token "elden ring style" in the prompt
-negative_prompt = "ugly, deformed, disfigured, poor details, bad anatomy"
-
-image = pipeline(prompt, negative_prompt=negative_prompt, image=init_image, strength=0.45, guidance_scale=10.5).images[0]
+```python
+prompt = "ghibli style, a fantasy landscape with castles"
+generator = torch.Generator(device=device).manual_seed(1024)
+image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
 image
 ```
 
 <div class="flex justify-center">
-  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/img2img-elden-ring.png">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ghibli-castles.png"/>
 </div>
 
-## Optimize
+You can also try experimenting with a different scheduler to see how that affects the output:
 
-Running diffusion models is computationally expensive and intensive, but with a few optimization tricks, it is entirely possible to run them on consumer and free-tier GPUs. For example, you can use a more memory-efficient form of attention such as PyTorch 2.0's [scaled-dot product attention](../optimization/torch2.0#scaled-dot-product-attention) or [xFormers](../optimization/xformers) (you can use one or the other, but there's no need to use both). You can also offload the model to the GPU while the other pipeline components wait on the CPU.
+```python
+from diffusers import LMSDiscreteScheduler
 
-```diff
-+ pipeline.enable_model_cpu_offload()
-+ pipeline.enable_xformers_memory_efficient_attention()
+lms = LMSDiscreteScheduler.from_config(pipe.scheduler.config)
+pipe.scheduler = lms
+generator = torch.Generator(device=device).manual_seed(1024)
+image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
+image
 ```
 
-With [`torch.compile`](../optimization/torch2.0#torch.compile), you can boost your inference speed even more by wrapping your UNet with it:
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lms-ghibli.png"/>
+</div>
 
-```py
-pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
-```
+Check out the Spaces below, and try generating images with different values for `strength`. You'll notice that using lower values for `strength` produces images that are more similar to the original image.
 
-To learn more, take a look at the [Reduce memory usage](../optimization/memory) and [Torch 2.0](../optimization/torch2.0) guides.
+Feel free to also switch the scheduler to the [`LMSDiscreteScheduler`] and see how that affects the output.
+
+<iframe
+	src="https://stevhliu-ghibli-img2img.hf.space"
+	frameborder="0"
+	width="850"
+	height="500"
+></iframe>

docs/source/en/using-diffusers/inpaint.md

@@ -10,302 +10,87 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# Inpainting
+# Text-guided image-inpainting
 
 [[open-in-colab]]
 
-Inpainting replaces or edits specific areas of an image. This makes it a useful tool for image restoration like removing defects and artifacts, or even replacing an image area with something entirely new. Inpainting relies on a mask to determine which regions of an image to fill in; the area to inpaint is represented by white pixels and the area to keep is represented by black pixels. The white pixels are filled in by the prompt.
+The [`StableDiffusionInpaintPipeline`] allows you to edit specific parts of an image by providing a mask and a text prompt. It uses a version of Stable Diffusion, like [`runwayml/stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting) specifically trained for inpainting tasks.
 
-With 🤗 Diffusers, here is how you can do inpainting:
+Get started by loading an instance of the [`StableDiffusionInpaintPipeline`]:
 
-1. Load an inpainting checkpoint with the [`AutoPipelineForInpainting`] class. This'll automatically detect the appropriate pipeline class to load based on the checkpoint:
-
-```py
+```python
+import PIL
+import requests
 import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
+from io import BytesIO
 
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder-inpaint", torch_dtype=torch.float16
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
+from diffusers import StableDiffusionInpaintPipeline
+
+pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+    "runwayml/stable-diffusion-inpainting",
+    torch_dtype=torch.float16,
+    use_safetensors=True,
+    variant="fp16",
+)
+pipeline = pipeline.to("cuda")
 ```
 
-<Tip>
+Download an image and a mask of a dog which you'll eventually replace:
 
-You'll notice throughout the guide, we use [`~DiffusionPipeline.enable_model_cpu_offload`] and [`~DiffusionPipeline.enable_xformers_memory_efficient_attention`], to save memory and increase inference speed. If you're using PyTorch 2.0, it's not necessary to call [`~DiffusionPipeline.enable_xformers_memory_efficient_attention`] on your pipeline because it'll already be using PyTorch 2.0's native [scaled-dot product attention](../optimization/torch2.0#scaled-dot-product-attention).
+```python
+def download_image(url):
+    response = requests.get(url)
+    return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+
+img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
+mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"
+
+init_image = download_image(img_url).resize((512, 512))
+mask_image = download_image(mask_url).resize((512, 512))
+```
+
+Now you can create a prompt to replace the mask with something else:
+
+```python
+prompt = "Face of a yellow cat, high resolution, sitting on a park bench"
+image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
+```
+
+`image`          | `mask_image` | `prompt` | output |
+:-------------------------:|:-------------------------:|:-------------------------:|-------------------------:|
+<img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" alt="drawing" width="250"/> | <img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" alt="drawing" width="250"/> | ***Face of a yellow cat, high resolution, sitting on a park bench*** | <img src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint/yellow_cat_sitting_on_a_park_bench.png" alt="drawing" width="250"/> |
+
+
+<Tip warning={true}>
+
+A previous experimental implementation of inpainting used a different, lower-quality process. To ensure backwards compatibility, loading a pretrained pipeline that doesn't contain the new model will still apply the old inpainting method.
 
 </Tip>
 
-2. Load the base and mask images:
-
-```py
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-```
-
-3. Create a prompt to inpaint the image with and pass it to the pipeline with the base and mask images:
-
-```py
-prompt = "a black cat with glowing eyes, cute, adorable, disney, pixar, highly detailed, 8k"
-negative_prompt = "bad anatomy, deformed, ugly, disfigured"
-image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=init_image, mask_image=mask_image).images[0]
-```
-
-<div class="flex gap-4">
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">base image</figcaption>
-  </div>
-  <div>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-cat.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">generated image</figcaption>
-  </div>
-</div>
-
-## Create a mask image
-
-Throughout this guide, the mask image is provided in all of the code examples for convenience. You can inpaint on your own images, but you'll need to create a mask image for it. Use the Space below to easily create a mask image.
-
-Upload a base image to inpaint on and use the sketch tool to draw a mask. Once you're done, click **Run** to generate and download the mask image.
+Check out the Spaces below to try out image inpainting yourself!
 
 <iframe
-	src="https://stevhliu-inpaint-mask-maker.hf.space"
+	src="https://runwayml-stable-diffusion-inpainting.hf.space"
 	frameborder="0"
 	width="850"
-	height="450"
+	height="500"
 ></iframe>
 
-## Popular models
+## Preserving the Unmasked Area of the Image
 
-[Stable Diffusion Inpainting](https://huggingface.co/runwayml/stable-diffusion-inpainting), [Stable Diffusion XL (SDXL) Inpainting](https://huggingface.co/diffusers/stable-diffusion-xl-1.0-inpainting-0.1), and [Kandinsky 2.2](https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder-inpaint) are among the most popular models for inpainting. SDXL typically produces higher resolution images than Stable Diffusion v1.5, and Kandinsky 2.2 is also capable of generating high-quality images.
+Generally speaking, [`StableDiffusionInpaintPipeline`] (and other inpainting pipelines) will change the unmasked part of the image as well. If this behavior is undesirable, you can force the unmasked area to remain the same as follows:
 
-### Stable Diffusion Inpainting
-
-Stable Diffusion Inpainting is a latent diffusion model finetuned on 512x512 images on inpainting. It is a good starting point because it is relatively fast and generates good quality images. To use this model for inpainting, you'll need to pass a prompt, base and mask image to the pipeline:
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-generator = torch.Generator("cuda").manual_seed(92)
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[0]
-```
-
-### Stable Diffusion XL (SDXL) Inpainting
-
-SDXL is a larger and more powerful version of Stable Diffusion v1.5. This model can follow a two-stage model process (though each model can also be used alone); the base model generates an image, and a refiner model takes that image and further enhances its details and quality. Take a look at the [SDXL](sdxl) guide for a more comprehensive guide on how to use SDXL and configure it's parameters.
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "diffusers/stable-diffusion-xl-1.0-inpainting-0.1", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-generator = torch.Generator("cuda").manual_seed(92)
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[0]
-```
-
-### Kandinsky 2.2 Inpainting
-
-The Kandinsky model family is similar to SDXL because it uses two models as well; the image prior model creates image embeddings, and the diffusion model generates images from them. You can load the image prior and diffusion model separately, but the easiest way to use Kandinsky 2.2 is to load it into the [`AutoPipelineForInpainting`] class which uses the [`KandinskyV22InpaintCombinedPipeline`] under the hood.
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder-inpaint", torch_dtype=torch.float16
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-generator = torch.Generator("cuda").manual_seed(92)
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, generator=generator).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">base image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-sdv1.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion Inpainting</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-sdxl.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Stable Diffusion XL Inpainting</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-kandinsky.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">Kandinsky 2.2 Inpainting</figcaption>
-  </div>
-</div>
-
-## Configure pipeline parameters
-
-Image features - like quality and "creativity" - are dependent on pipeline parameters. Knowing what these parameters do is important for getting the results you want. Let's take a look at the most important parameters and see how changing them affects the output.
-
-### Strength
-
-`strength` is a measure of how much noise is added to the base image, which influences how similar the output is to the base image.
-
-* 📈 a high `strength` value means more noise is added to an image and the denoising process takes longer, but you'll get higher quality images that are more different from the base image
-* 📉 a low `strength` value means less noise is added to an image and the denoising process is faster, but the image quality may not be as great and the generated image resembles the base image more
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, strength=0.6).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-0.6.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.6</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-0.8.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 0.8</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-strength-1.0.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">strength = 1.0</figcaption>
-  </div>
-</div>
-
-### Guidance scale
-
-`guidance_scale` affects how aligned the text prompt and generated image are.
-
-* 📈 a high `guidance_scale` value means the prompt and generated image are closely aligned, so the output is a stricter interpretation of the prompt
-* 📉 a low `guidance_scale` value means the prompt and generated image are more loosely aligned, so the output may be more varied from the prompt
-
-You can use `strength` and `guidance_scale` together for more control over how expressive the model is. For example, a combination high `strength` and `guidance_scale` values gives the model the most creative freedom.
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, guidance_scale=2.5).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-2.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 2.5</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-7.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 7.5</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-guidance-12.5.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">guidance_scale = 12.5</figcaption>
-  </div>
-</div>
-
-### Negative prompt
-
-A negative prompt assumes the opposite role of a prompt; it guides the model away from generating certain things in an image. This is useful for quickly improving image quality and preventing the model from generating things you don't want.
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-negative_prompt = "bad architecture, unstable, poor details, blurry"
-image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=init_image, mask_image=mask_image).images[0]
-image
-```
-
-<div class="flex justify-center">
-  <figure>
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-negative.png" />
-    <figcaption class="text-center">negative_prompt = "bad architecture, unstable, poor details, blurry"</figcaption>
-  </figure>
-</div>
-
-## Preserve unmasked areas
-
-The [`AutoPipelineForInpainting`] (and other inpainting pipelines) generally changes the unmasked parts of an image to create a more natural transition between the masked and unmasked region. If this behavior is undesirable, you can force the unmasked area to remain the same. However, forcing the unmasked portion of the image to remain the same may result in some unusual transitions between the unmasked and masked areas.
-
-```py
+```python
 import PIL
 import numpy as np
 import torch
 
-from diffusers import AutoPipelineForInpainting
+from diffusers import StableDiffusionInpaintPipeline
 from diffusers.utils import load_image
 
 device = "cuda"
-pipeline = AutoPipelineForInpainting.from_pretrained(
+pipeline = StableDiffusionInpaintPipeline.from_pretrained(
     "runwayml/stable-diffusion-inpainting",
     torch_dtype=torch.float16,
 )
@@ -336,257 +121,4 @@ unmasked_unchanged_image = PIL.Image.fromarray(unmasked_unchanged_image_arr.roun
 unmasked_unchanged_image.save("force_unmasked_unchanged.png")
 ```
 
-## Chained inpainting pipelines
-
-[`AutoPipelineForInpainting`] can be chained with other 🤗 Diffusers pipelines to edit their outputs. This is often useful for improving the output quality from your other diffusion pipelines, and if you're using multiple pipelines, it can be more memory-efficient to chain them together to keep the outputs in latent space and reuse the same pipeline components.
-
-### Text-to-image-to-inpaint
-
-Chaining a text-to-image and inpainting pipeline allows you to inpaint the generated image, and you don't have to provide a base image to begin with. This makes it convenient to edit your favorite text-to-image outputs without having to generate an entirely new image.
-
-Start with the text-to-image pipeline to create a castle:
-
-```py
-import torch
-from diffusers import AutoPipelineForText2Image, AutoPipelineForInpainting
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForText2Image.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline("concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k").images[0]
-```
-
-Load the mask image of the output from above:
-
-```py
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_text-chain-mask.png").convert("RGB")
-```
-
-And let's inpaint the masked area with a waterfall:
-
-```py
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "kandinsky-community/kandinsky-2-2-decoder-inpaint", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-prompt = "digital painting of a fantasy waterfall, cloudy"
-image = pipeline(prompt=prompt, image=image, mask_image=mask_image).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-text-chain.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">text-to-image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-text-chain-out.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">inpaint</figcaption>
-  </div>
-</div>
-
-
-### Inpaint-to-image-to-image
-
-You can also chain an inpainting pipeline before another pipeline like image-to-image or an upscaler to improve the quality.
-
-Begin by inpainting an image:
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting, AutoPipelineForImage2Image
-from diffusers.utils import load_image
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
-
-# resize image to 1024x1024 for SDXL
-image = image.resize((1024, 1024))
-```
-
-Now let's pass the image to another inpainting pipeline with SDXL's refiner model to enhance the image details and quality:
-
-```py
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline(prompt=prompt, image=image, mask_image=mask_image, output_type="latent").images[0]
-```
-
-<Tip>
-
-It is important to specify `output_type="latent"` in the pipeline to keep all the outputs in latent space to avoid an unnecessary decode-encode step. This only works if the chained pipelines are using the same VAE. For example, in the [Text-to-image-to-inpaint](#text-to-image-to-inpaint) section, Kandinsky 2.2 uses a different VAE class than the Stable Diffusion model so it won't work. But if you use Stable Diffusion v1.5 for both pipelines, then you can keep everything in latent space because they both use [`AutoencoderKL`].
-
-</Tip>
-
-Finally, you can pass this image to an image-to-image pipeline to put the finishing touches on it. It is more efficient to use the [`~AutoPipelineForImage2Image.from_pipe`] method to reuse the existing pipeline components, and avoid unnecessarily loading all the pipeline components into memory again.
-
-```py
-pipeline = AutoPipelineForImage2Image.from_pipe(pipeline)
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline(prompt=prompt, image=image).images[0]
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-to-image-chain.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">inpaint</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-to-image-final.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">image-to-image</figcaption>
-  </div>
-</div>
-
-Image-to-image and inpainting are actually very similar tasks. Image-to-image generates a new image that resembles the existing provided image. Inpainting does the same thing, but it only transforms the image area defined by the mask and the rest of the image is unchanged. You can think of inpainting as a more precise tool for making specific changes and image-to-image has a broader scope for making more sweeping changes.
-
-## Control image generation
-
-Getting an image to look exactly the way you want is challenging because the denoising process is random. While you can control certain aspects of generation by configuring parameters like `negative_prompt`, there are better and more efficient methods for controlling image generation.
-
-### Prompt weighting
-
-Prompt weighting provides a quantifiable way to scale the representation of concepts in a prompt. You can use it to increase or decrease the magnitude of the text embedding vector for each concept in the prompt, which subsequently determines how much of each concept is generated. The [Compel](https://github.com/damian0815/compel) library offers an intuitive syntax for scaling the prompt weights and generating the embeddings. Learn how to create the embeddings in the [Prompt weighting](../using-diffusers/weighted_prompts) guide.
-
-Once you've generated the embeddings, pass them to the `prompt_embeds` (and `negative_prompt_embeds` if you're using a negative prompt) parameter in the [`AutoPipelineForInpainting`]. The embeddings replace the `prompt` parameter:
-
-```py
-import torch
-from diffusers import AutoPipelineForInpainting
-
-pipeline = AutoPipelineForInpainting.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float16,
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-image = pipeline(prompt_emebds=prompt_embeds, # generated from Compel
-    negative_prompt_embeds, # generated from Compel
-    image=init_image,
-    mask_image=mask_image
-).images[0]
-```
-
-### ControlNet
-
-ControlNet models are used with other diffusion models like Stable Diffusion, and they provide an even more flexible and accurate way to control how an image is generated. A ControlNet accepts an additional conditioning image input that guides the diffusion model to preserve the features in it.
-
-For example, let's condition an image with a ControlNet pretrained on inpaint images:
-
-```py
-import torch
-import numpy as np
-from diffusers import ControlNetModel, StableDiffusionControlNetInpaintPipeline
-from diffusers.utils import load_image
-
-# load ControlNet
-controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11p_sd15_inpaint", torch_dtype=torch.float16, variant="fp16")
-
-# pass ControlNet to the pipeline
-pipeline = StableDiffusionControlNetInpaintPipeline.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", controlnet=controlnet, torch_dtype=torch.float16, variant="fp16"
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-# load base and mask image
-init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png").convert("RGB")
-mask_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint_mask.png").convert("RGB")
-
-# prepare control image
-def make_inpaint_condition(init_image, mask_image):
-    init_image = np.array(init_image.convert("RGB")).astype(np.float32) / 255.0
-    mask_image = np.array(mask_image.convert("L")).astype(np.float32) / 255.0
-
-    assert init_image.shape[0:1] == mask_image.shape[0:1], "image and image_mask must have the same image size"
-    init_image[mask_image > 0.5] = -1.0  # set as masked pixel
-    init_image = np.expand_dims(init_image, 0).transpose(0, 3, 1, 2)
-    init_image = torch.from_numpy(init_image)
-    return init_image
-
-control_image = make_inpaint_condition(init_image, mask_image)
-```
-
-Now generate an image from the base, mask and control images. You'll notice features of the base image are strongly preserved in the generated image.
-
-```py
-prompt = "concept art digital painting of an elven castle, inspired by lord of the rings, highly detailed, 8k"
-image = pipeline(prompt=prompt, image=init_image, mask_image=mask_image, control_image=control_image).images[0]
-image
-```
-
-You can take this a step further and chain it with an image-to-image pipeline to apply a new [style](https://huggingface.co/nitrosocke/elden-ring-diffusion):
-
-```py
-from diffusers import AutoPipelineForImage2Image
-
-pipeline = AutoPipelineForImage2Image.from_pretrained(
-    "nitrosocke/elden-ring-diffusion", torch_dtype=torch.float16,
-).to("cuda")
-pipeline.enable_model_cpu_offload()
-pipeline.enable_xformers_memory_efficient_attention()
-
-prompt = "elden ring style castle" # include the token "elden ring style" in the prompt
-negative_prompt = "bad architecture, deformed, disfigured, poor details"
-
-image = pipeline(prompt, negative_prompt=negative_prompt, image=image).images[0]
-image
-```
-
-<div class="flex flex-row gap-4">
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">initial image</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-controlnet.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">ControlNet inpaint</figcaption>
-  </div>
-  <div class="flex-1">
-    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/inpaint-img2img.png"/>
-    <figcaption class="mt-2 text-center text-sm text-gray-500">image-to-image</figcaption>
-  </div>
-</div>
-
-## Optimize
-
-It can be difficult and slow to run diffusion models if you're resource constrained, but it dosen't have to be with a few optimization tricks. One of the biggest (and easiest) optimizations you can enable is switching to memory-efficient attention. If you're using PyTorch 2.0, [scaled-dot product attention](../optimization/torch2.0#scaled-dot-product-attention) is automatically enabled and you don't need to do anything else. For non-PyTorch 2.0 users, you can install and use [xFormers](../optimization/xformers)'s implementation of memory-efficient attention. Both options reduce memory usage and accelerate inference.
-
-You can also offload the model to the GPU to save even more memory:
-
-```diff
-+ pipeline.enable_xformers_memory_efficient_attention()
-+ pipeline.enable_model_cpu_offload()
-```
-
-To speed-up your inference code even more, use [`torch_compile`](../optimization/torch2.0#torch.compile). You should wrap `torch.compile` around the most intensive component in the pipeline which is typically the UNet:
-
-```py
-pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
-```
-
-Learn more in the [Reduce memory usage](../optimization/memory) and [Torch 2.0](../optimization/torch2.0) guides.
+Forcing the unmasked portion of the image to remain the same might result in some weird transitions between the unmasked and masked areas, since the model will typically change the masked and unmasked areas to make the transition more natural.

docs/source/en/using-diffusers/sdxl.md

@@ -39,7 +39,7 @@ pipeline = StableDiffusionXLPipeline.from_pretrained(
     "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
 ).to("cuda")
 
-refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+refiner = StableDiffusionXLImg2ImgPipeline.from_single_file(
     "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, use_safetensors=True, variant="fp16"
 ).to("cuda")
 ```
@@ -397,8 +397,6 @@ image = pipeline(prompt=prompt, prompt_2=prompt_2).images[0]
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl-double-prompt.png" alt="generated image of an astronaut in a jungle in the style of a van gogh painting"/>
 </div>
 
-The dual text-encoders also support textual inversion embeddings that need to be loaded separately as explained in the [SDXL textual inversion](textual_inversion_inference#stable-diffusion-xl] section.
-
 ## Optimizations
 
 SDXL is a large model, and you may need to optimize memory to get it to run on your hardware. Here are some tips to save memory and speed up inference.
@@ -428,4 +426,4 @@ SDXL is a large model, and you may need to optimize memory to get it to run on y
 
 ## Other resources
 
-If you're interested in experimenting with a minimal version of the [`UNet2DConditionModel`] used in SDXL, take a look at the [minSDXL](https://github.com/cloneofsimo/minSDXL) implementation which is written in PyTorch and directly compatible with 🤗 Diffusers.
+If you're interested in experimenting with a minimal version of the [`UNet2DConditionModel`] used in SDXL, take a look at the [minSDXL](https://github.com/cloneofsimo/minSDXL) implementation which is written in PyTorch and directly compatible with 🤗 Diffusers.

docs/source/en/using-diffusers/stable_diffusion_jax_how_to.md

@@ -1,41 +1,51 @@
-# JAX/Flax
+# 🧨 Stable Diffusion in JAX / Flax !
 
 [[open-in-colab]]
 
-🤗 Diffusers supports Flax for super fast inference on Google TPUs, such as those available in Colab, Kaggle or Google Cloud Platform. This guide shows you how to run inference with Stable Diffusion using JAX/Flax.
+🤗 Hugging Face [Diffusers](https://github.com/huggingface/diffusers) supports Flax since version `0.5.1`! This allows for super fast inference on Google TPUs, such as those available in Colab, Kaggle or Google Cloud Platform.
 
-Before you begin, make sure you have the necessary libraries installed:
+This notebook shows how to run inference using JAX / Flax. If you want more details about how Stable Diffusion works or want to run it in GPU, please refer to [this notebook](https://huggingface.co/docs/diffusers/stable_diffusion).
+
+First, make sure you are using a TPU backend. If you are running this notebook in Colab, select `Runtime` in the menu above, then select the option "Change runtime type" and then select `TPU` under the `Hardware accelerator` setting.
+
+Note that JAX is not exclusive to TPUs, but it shines on that hardware because each TPU server has 8 TPU accelerators working in parallel.
+
+## Setup
+
+First make sure diffusers is installed.
 
 ```py
 # uncomment to install the necessary libraries in Colab
-#!pip install -q jax==0.3.25 jaxlib==0.3.25 flax transformers ftfy
-#!pip install -q diffusers
+#!pip install jax==0.3.25 jaxlib==0.3.25 flax transformers ftfy
+#!pip install diffusers
 ```
 
-You should also make sure you're using a TPU backend. While JAX does not run exclusively on TPUs, you'll get the best performance on a TPU because each server has 8 TPU accelerators working in parallel.
+```python
+import jax.tools.colab_tpu
 
-If you are running this guide in Colab, select *Runtime* in the menu above, select the option *Change runtime type*, and then select *TPU* under the *Hardware accelerator* setting. Import JAX and quickly check whether you're using a TPU:
+jax.tools.colab_tpu.setup_tpu()
+import jax
+```
 
 ```python
-import jax
-import jax.tools.colab_tpu
-jax.tools.colab_tpu.setup_tpu()
-
 num_devices = jax.device_count()
 device_type = jax.devices()[0].device_kind
 
 print(f"Found {num_devices} JAX devices of type {device_type}.")
 assert (
-    "TPU" in device_type, 
-    "Available device is not a TPU, please select TPU from Edit > Notebook settings > Hardware accelerator"
-)
-"Found 8 JAX devices of type Cloud TPU."
+    "TPU" in device_type
+), "Available device is not a TPU, please select TPU from Edit > Notebook settings > Hardware accelerator"
 ```
 
-Great, now you can import the rest of the dependencies you'll need:
+```python out
+Found 8 JAX devices of type Cloud TPU.
+```
+
+Then we import all the dependencies.
 
 ```python
 import numpy as np
+import jax
 import jax.numpy as jnp
 
 from pathlib import Path
@@ -48,12 +58,17 @@ from huggingface_hub import notebook_login
 from diffusers import FlaxStableDiffusionPipeline
 ```
 
-## Load a model
+## Model Loading
 
-Flax is a functional framework, so models are stateless and parameters are stored outside of them. Loading a pretrained Flax pipeline returns *both* the pipeline and the model weights (or parameters). In this guide, you'll use `bfloat16`, a more efficient half-float type that is supported by TPUs (you can also use `float32` for full precision if you want).
+TPU devices support `bfloat16`, an efficient half-float type. We'll use it for our tests, but you can also use `float32` to use full precision instead.
 
 ```python
 dtype = jnp.bfloat16
+```
+
+Flax is a functional framework, so models are stateless and parameters are stored outside them. Loading the pre-trained Flax pipeline will return both the pipeline itself and the model weights (or parameters). We are using a `bf16` version of the weights, which leads to type warnings that you can safely ignore.
+
+```python
 pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
     "CompVis/stable-diffusion-v1-4",
     revision="bf16",
@@ -63,87 +78,95 @@ pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
 
 ## Inference
 
-TPUs usually have 8 devices working in parallel, so let's use the same prompt for each device. This means you can perform inference on 8 devices at once, with each device generating one image. As a result, you'll get 8 images in the same amount of time it takes for one chip to generate a single image!
+Since TPUs usually have 8 devices working in parallel, we'll replicate our prompt as many times as devices we have. Then we'll perform inference on the 8 devices at once, each responsible for generating one image. Thus, we'll get 8 images in the same amount of time it takes for one chip to generate a single one.
 
-<Tip>
-
-Learn more details in the [How does parallelization work?](#how-does-parallelization-work) section.
-
-</Tip>
-
-After replicating the prompt, get the tokenized text ids by calling the `prepare_inputs` function on the pipeline. The length of the tokenized text is set to 77 tokens as required by the configuration of the underlying CLIP text model.
+After replicating the prompt, we obtain the tokenized text ids by invoking the `prepare_inputs` function of the pipeline. The length of the tokenized text is set to 77 tokens, as required by the configuration of the underlying CLIP Text model.
 
 ```python
 prompt = "A cinematic film still of Morgan Freeman starring as Jimi Hendrix, portrait, 40mm lens, shallow depth of field, close up, split lighting, cinematic"
 prompt = [prompt] * jax.device_count()
 prompt_ids = pipeline.prepare_inputs(prompt)
 prompt_ids.shape
-"(8, 77)"
 ```
 
-Model parameters and inputs have to be replicated across the 8 parallel devices. The parameters dictionary is replicated with [`flax.jax_utils.replicate`](https://flax.readthedocs.io/en/latest/api_reference/flax.jax_utils.html#flax.jax_utils.replicate) which traverses the dictionary and changes the shape of the weights so they are repeated 8 times. Arrays are replicated using `shard`.
+```python out
+(8, 77)
+```
+
+### Replication and parallelization
+
+Model parameters and inputs have to be replicated across the 8 parallel devices we have. The parameters dictionary is replicated using `flax.jax_utils.replicate`, which traverses the dictionary and changes the shape of the weights so they are repeated 8 times. Arrays are replicated using `shard`.
 
 ```python
-# parameters
 p_params = replicate(params)
-
-# arrays
-prompt_ids = shard(prompt_ids)
-prompt_ids.shape
-"(8, 1, 77)"
 ```
 
-This shape means each one of the 8 devices receives as an input a `jnp` array with shape `(1, 77)`, where `1` is the batch size per device. On TPUs with sufficient memory, you could have a batch size larger than `1` if you want to generate multiple images (per chip) at once.
+```python
+prompt_ids = shard(prompt_ids)
+prompt_ids.shape
+```
 
-Next, create a random number generator to pass to the generation function. This is standard procedure in Flax, which is very serious and opinionated about random numbers. All functions that deal with random numbers are expected to receive a generator to ensure reproducibility, even when you're training across multiple distributed devices.
+```python out
+(8, 1, 77)
+```
 
-The helper function below uses a seed to initialize a random number generator. As long as you use the same seed, you'll get the exact same results. Feel free to use different seeds when exploring results later in the guide.
+That shape means that each one of the `8` devices will receive as an input a `jnp` array with shape `(1, 77)`. `1` is therefore the batch size per device. In TPUs with sufficient memory, it could be larger than `1` if we wanted to generate multiple images (per chip) at once.
+
+We are almost ready to generate images! We just need to create a random number generator to pass to the generation function. This is the standard procedure in Flax, which is very serious and opinionated about random numbers – all functions that deal with random numbers are expected to receive a generator. This ensures reproducibility, even when we are training across multiple distributed devices.
+
+The helper function below uses a seed to initialize a random number generator. As long as we use the same seed, we'll get the exact same results. Feel free to use different seeds when exploring results later in the notebook.
 
 ```python
 def create_key(seed=0):
     return jax.random.PRNGKey(seed)
 ```
 
-The helper function, or `rng`, is split 8 times so each device receives a different generator and generates a different image.
+We obtain a rng and then "split" it 8 times so each device receives a different generator. Therefore, each device will create a different image, and the full process is reproducible.
 
 ```python
 rng = create_key(0)
 rng = jax.random.split(rng, jax.device_count())
 ```
 
-To take advantage of JAX's optimized speed on a TPU, pass `jit=True` to the pipeline to compile the JAX code into an efficient representation and to ensure the model runs in parallel across the 8 devices.
+JAX code can be compiled to an efficient representation that runs very fast. However, we need to ensure that all inputs have the same shape in subsequent calls; otherwise, JAX will have to recompile the code, and we wouldn't be able to take advantage of the optimized speed.
 
-<Tip warning={true}>
+The Flax pipeline can compile the code for us if we pass `jit = True` as an argument. It will also ensure that the model runs in parallel in the 8 available devices.
 
-You need to ensure all your inputs have the same shape in subsequent calls, other JAX will need to recompile the code which is slower.
+The first time we run the following cell it will take a long time to compile, but subequent calls (even with different inputs) will be much faster. For example, it took more than a minute to compile in a TPU v2-8 when I tested, but then it takes about **`7s`** for future inference runs.
 
-</Tip>
-
-The first inference run takes more time because it needs to compile the code, but subsequent calls (even with different inputs) are much faster. For example, it took more than a minute to compile on a TPU v2-8, but then it takes about **7s** on a future inference run!
-
-```py
+```
 %%time
 images = pipeline(prompt_ids, p_params, rng, jit=True)[0]
-
-"CPU times: user 56.2 s, sys: 42.5 s, total: 1min 38s"
-"Wall time: 1min 29s"
 ```
 
-The returned array has shape `(8, 1, 512, 512, 3)` which should be reshaped to remove the second dimension and get 8 images of `512 × 512 × 3`. Then you can use the [`~utils.numpy_to_pil`] function to convert the arrays into images.
+```python out
+CPU times: user 56.2 s, sys: 42.5 s, total: 1min 38s
+Wall time: 1min 29s
+```
+
+The returned array has shape `(8, 1, 512, 512, 3)`. We reshape it to get rid of the second dimension and obtain 8 images of `512 × 512 × 3` and then convert them to PIL.
+
+```python
+images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:])
+images = pipeline.numpy_to_pil(images)
+```
+
+### Visualization
 
 ```python
 from diffusers import make_image_grid
 
-images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:])
-images = pipeline.numpy_to_pil(images)
 make_image_grid(images, 2, 4)
 ```
 
 ![img](https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/stable_diffusion_jax_how_to_cell_38_output_0.jpeg)
 
+
 ## Using different prompts
 
-You don't necessarily have to use the same prompt on all devices. For example, to generate 8 different prompts:
+We don't have to replicate the _same_ prompt in all the devices. We can do whatever we want: generate 2 prompts 4 times each, or even generate 8 different prompts at once. Let's do that!
+
+First, we'll refactor the input preparation code into a handy function:
 
 ```python
 prompts = [
@@ -156,7 +179,9 @@ prompts = [
     "Armchair in the shape of an avocado",
     "Clown astronaut in space, with Earth in the background",
 ]
+```
 
+```python
 prompt_ids = pipeline.prepare_inputs(prompts)
 prompt_ids = shard(prompt_ids)
 
@@ -172,41 +197,46 @@ make_image_grid(images, 2, 4)
 
 ## How does parallelization work?
 
-The Flax pipeline in 🤗 Diffusers automatically compiles the model and runs it in parallel on all available devices. Let's take a closer look at how that process works.
+We said before that the `diffusers` Flax pipeline automatically compiles the model and runs it in parallel on all available devices. We'll now briefly look inside that process to show how it works.
 
-JAX parallelization can be done in multiple ways. The easiest one revolves around using the [`jax.pmap`](https://jax.readthedocs.io/en/latest/_autosummary/jax.pmap.html) function to achieve single-program multiple-data (SPMD) parallelization. It means running several copies of the same code, each on different data inputs. More sophisticated approaches are possible, and you can go over to the JAX [documentation](https://jax.readthedocs.io/en/latest/index.html) to explore this topic in more detail if you are interested!
+JAX parallelization can be done in multiple ways. The easiest one revolves around using the `jax.pmap` function to achieve single-program, multiple-data (SPMD) parallelization. It means we'll run several copies of the same code, each on different data inputs. More sophisticated approaches are possible, we invite you to go over the [JAX documentation](https://jax.readthedocs.io/en/latest/index.html) and the [`pjit` pages](https://jax.readthedocs.io/en/latest/jax-101/08-pjit.html?highlight=pjit) to explore this topic if you are interested!
 
-`jax.pmap` does two things:
+`jax.pmap` does two things for us:
+- Compiles (or `jit`s) the code, as if we had invoked `jax.jit()`. This does not happen when we call `pmap`, but the first time the pmapped function is invoked.
+- Ensures the compiled code runs in parallel in all the available devices.
 
-1. Compiles (or "`jit`s") the code which is similar to `jax.jit()`. This does not happen when you call `pmap`, and only the first time the `pmap`ped function is called.
-2. Ensures the compiled code runs in parallel on all available devices.
-
-To demonstrate, call `pmap` on the pipeline's `_generate` method (this is a private method that generates images and may be renamed or removed in future releases of 🤗 Diffusers):
+To show how it works we `pmap` the `_generate` method of the pipeline, which is the private method that runs generates images. Please, note that this method may be renamed or removed in future releases of `diffusers`.
 
 ```python
 p_generate = pmap(pipeline._generate)
 ```
 
-After calling `pmap`, the prepared function `p_generate` will:
+After we use `pmap`, the prepared function `p_generate` will conceptually do the following:
+* Invoke a copy of the underlying function `pipeline._generate` in each device.
+* Send each device a different portion of the input arguments. That's what sharding is used for. In our case, `prompt_ids` has shape `(8, 1, 77, 768)`. This array will be split in `8` and each copy of `_generate` will receive an input with shape `(1, 77, 768)`.
 
-1. Make a copy of the underlying function, `pipeline._generate`, on each device.
-2. Send each device a different portion of the input arguments (this is why its necessary to call the *shard* function). In this case, `prompt_ids` has shape `(8, 1, 77, 768)` so the array is split into 8 and each copy of `_generate` receives an input with shape `(1, 77, 768)`.
+We can code `_generate` completely ignoring the fact that it will be invoked in parallel. We just care about our batch size (`1` in this example) and the dimensions that make sense for our code, and don't have to change anything to make it work in parallel.
 
-The most important thing to pay attention to here is the batch size (1 in this example), and the input dimensions that make sense for your code. You don't have to change anything else to make the code work in parallel.
+The same way as when we used the pipeline call, the first time we run the following cell it will take a while, but then it will be much faster.
 
-The first time you call the pipeline takes more time, but the calls afterward are much faster. The `block_until_ready` function is used to correctly measure inference time because JAX uses asynchronous dispatch and returns control to the Python loop as soon as it can. You don't need to use that in your code; blocking occurs automatically when you want to use the result of a computation that has not yet been materialized.
-
-```py
+```
 %%time
 images = p_generate(prompt_ids, p_params, rng)
 images = images.block_until_ready()
-"CPU times: user 1min 15s, sys: 18.2 s, total: 1min 34s"
-"Wall time: 1min 15s"
+images.shape
 ```
 
-Check your image dimensions to see if they're correct:
+```python out
+CPU times: user 1min 15s, sys: 18.2 s, total: 1min 34s
+Wall time: 1min 15s
+```
 
 ```python
 images.shape
-"(8, 1, 512, 512, 3)"
-```
+```
+
+```python out
+(8, 1, 512, 512, 3)
+```
+
+We use `block_until_ready()` to correctly measure inference time, because JAX uses asynchronous dispatch and returns control to the Python loop as soon as it can. You don't need to use that in your code; blocking will occur automatically when you want to use the result of a computation that has not yet been materialized.

docs/source/en/using-diffusers/textual_inversion_inference.md

@@ -28,8 +28,6 @@ from diffusers.utils import make_image_grid
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 ```
 
-## Stable Diffusion 1 and 2
-
 Pick a Stable Diffusion checkpoint and a pre-learned concept from the [Stable Diffusion Conceptualizer](https://huggingface.co/spaces/sd-concepts-library/stable-diffusion-conceptualizer):
 
 ```py
@@ -71,50 +69,3 @@ grid
 <div class="flex justify-center">
     <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/textual_inversion_inference.png">
 </div>
-
-
-## Stable Diffusion XL
-
-Stable Diffusion XL (SDXL) can also use textual inversion vectors for inference. In contrast to Stable Diffusion 1 and 2, SDXL has two text encoders so you'll need two textual inversion embeddings - one for each text encoder model.
-
-Let's download the SDXL textual inversion embeddings and have a closer look at it's structure:
-
-```py
-from huggingface_hub import hf_hub_download
-from safetensors.torch import load_file
-
-file = hf_hub_download("dn118/unaestheticXL", filename="unaestheticXLv31.safetensors")
-state_dict = load_file(file)
-state_dict
-```
-
-```
-{'clip_g': tensor([[ 0.0077, -0.0112,  0.0065,  ...,  0.0195,  0.0159,  0.0275],
-         ...,
-         [-0.0170,  0.0213,  0.0143,  ..., -0.0302, -0.0240, -0.0362]],
- 'clip_l': tensor([[ 0.0023,  0.0192,  0.0213,  ..., -0.0385,  0.0048, -0.0011],
-         ...,
-         [ 0.0475, -0.0508, -0.0145,  ...,  0.0070, -0.0089, -0.0163]],
-```
-
-There are two tensors, `"clip-g"` and `"clip-l"`.
-`"clip-g"` corresponds to the bigger text encoder in SDXL and refers to 
-`pipe.text_encoder_2` and `"clip-l"` refers to `pipe.text_encoder`.
-
-Now you can load each tensor separately by passing them along with the correct text encoder and tokenizer
-to [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`]:
-
-```py
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipe = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", variant="fp16", torch_dtype=torch.float16)
-pipe.to("cuda")
-
-pipe.load_textual_inversion(state_dict["clip_g"], token="unaestheticXLv31", text_encoder=pipe.text_encoder_2, tokenizer=pipe.tokenizer_2)
-pipe.load_textual_inversion(state_dict["clip_l"], token="unaestheticXLv31", text_encoder=pipe.text_encoder, tokenizer=pipe.tokenizer)
-
-# the embedding should be used as a negative embedding, so we pass it as a negative prompt
-generator = torch.Generator().manual_seed(33)
-image = pipe("a woman standing in front of a mountain", negative_prompt="unaestheticXLv31", generator=generator).images[0]
-```

docs/source/en/using-diffusers/using_safetensors

@@ -0,0 +1,19 @@
+# What is safetensors ? 
+
+[safetensors](https://github.com/huggingface/safetensors) is a different format
+from the classic `.bin` which uses Pytorch which uses pickle.
+
+Pickle is notoriously unsafe which allow any malicious file to execute arbitrary code.
+The hub itself tries to prevent issues from it, but it's not a silver bullet.
+
+`safetensors` first and foremost goal is to make loading machine learning models *safe*
+in the sense that no takeover of your computer can be done.
+
+# Why use safetensors ?
+
+**Safety** can be one reason, if you're attempting to use a not well known model and
+you're not sure about the source of the file.
+
+And a secondary reason, is **the speed of loading**. Safetensors can load models much faster
+than regular pickle files. If you spend a lot of times switching models, this can be
+a huge timesave.

docs/source/en/using-diffusers/write_own_pipeline.md

@@ -112,7 +112,7 @@ As you can see, this is already more complex than the DDPM pipeline which only c
 
 <Tip>
 
-💡 Read the [How does Stable Diffusion work?](https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work) blog for more details about how the VAE, UNet, and text encoder models work.
+💡 Read the [How does Stable Diffusion work?](https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work) blog for more details about how the VAE, UNet, and text encoder models.
 
 </Tip>
 
@@ -214,7 +214,7 @@ Next, generate some initial random noise as a starting point for the diffusion p
 
 ```py
 >>> latents = torch.randn(
-...     (batch_size, unet.config.in_channels, height // 8, width // 8),
+...     (batch_size, unet.in_channels, height // 8, width // 8),
 ...     generator=generator,
 ... )
 >>> latents = latents.to(torch_device)

docs/source/ko/installation.md

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 사용하시는 라이브러리에 맞는 🤗 Diffusers를 설치하세요.
 
-🤗 Diffusers는 Python 3.8+, PyTorch 1.7.0+ 및 flax에서 테스트되었습니다. 사용중인 딥러닝 라이브러리에 대한 아래의 설치 안내를 따르세요.
+🤗 Diffusers는 Python 3.7+, PyTorch 1.7.0+ 및 flax에서 테스트되었습니다. 사용중인 딥러닝 라이브러리에 대한 아래의 설치 안내를 따르세요.
 
 - [PyTorch 설치 안내](https://pytorch.org/get-started/locally/)
 - [Flax 설치 안내](https://flax.readthedocs.io/en/latest/)
@@ -105,7 +105,7 @@ pip install -e ".[flax]"
 
 이러한 명령어들은 저장소를 복제한 폴더와 Python 라이브러리 경로를 연결합니다.
 Python은 이제 일반 라이브러리 경로에 더하여 복제한 폴더 내부를 살펴봅니다.
-예를들어 Python 패키지가 `~/anaconda3/envs/main/lib/python3.8/site-packages/`에 설치되어 있는 경우 Python은 복제한 폴더인 `~/diffusers/`도 검색합니다.
+예를들어 Python 패키지가 `~/anaconda3/envs/main/lib/python3.7/site-packages/`에 설치되어 있는 경우 Python은 복제한 폴더인 `~/diffusers/`도 검색합니다.
 
 <Tip warning={true}>
 

docs/source/ko/using-diffusers/unconditional_image_generation.md

@@ -29,32 +29,26 @@ Unconditional 이미지 생성은 비교적 간단한 작업입니다. 모델이
 
 이 가이드에서는 unconditional 이미지 생성에 ['DiffusionPipeline']과 [DDPM](https://arxiv.org/abs/2006.11239)을 사용합니다:
 
-```python
+	```python
  >>> from diffusers import DiffusionPipeline
 
  >>> generator = DiffusionPipeline.from_pretrained("anton-l/ddpm-butterflies-128")
-```
-
+	```
 [diffusion 파이프라인]은 모든 모델링, 토큰화, 스케줄링 구성 요소를 다운로드하고 캐시합니다. 이 모델은 약 14억 개의 파라미터로 구성되어 있기 때문에 GPU에서 실행할 것을 강력히 권장합니다. PyTorch에서와 마찬가지로 제너레이터 객체를 GPU로 옮길 수 있습니다:
-
-```python
+	```python
  >>> generator.to("cuda")
-```
-
+	```
 이제 제너레이터를 사용하여 이미지를 생성할 수 있습니다:
-
-```python
+	```python
  >>> image = generator().images[0]
-```
-
+	```
 출력은 기본적으로 [PIL.Image](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) 객체로 감싸집니다.
 
 다음을 호출하여 이미지를 저장할 수 있습니다:
-
-```python
+	```python
  >>> image.save("generated_image.png")
-```
+	```
 	
 아래 스페이스(데모 링크)를 이용해 보고, 추론 단계의 매개변수를 자유롭게 조절하여 이미지 품질에 어떤 영향을 미치는지 확인해 보세요!
 
-<iframe src="https://stevhliu-ddpm-butterflies-128.hf.space" frameborder="0" width="850" height="500"></iframe>
+<iframe src="https://stevhliu-ddpm-butterflies-128.hf.space" frameborder="0" width="850" height="500"></iframe>

docs/source/zh/_toctree.yml

@@ -3,8 +3,6 @@
     title: 🧨 Diffusers
   - local: quicktour
     title: 快速入门
-  - local: stable_diffusion
-    title: 有效和高效的扩散
   - local: installation
     title: 安装
   title: 开始

docs/source/zh/installation.md

@@ -14,7 +14,7 @@ specific language governing permissions and limitations under the License.
 
 在你正在使用的任意深度学习框架中安装 🤗 Diffusers 。
 
-🤗 Diffusers已在Python 3.8+、PyTorch 1.7.0+和Flax上进行了测试。按照下面的安装说明，针对你正在使用的深度学习框架进行安装：
+🤗 Diffusers已在Python 3.7+、PyTorch 1.7.0+和Flax上进行了测试。按照下面的安装说明，针对你正在使用的深度学习框架进行安装：
 
 - [PyTorch](https://pytorch.org/get-started/locally/) installation instructions.
 - [Flax](https://flax.readthedocs.io/en/latest/) installation instructions.
@@ -107,7 +107,7 @@ pip install -e ".[flax]"
 
 这些命令将连接到你克隆的版本库和你的 Python 库路径。
 现在，不只是在通常的库路径，Python 还会在你克隆的文件夹内寻找包。
-例如，如果你的 Python 包通常安装在 `~/anaconda3/envs/main/lib/python3.8/Site-packages/`，Python 也会搜索你克隆到的文件夹。`~/diffusers/`。
+例如，如果你的 Python 包通常安装在 `~/anaconda3/envs/main/lib/python3.7/Site-packages/`，Python 也会搜索你克隆到的文件夹。`~/diffusers/`。
 
 <Tip warning={true}>
 

docs/source/zh/stable_diffusion.md

@@ -1,264 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
--->
-                                                               
-# 有效且高效的扩散
-
-[[open-in-colab]]
-
-让 [`DiffusionPipeline`] 生成特定风格或包含你所想要的内容的图像可能会有些棘手。 通常情况下，你需要多次运行 [`DiffusionPipeline`] 才能得到满意的图像。但是从无到有生成图像是一个计算密集的过程，特别是如果你要一遍又一遍地进行推理运算。
-
-这就是为什么从pipeline中获得最高的 *computational* (speed) 和 *memory* (GPU RAM) 非常重要 ，以减少推理周期之间的时间，从而使迭代速度更快。
-
-
-本教程将指导您如何通过 [`DiffusionPipeline`]  更快、更好地生成图像。
-
-
-首先，加载 [`runwayml/stable-diffusion-v1-5`](https://huggingface.co/runwayml/stable-diffusion-v1-5) 模型:
-
-```python
-from diffusers import DiffusionPipeline
-
-model_id = "runwayml/stable-diffusion-v1-5"
-pipeline = DiffusionPipeline.from_pretrained(model_id, use_safetensors=True)
-```
-
-本教程将使用的提示词是 [`portrait photo of a old warrior chief`] ，但是你可以随心所欲的想象和构造自己的提示词：
-
-```python
-prompt = "portrait photo of a old warrior chief"
-```
-
-## 速度
-
-<Tip>
-
-💡 如果你没有 GPU, 你可以从像 [Colab](https://colab.research.google.com/) 这样的 GPU 提供商获取免费的 GPU !
-
-</Tip>
-
-加速推理的最简单方法之一是将 pipeline 放在 GPU 上 ，就像使用任何 PyTorch 模块一样：
-
-```python
-pipeline = pipeline.to("cuda")
-```
-
-为了确保您可以使用相同的图像并对其进行改进，使用 [`Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) 方法，然后设置一个随机数种子 以确保其 [复现性](./using-diffusers/reproducibility):
-
-```python
-import torch
-
-generator = torch.Generator("cuda").manual_seed(0)
-```
-
-现在，你可以生成一个图像：
-
-```python
-image = pipeline(prompt, generator=generator).images[0]
-image
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_1.png">
-</div>
-
-在 T4 GPU 上，这个过程大概要30秒（如果你的 GPU 比 T4 好，可能会更快）。在默认情况下，[`DiffusionPipeline`] 使用完整的 `float32` 精度进行 50 步推理。你可以通过降低精度（如 `float16` ）或者减少推理步数来加速整个过程
-
-
-让我们把模型的精度降低至 `float16` ，然后生成一张图像：
-
-```python
-import torch
-
-pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16, use_safetensors=True)
-pipeline = pipeline.to("cuda")
-generator = torch.Generator("cuda").manual_seed(0)
-image = pipeline(prompt, generator=generator).images[0]
-image
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_2.png">
-</div>
-
-这一次，生成图像只花了约 11 秒，比之前快了近 3 倍！
-
-<Tip>
-
-💡 我们强烈建议把 pipeline 精度降低至 `float16` , 到目前为止, 我们很少看到输出质量有任何下降。
-
-</Tip>
-
-另一个选择是减少推理步数。 你可以选择一个更高效的调度器 (*scheduler*) 可以减少推理步数同时保证输出质量。您可以在 [DiffusionPipeline] 中通过调用compatibles方法找到与当前模型兼容的调度器 (*scheduler*)。 
-
-```python
-pipeline.scheduler.compatibles
-[
-    diffusers.schedulers.scheduling_lms_discrete.LMSDiscreteScheduler,
-    diffusers.schedulers.scheduling_unipc_multistep.UniPCMultistepScheduler,
-    diffusers.schedulers.scheduling_k_dpm_2_discrete.KDPM2DiscreteScheduler,
-    diffusers.schedulers.scheduling_deis_multistep.DEISMultistepScheduler,
-    diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler,
-    diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler,
-    diffusers.schedulers.scheduling_ddpm.DDPMScheduler,
-    diffusers.schedulers.scheduling_dpmsolver_singlestep.DPMSolverSinglestepScheduler,
-    diffusers.schedulers.scheduling_k_dpm_2_ancestral_discrete.KDPM2AncestralDiscreteScheduler,
-    diffusers.schedulers.scheduling_heun_discrete.HeunDiscreteScheduler,
-    diffusers.schedulers.scheduling_pndm.PNDMScheduler,
-    diffusers.schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteScheduler,
-    diffusers.schedulers.scheduling_ddim.DDIMScheduler,
-]
-```
-
-Stable Diffusion 模型默认使用的是 [`PNDMScheduler`] ，通常要大概50步推理, 但是像 [`DPMSolverMultistepScheduler`] 这样更高效的调度器只要大概 20 或 25 步推理. 使用 [`ConfigMixin.from_config`] 方法加载新的调度器:
-
-```python
-from diffusers import DPMSolverMultistepScheduler
-
-pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
-```
-
-现在将 `num_inference_steps` 设置为 20:
-
-```python
-generator = torch.Generator("cuda").manual_seed(0)
-image = pipeline(prompt, generator=generator, num_inference_steps=20).images[0]
-image
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_3.png">
-</div>
-
-太棒了！你成功把推理时间缩短到 4 秒！⚡️
-
-## 内存
-
-改善 pipeline 性能的另一个关键是减少内存的使用量，这间接意味着速度更快，因为你经常试图最大化每秒生成的图像数量。要想知道你一次可以生成多少张图片，最简单的方法是尝试不同的batch size，直到出现`OutOfMemoryError` (OOM)。
-
-创建一个函数，为每一批要生成的图像分配提示词和 `Generators` 。请务必为每个`Generator` 分配一个种子，以便于复现良好的结果。
-
-
-```python
-def get_inputs(batch_size=1):
-    generator = [torch.Generator("cuda").manual_seed(i) for i in range(batch_size)]
-    prompts = batch_size * [prompt]
-    num_inference_steps = 20
-
-    return {"prompt": prompts, "generator": generator, "num_inference_steps": num_inference_steps}
-```
-
-设置 `batch_size=4` ，然后看一看我们消耗了多少内存:
-
-```python
-from diffusers.utils import make_image_grid 
-
-images = pipeline(**get_inputs(batch_size=4)).images
-make_image_grid(images, 2, 2)
-```
-
-除非你有一个更大内存的GPU, 否则上述代码会返回 `OOM` 错误! 大部分内存被 cross-attention 层使用。按顺序运行可以节省大量内存，而不是在批处理中进行。你可以为 pipeline 配置 [`~DiffusionPipeline.enable_attention_slicing`] 函数:
-
-```python
-pipeline.enable_attention_slicing()
-```
-
-现在尝试把 `batch_size` 增加到 8!
-
-```python
-images = pipeline(**get_inputs(batch_size=8)).images
-make_image_grid(images, rows=2, cols=4)
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_5.png">
-</div>
-
-以前你不能一批生成 4 张图片，而现在你可以在一张图片里面生成八张图片而只需要大概3.5秒！这可能是 T4 GPU 在不牺牲质量的情况运行速度最快的一种方法。
-
-## 质量
-
-在最后两节中, 你要学习如何通过 `fp16` 来优化 pipeline 的速度, 通过使用性能更高的调度器来减少推理步数, 使用注意力切片（*enabling attention slicing*）方法来节省内存。现在，你将关注的是如何提高图像的质量。
-
-### 更好的 checkpoints
-
-有个显而易见的方法是使用更好的 checkpoints。 Stable Diffusion 模型是一个很好的起点, 自正式发布以来，还发布了几个改进版本。然而, 使用更新的版本并不意味着你会得到更好的结果。你仍然需要尝试不同的 checkpoints ，并做一些研究 (例如使用 [negative prompts](https://minimaxir.com/2022/11/stable-diffusion-negative-prompt/)) 来获得更好的结果。
-
-随着该领域的发展, 有越来越多经过微调的高质量的 checkpoints 用来生成不一样的风格. 在 [Hub](https://huggingface.co/models?library=diffusers&sort=downloads) 和 [Diffusers Gallery](https://huggingface.co/spaces/huggingface-projects/diffusers-gallery) 寻找你感兴趣的一种!
-
-### 更好的 pipeline 组件
-
-也可以尝试用新版本替换当前 pipeline 组件。让我们加载最新的 [autodecoder](https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main/vae) 从 Stability AI 加载到 pipeline, 并生成一些图像:
-
-```python
-from diffusers import AutoencoderKL
-
-vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16).to("cuda")
-pipeline.vae = vae
-images = pipeline(**get_inputs(batch_size=8)).images
-make_image_grid(images, rows=2, cols=4)
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_6.png">
-</div>
-
-### 更好的提示词工程
-
-用于生成图像的文本非常重要, 因此被称为 *提示词工程*。 在设计提示词工程应注意如下事项:
-
-- 我想生成的图像或类似图像如何存储在互联网上？
-- 我可以提供哪些额外的细节来引导模型朝着我想要的风格生成？
-
-考虑到这一点，让我们改进提示词，以包含颜色和更高质量的细节：
-
-```python
-prompt += ", tribal panther make up, blue on red, side profile, looking away, serious eyes"
-prompt += " 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta"
-```
-
-使用新的提示词生成一批图像:
-
-```python
-images = pipeline(**get_inputs(batch_size=8)).images
-make_image_grid(images, rows=2, cols=4)
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_7.png">
-</div>
-
-非常的令人印象深刻! Let's tweak the second image - 把 `Generator` 的种子设置为 `1` - 添加一些关于年龄的主题文本:
-
-```python
-prompts = [
-    "portrait photo of the oldest warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
-    "portrait photo of a old warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
-    "portrait photo of a warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
-    "portrait photo of a young warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
-]
-
-generator = [torch.Generator("cuda").manual_seed(1) for _ in range(len(prompts))]
-images = pipeline(prompt=prompts, generator=generator, num_inference_steps=25).images
-make_image_grid(images, 2, 2)
-```
-
-<div class="flex justify-center">
-    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_8.png">
-</div>
-
-## 最后
-
-在本教程中, 您学习了如何优化[`DiffusionPipeline`]以提高计算和内存效率，以及提高生成输出的质量. 如果你有兴趣让你的 pipeline 更快, 可以看一看以下资源:
-
-- 学习 [PyTorch 2.0](./optimization/torch2.0) 和 [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html) 可以让推理速度提高 5 - 300% . 在 A100 GPU 上, 推理速度可以提高 50% !
-- 如果你没法用 PyTorch 2, 我们建议你安装 [xFormers](./optimization/xformers)。它的内存高效注意力机制（*memory-efficient attention mechanism*）与PyTorch 1.13.1配合使用，速度更快，内存消耗更少。
-- 其他的优化技术, 如：模型卸载（*model offloading*）, 包含在 [这份指南](./optimization/fp16).

examples/community/README.md

@@ -43,7 +43,6 @@ If a community doesn't work as expected, please open an issue and ping the autho
 Stable Diffusion XL Long Weighted Prompt Pipeline | A pipeline support unlimited length of prompt and negative prompt, use A1111 style of prompt weighting | [Stable Diffusion XL Long Weighted Prompt Pipeline](#stable-diffusion-xl-long-weighted-prompt-pipeline) | - | [Andrew Zhu](https://xhinker.medium.com/) | 
 FABRIC - Stable Diffusion with feedback Pipeline | pipeline supports feedback from liked and disliked images | [Stable Diffusion Fabric Pipline](#stable-diffusion-fabric-pipeline) | - | [Shauray Singh](https://shauray8.github.io/about_shauray/) | 
 sketch inpaint - Inpainting with non-inpaint Stable Diffusion | sketch inpaint much like in automatic1111 | [Masked Im2Im Stable Diffusion Pipeline](#stable-diffusion-masked-im2im) | - | [Anatoly Belikov](https://github.com/noskill) | 
-prompt-to-prompt | change parts of a prompt and retain image structure (see [paper page](https://prompt-to-prompt.github.io/)) | [Prompt2Prompt Pipeline](#prompt2prompt-pipeline) | - | [Umer H. Adil](https://twitter.com/UmerHAdil) | 
 
 
 To load a custom pipeline you just need to pass the `custom_pipeline` argument to `DiffusionPipeline`, as one of the files in `diffusers/examples/community`. Feel free to send a PR with your own pipelines, we will merge them quickly.
@@ -2061,89 +2060,3 @@ result:
 
 <img src=https://github.com/noskill/diffusers/assets/733626/23a0a71d-51db-471e-926a-107ac62512a8 width="25%" >
 
-
-### Prompt2Prompt Pipeline
-
-Prompt2Prompt allows the following edits:
-- ReplaceEdit (change words in prompt)
-- ReplaceEdit with local blend (change words in prompt, keep image part unrelated to changes constant)
-- RefineEdit (add words to prompt)
-- RefineEdit with local blend (add words to prompt, keep image part unrelated to changes constant)
-- ReweightEdit (modulate importance of words)
-
-Here's a full example for `ReplaceEdit``:
-
-```python
-import torch
-import numpy as np
-import matplotlib.pyplot as plt
-from diffusers.pipelines import Prompt2PromptPipeline
-
-pipe = Prompt2PromptPipeline.from_pretrained("CompVis/stable-diffusion-v1-4").to("cuda")
-
-prompts = ["A turtle playing with a ball",
-           "A monkey playing with a ball"]
-
-cross_attention_kwargs = {
-    "edit_type": "replace",
-    "cross_replace_steps": 0.4,
-    "self_replace_steps": 0.4
-}
-
-outputs = pipe(prompt=prompts, height=512, width=512, num_inference_steps=50, cross_attention_kwargs=cross_attention_kwargs)
-```
-
-And abbreviated examples for the other edits:
-
-`ReplaceEdit with local blend`
-```python
-prompts = ["A turtle playing with a ball",
-           "A monkey playing with a ball"]
-
-cross_attention_kwargs = {
-    "edit_type": "replace",
-    "cross_replace_steps": 0.4,
-    "self_replace_steps": 0.4,
-    "local_blend_words": ["turtle", "monkey"]
-}
-```
-
-`RefineEdit`
-```python
-prompts = ["A turtle",
-           "A turtle in a forest"]
-
-cross_attention_kwargs = {
-    "edit_type": "refine",
-    "cross_replace_steps": 0.4,
-    "self_replace_steps": 0.4,
-}
-```
-
-`RefineEdit with local blend`
-```python
-prompts = ["A turtle",
-           "A turtle in a forest"]
-
-cross_attention_kwargs = {
-    "edit_type": "refine",
-    "cross_replace_steps": 0.4,
-    "self_replace_steps": 0.4,
-    "local_blend_words": ["in", "a" , "forest"]
-}
-```
-
-`ReweightEdit`
-```python
-prompts = ["A smiling turtle"] * 2
-
-edit_kcross_attention_kwargswargs = {
-    "edit_type": "reweight",
-    "cross_replace_steps": 0.4,
-    "self_replace_steps": 0.4,
-    "equalizer_words": ["smiling"],
-    "equalizer_strengths": [5]
-}
-```
-
-Side note: See [this GitHub gist](https://gist.github.com/UmerHA/b65bb5fb9626c9c73f3ade2869e36164) if you want to visualize the attention maps.

examples/community/clip_guided_images_mixing_stable_diffusion.py

@@ -3,7 +3,7 @@ import inspect
 from typing import Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from torch.nn import functional as F
 from torchvision import transforms

examples/community/clip_guided_stable_diffusion_img2img.py

@@ -2,7 +2,7 @@ import inspect
 from typing import List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from torch import nn
 from torch.nn import functional as F

examples/community/composable_stable_diffusion.py

@@ -562,8 +562,7 @@ class ComposableStableDiffusionPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # 8. Post-processing
         image = self.decode_latents(latents)

examples/community/ddim_noise_comparative_analysis.py

@@ -14,7 +14,7 @@
 
 from typing import List, Optional, Tuple, Union
 
-import PIL.Image
+import PIL
 import torch
 from torchvision import transforms
 

examples/community/imagic_stable_diffusion.py

@@ -7,7 +7,7 @@ import warnings
 from typing import List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 import torch.nn.functional as F
 from accelerate import Accelerator

examples/community/img2img_inpainting.py

@@ -2,7 +2,7 @@ import inspect
 from typing import Callable, List, Optional, Tuple, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
@@ -434,8 +434,7 @@ class ImageToImageInpaintingPipeline(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/community/interpolate_stable_diffusion.py

@@ -372,8 +372,7 @@ class StableDiffusionWalkPipeline(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/community/lpw_stable_diffusion.py

@@ -3,7 +3,7 @@ import re
 from typing import Any, Callable, Dict, List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from packaging import version
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
@@ -1088,8 +1088,7 @@ class StableDiffusionLongPromptWeightingPipeline(
                     progress_bar.update()
                     if i % callback_steps == 0:
                         if callback is not None:
-                            step_idx = i // getattr(self.scheduler, "order", 1)
-                            callback(step_idx, t, latents)
+                            callback(i, t, latents)
                         if is_cancelled_callback is not None and is_cancelled_callback():
                             return None
 

examples/community/lpw_stable_diffusion_onnx.py

@@ -3,7 +3,7 @@ import re
 from typing import Callable, List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from packaging import version
 from transformers import CLIPImageProcessor, CLIPTokenizer
@@ -846,8 +846,7 @@ class OnnxStableDiffusionLongPromptWeightingPipeline(OnnxStableDiffusionPipeline
             # call the callback, if provided
             if i % callback_steps == 0:
                 if callback is not None:
-                    step_idx = i // getattr(self.scheduler, "order", 1)
-                    callback(step_idx, t, latents)
+                    callback(i, t, latents)
                 if is_cancelled_callback is not None and is_cancelled_callback():
                     return None
 

examples/community/lpw_stable_diffusion_xl.py

@@ -1022,14 +1022,14 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
                 A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                 `self.processor` in
                 [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
-            guidance_rescale (`float`, *optional*, defaults to 0.0):
+            guidance_rescale (`float`, *optional*, defaults to 0.7):
                 Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
                 Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
                 [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
                 Guidance rescale factor should fix overexposure when using zero terminal SNR.
             original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                 If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
-                `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+                `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
                 explained in section 2.2 of
                 [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
             crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
@@ -1039,7 +1039,7 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
                 [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
             target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
                 For most cases, `target_size` should be set to the desired height and width of the generated image. If
-                not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+                not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
                 section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
 
         Examples:
@@ -1182,8 +1182,7 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             # make sure the VAE is in float32 mode, as it overflows in float16

examples/community/masked_stable_diffusion_img2img.py

@@ -1,7 +1,7 @@
 from typing import Any, Callable, Dict, List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 
 from diffusers import StableDiffusionImg2ImgPipeline
@@ -202,8 +202,7 @@ class MaskedStableDiffusionImg2ImgPipeline(StableDiffusionImg2ImgPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             scaled = latents / self.vae.config.scaling_factor

examples/community/multilingual_stable_diffusion.py

@@ -407,8 +407,7 @@ class MultilingualStableDiffusion(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/community/pipeline_prompt2prompt.py

@@ -1,860 +0,0 @@
-# Copyright 2023 The HuggingFace Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from __future__ import annotations
-
-import abc
-from typing import Any, Callable, Dict, List, Optional, Tuple, Union
-
-import numpy as np
-import torch
-import torch.nn.functional as F
-
-from ...src.diffusers.models.attention import Attention
-from ...src.diffusers.pipelines.stable_diffusion import StableDiffusionPipeline, StableDiffusionPipelineOutput
-
-
-# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
-def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
-    """
-    std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
-    std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
-    # rescale the results from guidance (fixes overexposure)
-    noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
-    # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
-    noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
-    return noise_cfg
-
-
-class Prompt2PromptPipeline(StableDiffusionPipeline):
-    r"""
-    Args:
-    Prompt-to-Prompt-Pipeline for text-to-image generation using Stable Diffusion. This model inherits from
-    [`StableDiffusionPipeline`]. Check the superclass documentation for the generic methods the library implements for
-    all the pipelines (such as downloading or saving, running on a particular device, etc.)
-        vae ([`AutoencoderKL`]):
-            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
-        text_encoder ([`CLIPTextModel`]):
-            Frozen text-encoder. Stable Diffusion uses the text portion of
-            [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
-            the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
-        tokenizer (`CLIPTokenizer`):
-            Tokenizer of class
-            [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
-        unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents. scheduler
-        ([`SchedulerMixin`]):
-            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
-            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
-        safety_checker ([`StableDiffusionSafetyChecker`]):
-            Classification module that estimates whether generated images could be considered offensive or harmful.
-            Please, refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for details.
-        feature_extractor ([`CLIPFeatureExtractor`]):
-            Model that extracts features from generated images to be used as inputs for the `safety_checker`.
-    """
-    _optional_components = ["safety_checker", "feature_extractor"]
-
-    @torch.no_grad()
-    def __call__(
-        self,
-        prompt: Union[str, List[str]],
-        height: Optional[int] = None,
-        width: Optional[int] = None,
-        num_inference_steps: int = 50,
-        guidance_scale: float = 7.5,
-        negative_prompt: Optional[Union[str, List[str]]] = None,
-        num_images_per_prompt: Optional[int] = 1,
-        eta: float = 0.0,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        prompt_embeds: Optional[torch.FloatTensor] = None,
-        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
-        output_type: Optional[str] = "pil",
-        return_dict: bool = True,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
-        callback_steps: Optional[int] = 1,
-        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
-        guidance_rescale: float = 0.0,
-    ):
-        r"""
-        Function invoked when calling the pipeline for generation.
-
-        Args:
-            prompt (`str` or `List[str]`):
-                The prompt or prompts to guide the image generation.
-            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The height in pixels of the generated image.
-            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The width in pixels of the generated image.
-            num_inference_steps (`int`, *optional*, defaults to 50):
-                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
-                expense of slower inference.
-            guidance_scale (`float`, *optional*, defaults to 7.5):
-                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
-                `guidance_scale` is defined as `w` of equation 2. of [Imagen
-                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
-                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
-                usually at the expense of lower image quality.
-            negative_prompt (`str` or `List[str]`, *optional*):
-                The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
-                if `guidance_scale` is less than `1`).
-            num_images_per_prompt (`int`, *optional*, defaults to 1):
-                The number of images to generate per prompt.
-            eta (`float`, *optional*, defaults to 0.0):
-                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
-                [`schedulers.DDIMScheduler`], will be ignored for others.
-            generator (`torch.Generator`, *optional*):
-                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
-                to make generation deterministic.
-            latents (`torch.FloatTensor`, *optional*):
-                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
-                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
-                tensor will ge generated by sampling using the supplied random `generator`.
-            output_type (`str`, *optional*, defaults to `"pil"`):
-                The output format of the generate image. Choose between
-                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
-            return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
-                plain tuple.
-            callback (`Callable`, *optional*):
-                A function that will be called every `callback_steps` steps during inference. The function will be
-                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
-            callback_steps (`int`, *optional*, defaults to 1):
-                The frequency at which the `callback` function will be called. If not specified, the callback will be
-                called at every step.
-            cross_attention_kwargs (`dict`, *optional*):
-                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
-                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
-
-                The keyword arguments to configure the edit are:
-                - edit_type (`str`). The edit type to apply. Can be either of `replace`, `refine`, `reweight`.
-                - n_cross_replace (`int`): Number of diffusion steps in which cross attention should be replaced
-                - n_self_replace (`int`): Number of diffusion steps in which self attention should be replaced
-                - local_blend_words(`List[str]`, *optional*, default to `None`): Determines which area should be
-                  changed. If None, then the whole image can be changed.
-                - equalizer_words(`List[str]`, *optional*, default to `None`): Required for edit type `reweight`.
-                  Determines which words should be enhanced.
-                - equalizer_strengths (`List[float]`, *optional*, default to `None`) Required for edit type `reweight`.
-                  Determines which how much the words in `equalizer_words` should be enhanced.
-
-            guidance_rescale (`float`, *optional*, defaults to 0.0):
-                Guidance rescale factor from [Common Diffusion Noise Schedules and Sample Steps are
-                Flawed](https://arxiv.org/pdf/2305.08891.pdf). Guidance rescale factor should fix overexposure when
-                using zero terminal SNR.
-
-        Returns:
-            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
-            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple.
-            When returning a tuple, the first element is a list with the generated images, and the second element is a
-            list of `bool`s denoting whether the corresponding generated image likely represents "not-safe-for-work"
-            (nsfw) content, according to the `safety_checker`.
-        """
-
-        self.controller = create_controller(
-            prompt, cross_attention_kwargs, num_inference_steps, tokenizer=self.tokenizer, device=self.device
-        )
-        self.register_attention_control(self.controller)  # add attention controller
-
-        # 0. Default height and width to unet
-        height = height or self.unet.config.sample_size * self.vae_scale_factor
-        width = width or self.unet.config.sample_size * self.vae_scale_factor
-
-        # 1. Check inputs. Raise error if not correct
-        self.check_inputs(prompt, height, width, callback_steps)
-
-        # 2. Define call parameters
-        if prompt is not None and isinstance(prompt, str):
-            batch_size = 1
-        elif prompt is not None and isinstance(prompt, list):
-            batch_size = len(prompt)
-        else:
-            batch_size = prompt_embeds.shape[0]
-
-        device = self._execution_device
-        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
-        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
-        # corresponds to doing no classifier free guidance.
-        do_classifier_free_guidance = guidance_scale > 1.0
-
-        # 3. Encode input prompt
-        text_encoder_lora_scale = (
-            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
-        )
-        prompt_embeds = self._encode_prompt(
-            prompt,
-            device,
-            num_images_per_prompt,
-            do_classifier_free_guidance,
-            negative_prompt,
-            prompt_embeds=prompt_embeds,
-            negative_prompt_embeds=negative_prompt_embeds,
-            lora_scale=text_encoder_lora_scale,
-        )
-
-        # 4. Prepare timesteps
-        self.scheduler.set_timesteps(num_inference_steps, device=device)
-        timesteps = self.scheduler.timesteps
-
-        # 5. Prepare latent variables
-        num_channels_latents = self.unet.config.in_channels
-        latents = self.prepare_latents(
-            batch_size * num_images_per_prompt,
-            num_channels_latents,
-            height,
-            width,
-            prompt_embeds.dtype,
-            device,
-            generator,
-            latents,
-        )
-
-        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
-        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
-
-        # 7. Denoising loop
-        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
-        with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for i, t in enumerate(timesteps):
-                # expand the latents if we are doing classifier free guidance
-                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
-
-                # predict the noise residual
-                noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=prompt_embeds).sample
-
-                # perform guidance
-                if do_classifier_free_guidance:
-                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
-                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
-
-                if do_classifier_free_guidance and guidance_rescale > 0.0:
-                    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
-                    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
-
-                # compute the previous noisy sample x_t -> x_t-1
-                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
-
-                # step callback
-                latents = self.controller.step_callback(latents)
-
-                # call the callback, if provided
-                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
-                    progress_bar.update()
-                    if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
-
-        # 8. Post-processing
-        if not output_type == "latent":
-            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
-            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
-        else:
-            image = latents
-            has_nsfw_concept = None
-
-        # 9. Run safety checker
-        if has_nsfw_concept is None:
-            do_denormalize = [True] * image.shape[0]
-        else:
-            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
-
-        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
-
-        # Offload last model to CPU
-        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
-            self.final_offload_hook.offload()
-
-        if not return_dict:
-            return (image, has_nsfw_concept)
-
-        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
-
-    def register_attention_control(self, controller):
-        attn_procs = {}
-        cross_att_count = 0
-        for name in self.unet.attn_processors.keys():
-            None if name.endswith("attn1.processor") else self.unet.config.cross_attention_dim
-            if name.startswith("mid_block"):
-                self.unet.config.block_out_channels[-1]
-                place_in_unet = "mid"
-            elif name.startswith("up_blocks"):
-                block_id = int(name[len("up_blocks.")])
-                list(reversed(self.unet.config.block_out_channels))[block_id]
-                place_in_unet = "up"
-            elif name.startswith("down_blocks"):
-                block_id = int(name[len("down_blocks.")])
-                self.unet.config.block_out_channels[block_id]
-                place_in_unet = "down"
-            else:
-                continue
-            cross_att_count += 1
-            attn_procs[name] = P2PCrossAttnProcessor(controller=controller, place_in_unet=place_in_unet)
-
-        self.unet.set_attn_processor(attn_procs)
-        controller.num_att_layers = cross_att_count
-
-
-class P2PCrossAttnProcessor:
-    def __init__(self, controller, place_in_unet):
-        super().__init__()
-        self.controller = controller
-        self.place_in_unet = place_in_unet
-
-    def __call__(self, attn: Attention, hidden_states, encoder_hidden_states=None, attention_mask=None):
-        batch_size, sequence_length, _ = hidden_states.shape
-        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
-
-        query = attn.to_q(hidden_states)
-
-        is_cross = encoder_hidden_states is not None
-        encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
-        key = attn.to_k(encoder_hidden_states)
-        value = attn.to_v(encoder_hidden_states)
-
-        query = attn.head_to_batch_dim(query)
-        key = attn.head_to_batch_dim(key)
-        value = attn.head_to_batch_dim(value)
-
-        attention_probs = attn.get_attention_scores(query, key, attention_mask)
-
-        # one line change
-        self.controller(attention_probs, is_cross, self.place_in_unet)
-
-        hidden_states = torch.bmm(attention_probs, value)
-        hidden_states = attn.batch_to_head_dim(hidden_states)
-
-        # linear proj
-        hidden_states = attn.to_out[0](hidden_states)
-        # dropout
-        hidden_states = attn.to_out[1](hidden_states)
-
-        return hidden_states
-
-
-def create_controller(
-    prompts: List[str], cross_attention_kwargs: Dict, num_inference_steps: int, tokenizer, device
-) -> AttentionControl:
-    edit_type = cross_attention_kwargs.get("edit_type", None)
-    local_blend_words = cross_attention_kwargs.get("local_blend_words", None)
-    equalizer_words = cross_attention_kwargs.get("equalizer_words", None)
-    equalizer_strengths = cross_attention_kwargs.get("equalizer_strengths", None)
-    n_cross_replace = cross_attention_kwargs.get("n_cross_replace", 0.4)
-    n_self_replace = cross_attention_kwargs.get("n_self_replace", 0.4)
-
-    # only replace
-    if edit_type == "replace" and local_blend_words is None:
-        return AttentionReplace(
-            prompts, num_inference_steps, n_cross_replace, n_self_replace, tokenizer=tokenizer, device=device
-        )
-
-    # replace + localblend
-    if edit_type == "replace" and local_blend_words is not None:
-        lb = LocalBlend(prompts, local_blend_words, tokenizer=tokenizer, device=device)
-        return AttentionReplace(
-            prompts, num_inference_steps, n_cross_replace, n_self_replace, lb, tokenizer=tokenizer, device=device
-        )
-
-    # only refine
-    if edit_type == "refine" and local_blend_words is None:
-        return AttentionRefine(
-            prompts, num_inference_steps, n_cross_replace, n_self_replace, tokenizer=tokenizer, device=device
-        )
-
-    # refine + localblend
-    if edit_type == "refine" and local_blend_words is not None:
-        lb = LocalBlend(prompts, local_blend_words, tokenizer=tokenizer, device=device)
-        return AttentionRefine(
-            prompts, num_inference_steps, n_cross_replace, n_self_replace, lb, tokenizer=tokenizer, device=device
-        )
-
-    # reweight
-    if edit_type == "reweight":
-        assert (
-            equalizer_words is not None and equalizer_strengths is not None
-        ), "To use reweight edit, please specify equalizer_words and equalizer_strengths."
-        assert len(equalizer_words) == len(
-            equalizer_strengths
-        ), "equalizer_words and equalizer_strengths must be of same length."
-        equalizer = get_equalizer(prompts[1], equalizer_words, equalizer_strengths, tokenizer=tokenizer)
-        return AttentionReweight(
-            prompts,
-            num_inference_steps,
-            n_cross_replace,
-            n_self_replace,
-            tokenizer=tokenizer,
-            device=device,
-            equalizer=equalizer,
-        )
-
-    raise ValueError(f"Edit type {edit_type} not recognized. Use one of: replace, refine, reweight.")
-
-
-class AttentionControl(abc.ABC):
-    def step_callback(self, x_t):
-        return x_t
-
-    def between_steps(self):
-        return
-
-    @property
-    def num_uncond_att_layers(self):
-        return 0
-
-    @abc.abstractmethod
-    def forward(self, attn, is_cross: bool, place_in_unet: str):
-        raise NotImplementedError
-
-    def __call__(self, attn, is_cross: bool, place_in_unet: str):
-        if self.cur_att_layer >= self.num_uncond_att_layers:
-            h = attn.shape[0]
-            attn[h // 2 :] = self.forward(attn[h // 2 :], is_cross, place_in_unet)
-        self.cur_att_layer += 1
-        if self.cur_att_layer == self.num_att_layers + self.num_uncond_att_layers:
-            self.cur_att_layer = 0
-            self.cur_step += 1
-            self.between_steps()
-        return attn
-
-    def reset(self):
-        self.cur_step = 0
-        self.cur_att_layer = 0
-
-    def __init__(self):
-        self.cur_step = 0
-        self.num_att_layers = -1
-        self.cur_att_layer = 0
-
-
-class EmptyControl(AttentionControl):
-    def forward(self, attn, is_cross: bool, place_in_unet: str):
-        return attn
-
-
-class AttentionStore(AttentionControl):
-    @staticmethod
-    def get_empty_store():
-        return {"down_cross": [], "mid_cross": [], "up_cross": [], "down_self": [], "mid_self": [], "up_self": []}
-
-    def forward(self, attn, is_cross: bool, place_in_unet: str):
-        key = f"{place_in_unet}_{'cross' if is_cross else 'self'}"
-        if attn.shape[1] <= 32**2:  # avoid memory overhead
-            self.step_store[key].append(attn)
-        return attn
-
-    def between_steps(self):
-        if len(self.attention_store) == 0:
-            self.attention_store = self.step_store
-        else:
-            for key in self.attention_store:
-                for i in range(len(self.attention_store[key])):
-                    self.attention_store[key][i] += self.step_store[key][i]
-        self.step_store = self.get_empty_store()
-
-    def get_average_attention(self):
-        average_attention = {
-            key: [item / self.cur_step for item in self.attention_store[key]] for key in self.attention_store
-        }
-        return average_attention
-
-    def reset(self):
-        super(AttentionStore, self).reset()
-        self.step_store = self.get_empty_store()
-        self.attention_store = {}
-
-    def __init__(self):
-        super(AttentionStore, self).__init__()
-        self.step_store = self.get_empty_store()
-        self.attention_store = {}
-
-
-class LocalBlend:
-    def __call__(self, x_t, attention_store):
-        k = 1
-        maps = attention_store["down_cross"][2:4] + attention_store["up_cross"][:3]
-        maps = [item.reshape(self.alpha_layers.shape[0], -1, 1, 16, 16, self.max_num_words) for item in maps]
-        maps = torch.cat(maps, dim=1)
-        maps = (maps * self.alpha_layers).sum(-1).mean(1)
-        mask = F.max_pool2d(maps, (k * 2 + 1, k * 2 + 1), (1, 1), padding=(k, k))
-        mask = F.interpolate(mask, size=(x_t.shape[2:]))
-        mask = mask / mask.max(2, keepdims=True)[0].max(3, keepdims=True)[0]
-        mask = mask.gt(self.threshold)
-        mask = (mask[:1] + mask[1:]).float()
-        x_t = x_t[:1] + mask * (x_t - x_t[:1])
-        return x_t
-
-    def __init__(
-        self, prompts: List[str], words: [List[List[str]]], tokenizer, device, threshold=0.3, max_num_words=77
-    ):
-        self.max_num_words = 77
-
-        alpha_layers = torch.zeros(len(prompts), 1, 1, 1, 1, self.max_num_words)
-        for i, (prompt, words_) in enumerate(zip(prompts, words)):
-            if isinstance(words_, str):
-                words_ = [words_]
-            for word in words_:
-                ind = get_word_inds(prompt, word, tokenizer)
-                alpha_layers[i, :, :, :, :, ind] = 1
-        self.alpha_layers = alpha_layers.to(device)
-        self.threshold = threshold
-
-
-class AttentionControlEdit(AttentionStore, abc.ABC):
-    def step_callback(self, x_t):
-        if self.local_blend is not None:
-            x_t = self.local_blend(x_t, self.attention_store)
-        return x_t
-
-    def replace_self_attention(self, attn_base, att_replace):
-        if att_replace.shape[2] <= 16**2:
-            return attn_base.unsqueeze(0).expand(att_replace.shape[0], *attn_base.shape)
-        else:
-            return att_replace
-
-    @abc.abstractmethod
-    def replace_cross_attention(self, attn_base, att_replace):
-        raise NotImplementedError
-
-    def forward(self, attn, is_cross: bool, place_in_unet: str):
-        super(AttentionControlEdit, self).forward(attn, is_cross, place_in_unet)
-        # FIXME not replace correctly
-        if is_cross or (self.num_self_replace[0] <= self.cur_step < self.num_self_replace[1]):
-            h = attn.shape[0] // (self.batch_size)
-            attn = attn.reshape(self.batch_size, h, *attn.shape[1:])
-            attn_base, attn_repalce = attn[0], attn[1:]
-            if is_cross:
-                alpha_words = self.cross_replace_alpha[self.cur_step]
-                attn_repalce_new = (
-                    self.replace_cross_attention(attn_base, attn_repalce) * alpha_words
-                    + (1 - alpha_words) * attn_repalce
-                )
-                attn[1:] = attn_repalce_new
-            else:
-                attn[1:] = self.replace_self_attention(attn_base, attn_repalce)
-            attn = attn.reshape(self.batch_size * h, *attn.shape[2:])
-        return attn
-
-    def __init__(
-        self,
-        prompts,
-        num_steps: int,
-        cross_replace_steps: Union[float, Tuple[float, float], Dict[str, Tuple[float, float]]],
-        self_replace_steps: Union[float, Tuple[float, float]],
-        local_blend: Optional[LocalBlend],
-        tokenizer,
-        device,
-    ):
-        super(AttentionControlEdit, self).__init__()
-        # add tokenizer and device here
-
-        self.tokenizer = tokenizer
-        self.device = device
-
-        self.batch_size = len(prompts)
-        self.cross_replace_alpha = get_time_words_attention_alpha(
-            prompts, num_steps, cross_replace_steps, self.tokenizer
-        ).to(self.device)
-        if isinstance(self_replace_steps, float):
-            self_replace_steps = 0, self_replace_steps
-        self.num_self_replace = int(num_steps * self_replace_steps[0]), int(num_steps * self_replace_steps[1])
-        self.local_blend = local_blend  # 在外面定义后传进来
-
-
-class AttentionReplace(AttentionControlEdit):
-    def replace_cross_attention(self, attn_base, att_replace):
-        return torch.einsum("hpw,bwn->bhpn", attn_base, self.mapper)
-
-    def __init__(
-        self,
-        prompts,
-        num_steps: int,
-        cross_replace_steps: float,
-        self_replace_steps: float,
-        local_blend: Optional[LocalBlend] = None,
-        tokenizer=None,
-        device=None,
-    ):
-        super(AttentionReplace, self).__init__(
-            prompts, num_steps, cross_replace_steps, self_replace_steps, local_blend, tokenizer, device
-        )
-        self.mapper = get_replacement_mapper(prompts, self.tokenizer).to(self.device)
-
-
-class AttentionRefine(AttentionControlEdit):
-    def replace_cross_attention(self, attn_base, att_replace):
-        attn_base_replace = attn_base[:, :, self.mapper].permute(2, 0, 1, 3)
-        attn_replace = attn_base_replace * self.alphas + att_replace * (1 - self.alphas)
-        return attn_replace
-
-    def __init__(
-        self,
-        prompts,
-        num_steps: int,
-        cross_replace_steps: float,
-        self_replace_steps: float,
-        local_blend: Optional[LocalBlend] = None,
-        tokenizer=None,
-        device=None,
-    ):
-        super(AttentionRefine, self).__init__(
-            prompts, num_steps, cross_replace_steps, self_replace_steps, local_blend, tokenizer, device
-        )
-        self.mapper, alphas = get_refinement_mapper(prompts, self.tokenizer)
-        self.mapper, alphas = self.mapper.to(self.device), alphas.to(self.device)
-        self.alphas = alphas.reshape(alphas.shape[0], 1, 1, alphas.shape[1])
-
-
-class AttentionReweight(AttentionControlEdit):
-    def replace_cross_attention(self, attn_base, att_replace):
-        if self.prev_controller is not None:
-            attn_base = self.prev_controller.replace_cross_attention(attn_base, att_replace)
-        attn_replace = attn_base[None, :, :, :] * self.equalizer[:, None, None, :]
-        return attn_replace
-
-    def __init__(
-        self,
-        prompts,
-        num_steps: int,
-        cross_replace_steps: float,
-        self_replace_steps: float,
-        equalizer,
-        local_blend: Optional[LocalBlend] = None,
-        controller: Optional[AttentionControlEdit] = None,
-        tokenizer=None,
-        device=None,
-    ):
-        super(AttentionReweight, self).__init__(
-            prompts, num_steps, cross_replace_steps, self_replace_steps, local_blend, tokenizer, device
-        )
-        self.equalizer = equalizer.to(self.device)
-        self.prev_controller = controller
-
-
-### util functions for all Edits
-def update_alpha_time_word(
-    alpha, bounds: Union[float, Tuple[float, float]], prompt_ind: int, word_inds: Optional[torch.Tensor] = None
-):
-    if isinstance(bounds, float):
-        bounds = 0, bounds
-    start, end = int(bounds[0] * alpha.shape[0]), int(bounds[1] * alpha.shape[0])
-    if word_inds is None:
-        word_inds = torch.arange(alpha.shape[2])
-    alpha[:start, prompt_ind, word_inds] = 0
-    alpha[start:end, prompt_ind, word_inds] = 1
-    alpha[end:, prompt_ind, word_inds] = 0
-    return alpha
-
-
-def get_time_words_attention_alpha(
-    prompts, num_steps, cross_replace_steps: Union[float, Dict[str, Tuple[float, float]]], tokenizer, max_num_words=77
-):
-    if not isinstance(cross_replace_steps, dict):
-        cross_replace_steps = {"default_": cross_replace_steps}
-    if "default_" not in cross_replace_steps:
-        cross_replace_steps["default_"] = (0.0, 1.0)
-    alpha_time_words = torch.zeros(num_steps + 1, len(prompts) - 1, max_num_words)
-    for i in range(len(prompts) - 1):
-        alpha_time_words = update_alpha_time_word(alpha_time_words, cross_replace_steps["default_"], i)
-    for key, item in cross_replace_steps.items():
-        if key != "default_":
-            inds = [get_word_inds(prompts[i], key, tokenizer) for i in range(1, len(prompts))]
-            for i, ind in enumerate(inds):
-                if len(ind) > 0:
-                    alpha_time_words = update_alpha_time_word(alpha_time_words, item, i, ind)
-    alpha_time_words = alpha_time_words.reshape(num_steps + 1, len(prompts) - 1, 1, 1, max_num_words)
-    return alpha_time_words
-
-
-### util functions for LocalBlend and ReplacementEdit
-def get_word_inds(text: str, word_place: int, tokenizer):
-    split_text = text.split(" ")
-    if isinstance(word_place, str):
-        word_place = [i for i, word in enumerate(split_text) if word_place == word]
-    elif isinstance(word_place, int):
-        word_place = [word_place]
-    out = []
-    if len(word_place) > 0:
-        words_encode = [tokenizer.decode([item]).strip("#") for item in tokenizer.encode(text)][1:-1]
-        cur_len, ptr = 0, 0
-
-        for i in range(len(words_encode)):
-            cur_len += len(words_encode[i])
-            if ptr in word_place:
-                out.append(i + 1)
-            if cur_len >= len(split_text[ptr]):
-                ptr += 1
-                cur_len = 0
-    return np.array(out)
-
-
-### util functions for ReplacementEdit
-def get_replacement_mapper_(x: str, y: str, tokenizer, max_len=77):
-    words_x = x.split(" ")
-    words_y = y.split(" ")
-    if len(words_x) != len(words_y):
-        raise ValueError(
-            f"attention replacement edit can only be applied on prompts with the same length"
-            f" but prompt A has {len(words_x)} words and prompt B has {len(words_y)} words."
-        )
-    inds_replace = [i for i in range(len(words_y)) if words_y[i] != words_x[i]]
-    inds_source = [get_word_inds(x, i, tokenizer) for i in inds_replace]
-    inds_target = [get_word_inds(y, i, tokenizer) for i in inds_replace]
-    mapper = np.zeros((max_len, max_len))
-    i = j = 0
-    cur_inds = 0
-    while i < max_len and j < max_len:
-        if cur_inds < len(inds_source) and inds_source[cur_inds][0] == i:
-            inds_source_, inds_target_ = inds_source[cur_inds], inds_target[cur_inds]
-            if len(inds_source_) == len(inds_target_):
-                mapper[inds_source_, inds_target_] = 1
-            else:
-                ratio = 1 / len(inds_target_)
-                for i_t in inds_target_:
-                    mapper[inds_source_, i_t] = ratio
-            cur_inds += 1
-            i += len(inds_source_)
-            j += len(inds_target_)
-        elif cur_inds < len(inds_source):
-            mapper[i, j] = 1
-            i += 1
-            j += 1
-        else:
-            mapper[j, j] = 1
-            i += 1
-            j += 1
-
-    return torch.from_numpy(mapper).float()
-
-
-def get_replacement_mapper(prompts, tokenizer, max_len=77):
-    x_seq = prompts[0]
-    mappers = []
-    for i in range(1, len(prompts)):
-        mapper = get_replacement_mapper_(x_seq, prompts[i], tokenizer, max_len)
-        mappers.append(mapper)
-    return torch.stack(mappers)
-
-
-### util functions for ReweightEdit
-def get_equalizer(
-    text: str, word_select: Union[int, Tuple[int, ...]], values: Union[List[float], Tuple[float, ...]], tokenizer
-):
-    if isinstance(word_select, (int, str)):
-        word_select = (word_select,)
-    equalizer = torch.ones(len(values), 77)
-    values = torch.tensor(values, dtype=torch.float32)
-    for word in word_select:
-        inds = get_word_inds(text, word, tokenizer)
-        equalizer[:, inds] = values
-    return equalizer
-
-
-### util functions for RefinementEdit
-class ScoreParams:
-    def __init__(self, gap, match, mismatch):
-        self.gap = gap
-        self.match = match
-        self.mismatch = mismatch
-
-    def mis_match_char(self, x, y):
-        if x != y:
-            return self.mismatch
-        else:
-            return self.match
-
-
-def get_matrix(size_x, size_y, gap):
-    matrix = np.zeros((size_x + 1, size_y + 1), dtype=np.int32)
-    matrix[0, 1:] = (np.arange(size_y) + 1) * gap
-    matrix[1:, 0] = (np.arange(size_x) + 1) * gap
-    return matrix
-
-
-def get_traceback_matrix(size_x, size_y):
-    matrix = np.zeros((size_x + 1, size_y + 1), dtype=np.int32)
-    matrix[0, 1:] = 1
-    matrix[1:, 0] = 2
-    matrix[0, 0] = 4
-    return matrix
-
-
-def global_align(x, y, score):
-    matrix = get_matrix(len(x), len(y), score.gap)
-    trace_back = get_traceback_matrix(len(x), len(y))
-    for i in range(1, len(x) + 1):
-        for j in range(1, len(y) + 1):
-            left = matrix[i, j - 1] + score.gap
-            up = matrix[i - 1, j] + score.gap
-            diag = matrix[i - 1, j - 1] + score.mis_match_char(x[i - 1], y[j - 1])
-            matrix[i, j] = max(left, up, diag)
-            if matrix[i, j] == left:
-                trace_back[i, j] = 1
-            elif matrix[i, j] == up:
-                trace_back[i, j] = 2
-            else:
-                trace_back[i, j] = 3
-    return matrix, trace_back
-
-
-def get_aligned_sequences(x, y, trace_back):
-    x_seq = []
-    y_seq = []
-    i = len(x)
-    j = len(y)
-    mapper_y_to_x = []
-    while i > 0 or j > 0:
-        if trace_back[i, j] == 3:
-            x_seq.append(x[i - 1])
-            y_seq.append(y[j - 1])
-            i = i - 1
-            j = j - 1
-            mapper_y_to_x.append((j, i))
-        elif trace_back[i][j] == 1:
-            x_seq.append("-")
-            y_seq.append(y[j - 1])
-            j = j - 1
-            mapper_y_to_x.append((j, -1))
-        elif trace_back[i][j] == 2:
-            x_seq.append(x[i - 1])
-            y_seq.append("-")
-            i = i - 1
-        elif trace_back[i][j] == 4:
-            break
-    mapper_y_to_x.reverse()
-    return x_seq, y_seq, torch.tensor(mapper_y_to_x, dtype=torch.int64)
-
-
-def get_mapper(x: str, y: str, tokenizer, max_len=77):
-    x_seq = tokenizer.encode(x)
-    y_seq = tokenizer.encode(y)
-    score = ScoreParams(0, 1, -1)
-    matrix, trace_back = global_align(x_seq, y_seq, score)
-    mapper_base = get_aligned_sequences(x_seq, y_seq, trace_back)[-1]
-    alphas = torch.ones(max_len)
-    alphas[: mapper_base.shape[0]] = mapper_base[:, 1].ne(-1).float()
-    mapper = torch.zeros(max_len, dtype=torch.int64)
-    mapper[: mapper_base.shape[0]] = mapper_base[:, 1]
-    mapper[mapper_base.shape[0] :] = len(y_seq) + torch.arange(max_len - len(y_seq))
-    return mapper, alphas
-
-
-def get_refinement_mapper(prompts, tokenizer, max_len=77):
-    x_seq = prompts[0]
-    mappers, alphas = [], []
-    for i in range(1, len(prompts)):
-        mapper, alpha = get_mapper(x_seq, prompts[i], tokenizer, max_len)
-        mappers.append(mapper)
-        alphas.append(alpha)
-    return torch.stack(mappers), torch.stack(alphas)

examples/community/pipeline_zero1to3.py

@@ -6,7 +6,7 @@ from typing import Any, Callable, Dict, List, Optional, Union
 
 import kornia
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from packaging import version
 from transformers import CLIPFeatureExtractor, CLIPVisionModelWithProjection
@@ -865,8 +865,7 @@ class Zero1to3StableDiffusionPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # 8. Post-processing
         has_nsfw_concept = None

examples/community/run_onnx_controlnet.py

@@ -8,6 +8,7 @@ from typing import Any, Callable, Dict, List, Optional, Union
 import numpy as np
 import PIL.Image
 import torch
+from diffuser.utils.torch_utils import randn_tensor
 from PIL import Image
 from transformers import CLIPTokenizer
 
@@ -21,7 +22,6 @@ from diffusers.utils import (
     logging,
     replace_example_docstring,
 )
-from diffusers.utils.torch_utils import randn_tensor
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -815,8 +815,7 @@ class OnnxStableDiffusionControlNetImg2ImgPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             _latents = latents.cpu().detach().numpy() / 0.18215
@@ -887,7 +886,7 @@ if __name__ == "__main__":
     onnx_pipeline = onnx_pipeline.to("cuda")
 
     prompt = "a cute cat fly to the moon"
-    negative_prompt = "paintings, sketches, worst quality, low quality, normal quality, lowres, normal quality, monochrome, grayscale, skin spots, acnes, skin blemishes, age spot, glans, nsfw, nipples, necklace, worst quality, low quality, watermark, username, signature, multiple breasts, lowres, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, single color, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, blurry, bad anatomy, bad proportions, extra limbs, disfigured, bad anatomy, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, mutated hands, fused fingers, too many fingers, long neck, bad body perspect"
+    negative_prompt = "paintings, sketches, worst quality, low quality, normal quality, lowres, normal quality, monochrome, grayscale, skin spots, acnes, skin blemishes, age spot, glans, nsfw, nipples, necklace, worst quality, low quality, watermark, username, signature, multiple breasts, lowres, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, single color, ugly, duplicate, morbid, mutilated, tranny, trans, trannsexual, hermaphrodite, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, blurry, bad anatomy, bad proportions, extra limbs, disfigured, bad anatomy, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, mutated hands, fused fingers, too many fingers, long neck, bad body perspect"
 
     for i in range(10):
         start_time = time.time()

examples/community/run_tensorrt_controlnet.py

@@ -11,6 +11,7 @@ import PIL.Image
 import pycuda.driver as cuda
 import tensorrt as trt
 import torch
+from diffuser.utils.torch_utils import randn_tensor
 from PIL import Image
 from pycuda.tools import make_default_context
 from transformers import CLIPTokenizer
@@ -25,7 +26,6 @@ from diffusers.utils import (
     logging,
     replace_example_docstring,
 )
-from diffusers.utils.torch_utils import randn_tensor
 
 
 # Initialize CUDA
@@ -919,8 +919,7 @@ class TensorRTStableDiffusionControlNetImg2ImgPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             _latents = latents.cpu().detach().numpy() / 0.18215
@@ -998,7 +997,7 @@ if __name__ == "__main__":
     onnx_pipeline = onnx_pipeline.to("cuda")
 
     prompt = "a cute cat fly to the moon"
-    negative_prompt = "paintings, sketches, worst quality, low quality, normal quality, lowres, normal quality, monochrome, grayscale, skin spots, acnes, skin blemishes, age spot, glans, nsfw, nipples, necklace, worst quality, low quality, watermark, username, signature, multiple breasts, lowres, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, single color, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, blurry, bad anatomy, bad proportions, extra limbs, disfigured, bad anatomy, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, mutated hands, fused fingers, too many fingers, long neck, bad body perspect"
+    negative_prompt = "paintings, sketches, worst quality, low quality, normal quality, lowres, normal quality, monochrome, grayscale, skin spots, acnes, skin blemishes, age spot, glans, nsfw, nipples, necklace, worst quality, low quality, watermark, username, signature, multiple breasts, lowres, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, single color, ugly, duplicate, morbid, mutilated, tranny, trans, trannsexual, hermaphrodite, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, ugly, blurry, bad anatomy, bad proportions, extra limbs, disfigured, bad anatomy, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, mutated hands, fused fingers, too many fingers, long neck, bad body perspect"
 
     for i in range(10):
         start_time = time.time()

examples/community/seed_resize_stable_diffusion.py

@@ -337,8 +337,7 @@ class SeedResizeStableDiffusionPipeline(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/community/speech_to_image_diffusion.py

@@ -242,8 +242,7 @@ class SpeechToImagePipeline(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/community/stable_diffusion_controlnet_img2img.py

@@ -951,8 +951,7 @@ class StableDiffusionControlNetImg2ImgPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings

examples/community/stable_diffusion_controlnet_inpaint.py

@@ -1100,8 +1100,7 @@ class StableDiffusionControlNetInpaintPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings

examples/community/stable_diffusion_controlnet_inpaint_img2img.py

@@ -1081,8 +1081,7 @@ class StableDiffusionControlNetInpaintImg2ImgPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings

examples/community/stable_diffusion_controlnet_reference.py

@@ -802,8 +802,7 @@ class StableDiffusionControlNetReferencePipeline(StableDiffusionControlNetPipeli
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings

examples/community/stable_diffusion_ipex.py

@@ -817,8 +817,7 @@ class StableDiffusionIPEXPipeline(DiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if output_type == "latent":
             image = latents

examples/community/stable_diffusion_reference.py

@@ -249,7 +249,7 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
                 `self.processor` in
                 [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
-            guidance_rescale (`float`, *optional*, defaults to 0.0):
+            guidance_rescale (`float`, *optional*, defaults to 0.7):
                 Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
                 Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
                 [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
@@ -770,8 +770,7 @@ class StableDiffusionReferencePipeline(StableDiffusionPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]

examples/community/stable_diffusion_repaint.py

@@ -16,7 +16,7 @@ import inspect
 from typing import Callable, List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from packaging import version
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
@@ -932,8 +932,7 @@ class StableDiffusionRepaintPipeline(DiffusionPipeline, TextualInversionLoaderMi
                 # call the callback, if provided
                 progress_bar.update()
                 if callback is not None and i % callback_steps == 0:
-                    step_idx = i // getattr(self.scheduler, "order", 1)
-                    callback(step_idx, t, latents)
+                    callback(i, t, latents)
 
                 t_last = t
 

examples/community/stable_diffusion_tensorrt_img2img.py

@@ -24,7 +24,7 @@ from typing import List, Optional, Union
 import numpy as np
 import onnx
 import onnx_graphsurgeon as gs
-import PIL.Image
+import PIL
 import tensorrt as trt
 import torch
 from huggingface_hub import snapshot_download

examples/community/stable_diffusion_tensorrt_inpaint.py

@@ -24,7 +24,7 @@ from typing import List, Optional, Union
 import numpy as np
 import onnx
 import onnx_graphsurgeon as gs
-import PIL.Image
+import PIL
 import tensorrt as trt
 import torch
 from huggingface_hub import snapshot_download

examples/community/stable_diffusion_xl_reference.py

@@ -771,8 +771,7 @@ class StableDiffusionXLReferencePipeline(StableDiffusionXLPipeline):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
                     if callback is not None and i % callback_steps == 0:
-                        step_idx = i // getattr(self.scheduler, "order", 1)
-                        callback(step_idx, t, latents)
+                        callback(i, t, latents)
 
         if not output_type == "latent":
             # make sure the VAE is in float32 mode, as it overflows in float16

examples/community/text_inpainting.py

@@ -1,6 +1,6 @@
 from typing import Callable, List, Optional, Union
 
-import PIL.Image
+import PIL
 import torch
 from transformers import (
     CLIPImageProcessor,

examples/community/tiled_upscaling.py

@@ -16,7 +16,7 @@ import math
 from typing import Callable, List, Optional, Union
 
 import numpy as np
-import PIL.Image
+import PIL
 import torch
 from PIL import Image
 from transformers import CLIPTextModel, CLIPTokenizer

examples/community/unclip_image_interpolation.py

@@ -1,7 +1,7 @@
 import inspect
 from typing import List, Optional, Union
 
-import PIL.Image
+import PIL
 import torch
 from torch.nn import functional as F
 from transformers import (

examples/community/wildcard_stable_diffusion.py

@@ -389,8 +389,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline):
 
             # call the callback, if provided
             if callback is not None and i % callback_steps == 0:
-                step_idx = i // getattr(self.scheduler, "order", 1)
-                callback(step_idx, t, latents)
+                callback(i, t, latents)
 
         latents = 1 / 0.18215 * latents
         image = self.vae.decode(latents).sample

examples/controlnet/train_controlnet.py

@@ -56,7 +56,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 

examples/controlnet/train_controlnet_flax.py

@@ -59,7 +59,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = logging.getLogger(__name__)
 
@@ -907,9 +907,6 @@ def main():
 
             if args.snr_gamma is not None:
                 snr = jnp.array(compute_snr(timesteps))
-                if noise_scheduler.config.prediction_type == "v_prediction":
-                    # Velocity objective requires that we add one to SNR values before we divide by them.
-                    snr = snr + 1
                 snr_loss_weights = jnp.where(snr < args.snr_gamma, snr, jnp.ones_like(snr) * args.snr_gamma) / snr
                 loss = loss * snr_loss_weights
 

examples/controlnet/train_controlnet_sdxl.py

@@ -58,7 +58,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 

examples/custom_diffusion/train_custom_diffusion.py

@@ -51,18 +51,14 @@ from diffusers import (
     UNet2DConditionModel,
 )
 from diffusers.loaders import AttnProcsLayers
-from diffusers.models.attention_processor import (
-    CustomDiffusionAttnProcessor,
-    CustomDiffusionAttnProcessor2_0,
-    CustomDiffusionXFormersAttnProcessor,
-)
+from diffusers.models.attention_processor import CustomDiffusionAttnProcessor, CustomDiffusionXFormersAttnProcessor
 from diffusers.optimization import get_scheduler
 from diffusers.utils import check_min_version, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 
@@ -207,7 +203,7 @@ class CustomDiffusionDataset(Dataset):
                     with open(concept["class_prompt"], "r") as f:
                         class_prompt = f.read().splitlines()
 
-                class_img_path = list(zip(class_images_path, class_prompt))
+                class_img_path = [(x, y) for (x, y) in zip(class_images_path, class_prompt)]
                 self.class_images_path.extend(class_img_path[:num_class_images])
 
         random.shuffle(self.instance_images_path)
@@ -874,9 +870,7 @@ def main(args):
     unet.to(accelerator.device, dtype=weight_dtype)
     vae.to(accelerator.device, dtype=weight_dtype)
 
-    attention_class = (
-        CustomDiffusionAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else CustomDiffusionAttnProcessor
-    )
+    attention_class = CustomDiffusionAttnProcessor
     if args.enable_xformers_memory_efficient_attention:
         if is_xformers_available():
             import xformers
@@ -1075,30 +1069,30 @@ def main(args):
                 f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
             )
             args.resume_from_checkpoint = None
-            initial_global_step = 0
         else:
             accelerator.print(f"Resuming from checkpoint {path}")
             accelerator.load_state(os.path.join(args.output_dir, path))
             global_step = int(path.split("-")[1])
 
-            initial_global_step = global_step
+            resume_global_step = global_step * args.gradient_accumulation_steps
             first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
 
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
 
     for epoch in range(first_epoch, args.num_train_epochs):
         unet.train()
         if args.modifier_token is not None:
             text_encoder.train()
         for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
             with accelerator.accumulate(unet), accelerator.accumulate(text_encoder):
                 # Convert images to latent space
                 latents = vae.encode(batch["pixel_values"].to(dtype=weight_dtype)).latent_dist.sample()
@@ -1214,52 +1208,50 @@ def main(args):
             if global_step >= args.max_train_steps:
                 break
 
-            if accelerator.is_main_process:
-                images = []
+        if accelerator.is_main_process:
+            images = []
 
-                if args.validation_prompt is not None and global_step % args.validation_steps == 0:
-                    logger.info(
-                        f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
-                        f" {args.validation_prompt}."
-                    )
-                    # create pipeline
-                    pipeline = DiffusionPipeline.from_pretrained(
-                        args.pretrained_model_name_or_path,
-                        unet=accelerator.unwrap_model(unet),
-                        text_encoder=accelerator.unwrap_model(text_encoder),
-                        tokenizer=tokenizer,
-                        revision=args.revision,
-                        torch_dtype=weight_dtype,
-                    )
-                    pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
-                    pipeline = pipeline.to(accelerator.device)
-                    pipeline.set_progress_bar_config(disable=True)
+            if args.validation_prompt is not None and global_step % args.validation_steps == 0:
+                logger.info(
+                    f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
+                    f" {args.validation_prompt}."
+                )
+                # create pipeline
+                pipeline = DiffusionPipeline.from_pretrained(
+                    args.pretrained_model_name_or_path,
+                    unet=accelerator.unwrap_model(unet),
+                    text_encoder=accelerator.unwrap_model(text_encoder),
+                    tokenizer=tokenizer,
+                    revision=args.revision,
+                    torch_dtype=weight_dtype,
+                )
+                pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
+                pipeline = pipeline.to(accelerator.device)
+                pipeline.set_progress_bar_config(disable=True)
 
-                    # run inference
-                    generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
-                    images = [
-                        pipeline(args.validation_prompt, num_inference_steps=25, generator=generator, eta=1.0).images[
-                            0
-                        ]
-                        for _ in range(args.num_validation_images)
-                    ]
+                # run inference
+                generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
+                images = [
+                    pipeline(args.validation_prompt, num_inference_steps=25, generator=generator, eta=1.0).images[0]
+                    for _ in range(args.num_validation_images)
+                ]
 
-                    for tracker in accelerator.trackers:
-                        if tracker.name == "tensorboard":
-                            np_images = np.stack([np.asarray(img) for img in images])
-                            tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
-                        if tracker.name == "wandb":
-                            tracker.log(
-                                {
-                                    "validation": [
-                                        wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
-                                        for i, image in enumerate(images)
-                                    ]
-                                }
-                            )
+                for tracker in accelerator.trackers:
+                    if tracker.name == "tensorboard":
+                        np_images = np.stack([np.asarray(img) for img in images])
+                        tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
+                    if tracker.name == "wandb":
+                        tracker.log(
+                            {
+                                "validation": [
+                                    wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
+                                    for i, image in enumerate(images)
+                                ]
+                            }
+                        )
 
-                    del pipeline
-                    torch.cuda.empty_cache()
+                del pipeline
+                torch.cuda.empty_cache()
 
     # Save the custom diffusion layers
     accelerator.wait_for_everyone()

examples/dreambooth/train_dreambooth.py

@@ -52,7 +52,6 @@ from diffusers import (
     UNet2DConditionModel,
 )
 from diffusers.optimization import get_scheduler
-from diffusers.training_utils import compute_snr
 from diffusers.utils import check_min_version, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
@@ -61,7 +60,7 @@ if is_wandb_available():
     import wandb
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 
@@ -525,13 +524,6 @@ def parse_args(input_args=None):
             " See: https://www.crosslabs.org//blog/diffusion-with-offset-noise for more information."
         ),
     )
-    parser.add_argument(
-        "--snr_gamma",
-        type=float,
-        default=None,
-        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
-        "More details here: https://arxiv.org/abs/2303.09556.",
-    )
     parser.add_argument(
         "--pre_compute_text_embeddings",
         action="store_true",
@@ -1178,30 +1170,30 @@ def main(args):
                 f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
             )
             args.resume_from_checkpoint = None
-            initial_global_step = 0
         else:
             accelerator.print(f"Resuming from checkpoint {path}")
             accelerator.load_state(os.path.join(args.output_dir, path))
             global_step = int(path.split("-")[1])
 
-            initial_global_step = global_step
+            resume_global_step = global_step * args.gradient_accumulation_steps
             first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
 
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
 
     for epoch in range(first_epoch, args.num_train_epochs):
         unet.train()
         if args.train_text_encoder:
             text_encoder.train()
         for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
             with accelerator.accumulate(unet):
                 pixel_values = batch["pixel_values"].to(dtype=weight_dtype)
 
@@ -1269,34 +1261,17 @@ def main(args):
                     # Chunk the noise and model_pred into two parts and compute the loss on each part separately.
                     model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
                     target, target_prior = torch.chunk(target, 2, dim=0)
+
+                    # Compute instance loss
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
                     # Compute prior loss
                     prior_loss = F.mse_loss(model_pred_prior.float(), target_prior.float(), reduction="mean")
 
-                # Compute instance loss
-                if args.snr_gamma is None:
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-                else:
-                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
-                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
-                    # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    base_weight = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )
-
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective needs to be floored to an SNR weight of one.
-                        mse_loss_weights = base_weight + 1
-                    else:
-                        # Epsilon and sample both use the same loss weights.
-                        mse_loss_weights = base_weight
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
-                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
-                    loss = loss.mean()
-
-                if args.with_prior_preservation:
                     # Add the prior loss to the instance loss.
                     loss = loss + args.prior_loss_weight * prior_loss
+                else:
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
 
                 accelerator.backward(loss)
                 if accelerator.sync_gradients:

examples/dreambooth/train_dreambooth_flax.py

@@ -36,7 +36,7 @@ from diffusers.utils import check_min_version
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 # Cache compiled models across invocations of this script.
 cc.initialize_cache(os.path.expanduser("~/.cache/jax/compilation_cache"))

examples/dreambooth/train_dreambooth_lora.py

@@ -24,6 +24,7 @@ import os
 import shutil
 import warnings
 from pathlib import Path
+from typing import Dict
 
 import numpy as np
 import torch
@@ -58,17 +59,18 @@ from diffusers.loaders import (
 from diffusers.models.attention_processor import (
     AttnAddedKVProcessor,
     AttnAddedKVProcessor2_0,
+    LoRAAttnAddedKVProcessor,
+    LoRAAttnProcessor,
+    LoRAAttnProcessor2_0,
     SlicedAttnAddedKVProcessor,
 )
-from diffusers.models.lora import LoRALinearLayer
 from diffusers.optimization import get_scheduler
-from diffusers.training_utils import unet_lora_state_dict
 from diffusers.utils import check_min_version, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 
@@ -656,6 +658,22 @@ def encode_prompt(text_encoder, input_ids, attention_mask, text_encoder_use_atte
     return prompt_embeds
 
 
+def unet_attn_processors_state_dict(unet) -> Dict[str, torch.tensor]:
+    r"""
+    Returns:
+        a state dict containing just the attention processor parameters.
+    """
+    attn_processors = unet.attn_processors
+
+    attn_processors_state_dict = {}
+
+    for attn_processor_key, attn_processor in attn_processors.items():
+        for parameter_key, parameter in attn_processor.state_dict().items():
+            attn_processors_state_dict[f"{attn_processor_key}.{parameter_key}"] = parameter
+
+    return attn_processors_state_dict
+
+
 def main(args):
     logging_dir = Path(args.output_dir, args.logging_dir)
 
@@ -836,64 +854,37 @@ def main(args):
     # For Stable Diffusion, it should be equal to:
     # - down blocks (2x attention layers) * (2x transformer layers) * (3x down blocks) = 12
     # - mid blocks (2x attention layers) * (1x transformer layers) * (1x mid blocks) = 2
-    # - up blocks (2x attention layers) * (3x transformer layers) * (3x up blocks) = 18
+    # - up blocks (2x attention layers) * (3x transformer layers) * (3x down blocks) = 18
     # => 32 layers
 
     # Set correct lora layers
+    unet_lora_attn_procs = {}
     unet_lora_parameters = []
-    for attn_processor_name, attn_processor in unet.attn_processors.items():
-        # Parse the attention module.
-        attn_module = unet
-        for n in attn_processor_name.split(".")[:-1]:
-            attn_module = getattr(attn_module, n)
-
-        # Set the `lora_layer` attribute of the attention-related matrices.
-        attn_module.to_q.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_q.in_features, out_features=attn_module.to_q.out_features, rank=args.rank
-            )
-        )
-        attn_module.to_k.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_k.in_features, out_features=attn_module.to_k.out_features, rank=args.rank
-            )
-        )
-        attn_module.to_v.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_v.in_features, out_features=attn_module.to_v.out_features, rank=args.rank
-            )
-        )
-        attn_module.to_out[0].set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_out[0].in_features,
-                out_features=attn_module.to_out[0].out_features,
-                rank=args.rank,
-            )
-        )
-
-        # Accumulate the LoRA params to optimize.
-        unet_lora_parameters.extend(attn_module.to_q.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_k.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_v.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_out[0].lora_layer.parameters())
+    for name, attn_processor in unet.attn_processors.items():
+        cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+        if name.startswith("mid_block"):
+            hidden_size = unet.config.block_out_channels[-1]
+        elif name.startswith("up_blocks"):
+            block_id = int(name[len("up_blocks.")])
+            hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+        elif name.startswith("down_blocks"):
+            block_id = int(name[len("down_blocks.")])
+            hidden_size = unet.config.block_out_channels[block_id]
 
         if isinstance(attn_processor, (AttnAddedKVProcessor, SlicedAttnAddedKVProcessor, AttnAddedKVProcessor2_0)):
-            attn_module.add_k_proj.set_lora_layer(
-                LoRALinearLayer(
-                    in_features=attn_module.add_k_proj.in_features,
-                    out_features=attn_module.add_k_proj.out_features,
-                    rank=args.rank,
-                )
+            lora_attn_processor_class = LoRAAttnAddedKVProcessor
+        else:
+            lora_attn_processor_class = (
+                LoRAAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else LoRAAttnProcessor
             )
-            attn_module.add_v_proj.set_lora_layer(
-                LoRALinearLayer(
-                    in_features=attn_module.add_v_proj.in_features,
-                    out_features=attn_module.add_v_proj.out_features,
-                    rank=args.rank,
-                )
-            )
-            unet_lora_parameters.extend(attn_module.add_k_proj.lora_layer.parameters())
-            unet_lora_parameters.extend(attn_module.add_v_proj.lora_layer.parameters())
+
+        module = lora_attn_processor_class(
+            hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=args.rank
+        )
+        unet_lora_attn_procs[name] = module
+        unet_lora_parameters.extend(module.parameters())
+
+    unet.set_attn_processor(unet_lora_attn_procs)
 
     # The text encoder comes from 🤗 transformers, so we cannot directly modify it.
     # So, instead, we monkey-patch the forward calls of its attention-blocks.
@@ -911,7 +902,7 @@ def main(args):
 
             for model in models:
                 if isinstance(model, type(accelerator.unwrap_model(unet))):
-                    unet_lora_layers_to_save = unet_lora_state_dict(model)
+                    unet_lora_layers_to_save = unet_attn_processors_state_dict(model)
                 elif isinstance(model, type(accelerator.unwrap_model(text_encoder))):
                     text_encoder_lora_layers_to_save = text_encoder_lora_state_dict(model)
                 else:
@@ -1117,30 +1108,30 @@ def main(args):
                 f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
             )
             args.resume_from_checkpoint = None
-            initial_global_step = 0
         else:
             accelerator.print(f"Resuming from checkpoint {path}")
             accelerator.load_state(os.path.join(args.output_dir, path))
             global_step = int(path.split("-")[1])
 
-            initial_global_step = global_step
+            resume_global_step = global_step * args.gradient_accumulation_steps
             first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
 
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
 
     for epoch in range(first_epoch, args.num_train_epochs):
         unet.train()
         if args.train_text_encoder:
             text_encoder.train()
         for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
             with accelerator.accumulate(unet):
                 pixel_values = batch["pixel_values"].to(dtype=weight_dtype)
 
@@ -1347,7 +1338,7 @@ def main(args):
     if accelerator.is_main_process:
         unet = accelerator.unwrap_model(unet)
         unet = unet.to(torch.float32)
-        unet_lora_layers = unet_lora_state_dict(unet)
+        unet_lora_layers = unet_attn_processors_state_dict(unet)
 
         if text_encoder is not None and args.train_text_encoder:
             text_encoder = accelerator.unwrap_model(text_encoder)

examples/dreambooth/train_dreambooth_lora_sdxl.py

@@ -23,6 +23,7 @@ import os
 import shutil
 import warnings
 from pathlib import Path
+from typing import Dict
 
 import numpy as np
 import torch
@@ -50,15 +51,14 @@ from diffusers import (
     UNet2DConditionModel,
 )
 from diffusers.loaders import LoraLoaderMixin, text_encoder_lora_state_dict
-from diffusers.models.lora import LoRALinearLayer
+from diffusers.models.attention_processor import LoRAAttnProcessor, LoRAAttnProcessor2_0
 from diffusers.optimization import get_scheduler
-from diffusers.training_utils import unet_lora_state_dict
 from diffusers.utils import check_min_version, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__)
 
@@ -575,6 +575,22 @@ def encode_prompt(text_encoders, tokenizers, prompt, text_input_ids_list=None):
     return prompt_embeds, pooled_prompt_embeds
 
 
+def unet_attn_processors_state_dict(unet) -> Dict[str, torch.tensor]:
+    """
+    Returns:
+        a state dict containing just the attention processor parameters.
+    """
+    attn_processors = unet.attn_processors
+
+    attn_processors_state_dict = {}
+
+    for attn_processor_key, attn_processor in attn_processors.items():
+        for parameter_key, parameter in attn_processor.state_dict().items():
+            attn_processors_state_dict[f"{attn_processor_key}.{parameter_key}"] = parameter
+
+    return attn_processors_state_dict
+
+
 def main(args):
     logging_dir = Path(args.output_dir, args.logging_dir)
 
@@ -745,52 +761,29 @@ def main(args):
 
     # now we will add new LoRA weights to the attention layers
     # Set correct lora layers
+    unet_lora_attn_procs = {}
     unet_lora_parameters = []
-    for attn_processor_name, attn_processor in unet.attn_processors.items():
-        # Parse the attention module.
-        attn_module = unet
-        for n in attn_processor_name.split(".")[:-1]:
-            attn_module = getattr(attn_module, n)
+    for name, attn_processor in unet.attn_processors.items():
+        cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+        if name.startswith("mid_block"):
+            hidden_size = unet.config.block_out_channels[-1]
+        elif name.startswith("up_blocks"):
+            block_id = int(name[len("up_blocks.")])
+            hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+        elif name.startswith("down_blocks"):
+            block_id = int(name[len("down_blocks.")])
+            hidden_size = unet.config.block_out_channels[block_id]
 
-        # Set the `lora_layer` attribute of the attention-related matrices.
-        attn_module.to_q.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_q.in_features,
-                out_features=attn_module.to_q.out_features,
-                rank=args.rank,
-                dtype=torch.float32,
-            )
+        lora_attn_processor_class = (
+            LoRAAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else LoRAAttnProcessor
         )
-        attn_module.to_k.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_k.in_features,
-                out_features=attn_module.to_k.out_features,
-                rank=args.rank,
-                dtype=torch.float32,
-            )
-        )
-        attn_module.to_v.set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_v.in_features,
-                out_features=attn_module.to_v.out_features,
-                rank=args.rank,
-                dtype=torch.float32,
-            )
-        )
-        attn_module.to_out[0].set_lora_layer(
-            LoRALinearLayer(
-                in_features=attn_module.to_out[0].in_features,
-                out_features=attn_module.to_out[0].out_features,
-                rank=args.rank,
-                dtype=torch.float32,
-            )
+        module = lora_attn_processor_class(
+            hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=args.rank
         )
+        unet_lora_attn_procs[name] = module
+        unet_lora_parameters.extend(module.parameters())
 
-        # Accumulate the LoRA params to optimize.
-        unet_lora_parameters.extend(attn_module.to_q.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_k.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_v.lora_layer.parameters())
-        unet_lora_parameters.extend(attn_module.to_out[0].lora_layer.parameters())
+    unet.set_attn_processor(unet_lora_attn_procs)
 
     # The text encoder comes from 🤗 transformers, so we cannot directly modify it.
     # So, instead, we monkey-patch the forward calls of its attention-blocks.
@@ -814,7 +807,7 @@ def main(args):
 
             for model in models:
                 if isinstance(model, type(accelerator.unwrap_model(unet))):
-                    unet_lora_layers_to_save = unet_lora_state_dict(model)
+                    unet_lora_layers_to_save = unet_attn_processors_state_dict(model)
                 elif isinstance(model, type(accelerator.unwrap_model(text_encoder_one))):
                     text_encoder_one_lora_layers_to_save = text_encoder_lora_state_dict(model)
                 elif isinstance(model, type(accelerator.unwrap_model(text_encoder_two))):
@@ -1055,25 +1048,18 @@ def main(args):
                 f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
             )
             args.resume_from_checkpoint = None
-            initial_global_step = 0
         else:
             accelerator.print(f"Resuming from checkpoint {path}")
             accelerator.load_state(os.path.join(args.output_dir, path))
             global_step = int(path.split("-")[1])
 
-            initial_global_step = global_step
+            resume_global_step = global_step * args.gradient_accumulation_steps
             first_epoch = global_step // num_update_steps_per_epoch
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
 
-    else:
-        initial_global_step = 0
-
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
 
     for epoch in range(first_epoch, args.num_train_epochs):
         unet.train()
@@ -1081,6 +1067,12 @@ def main(args):
             text_encoder_one.train()
             text_encoder_two.train()
         for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
             with accelerator.accumulate(unet):
                 pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
 
@@ -1281,7 +1273,7 @@ def main(args):
     if accelerator.is_main_process:
         unet = accelerator.unwrap_model(unet)
         unet = unet.to(torch.float32)
-        unet_lora_layers = unet_lora_state_dict(unet)
+        unet_lora_layers = unet_attn_processors_state_dict(unet)
 
         if args.train_text_encoder:
             text_encoder_one = accelerator.unwrap_model(text_encoder_one)

examples/instruct_pix2pix/train_instruct_pix2pix.py

@@ -52,7 +52,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__, log_level="INFO")
 

examples/instruct_pix2pix/train_instruct_pix2pix_sdxl.py

@@ -55,7 +55,7 @@ from diffusers.utils.import_utils import is_xformers_available
 
 
 # Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.22.0.dev0")
+check_min_version("0.21.0")
 
 logger = get_logger(__name__, log_level="INFO")
 
@@ -103,7 +103,7 @@ def parse_args():
     )
     parser.add_argument(
         "--vae_precision",
-        type=str,
+        type="choice",
         choices=["fp32", "fp16", "bf16"],
         default="fp32",
         help=(
@@ -726,9 +726,6 @@ def main():
     text_encoder_1.requires_grad_(False)
     text_encoder_2.requires_grad_(False)
 
-    # Set UNet to trainable.
-    unet.train()
-
     # Adapted from pipelines.StableDiffusionXLPipeline.encode_prompt
     def encode_prompt(text_encoders, tokenizers, prompt):
         prompt_embeds_list = []
@@ -936,28 +933,29 @@ def main():
                 f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
             )
             args.resume_from_checkpoint = None
-            initial_global_step = 0
         else:
             accelerator.print(f"Resuming from checkpoint {path}")
             accelerator.load_state(os.path.join(args.output_dir, path))
             global_step = int(path.split("-")[1])
 
-            initial_global_step = global_step
+            resume_global_step = global_step * args.gradient_accumulation_steps
             first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
+            resume_step = resume_global_step % (num_update_steps_per_epoch * args.gradient_accumulation_steps)
 
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
+    # Only show the progress bar once on each machine.
+    progress_bar = tqdm(range(global_step, args.max_train_steps), disable=not accelerator.is_local_main_process)
+    progress_bar.set_description("Steps")
 
     for epoch in range(first_epoch, args.num_train_epochs):
+        unet.train()
         train_loss = 0.0
         for step, batch in enumerate(train_dataloader):
+            # Skip steps until we reach the resumed step
+            if args.resume_from_checkpoint and epoch == first_epoch and step < resume_step:
+                if step % args.gradient_accumulation_steps == 0:
+                    progress_bar.update(1)
+                continue
+
             with accelerator.accumulate(unet):
                 # We want to learn the denoising process w.r.t the edited images which
                 # are conditioned on the original image (which was edited) and the edit instruction.

examples/kandinsky2_2/text_to_image/README.md

@@ -1,317 +0,0 @@
-# Kandinsky2.2 text-to-image fine-tuning
-
-Kandinsky 2.2 includes a prior pipeline that generates image embeddings from text prompts, and a decoder pipeline that generates the output image based on the image embeddings. We provide `train_text_to_image_prior.py` and `train_text_to_image_decoder.py` scripts to show you how to fine-tune the Kandinsky prior and decoder models separately based on your own dataset. To achieve the best results, you should fine-tune **_both_** your prior and decoder models.
-
-___Note___:
-
-___This script is experimental. The script fine-tunes the whole model and often times the model overfits and runs into issues like catastrophic forgetting. It's recommended to try different hyperparameters to get the best result on your dataset.___
-
-
-## Running locally with PyTorch
-
-Before running the scripts, make sure to install the library's training dependencies:
-
-**Important**
-
-To make sure you can successfully run the latest versions of the example scripts, we highly recommend **installing from source** and keeping the install up to date as we update the example scripts frequently and install some example-specific requirements. To do this, execute the following steps in a new virtual environment:
-```bash
-git clone https://github.com/huggingface/diffusers
-cd diffusers
-pip install .
-```
-
-Then cd in the example folder  and run
-```bash
-pip install -r requirements.txt
-```
-
-And initialize an [🤗Accelerate](https://github.com/huggingface/accelerate/) environment with:
-
-```bash
-accelerate config
-```
-For this example we want to directly store the trained LoRA embeddings on the Hub, so we need to be logged in and add the --push_to_hub flag.
-
-___
-
-### Pokemon example
-
-For all our examples, we will directly store the trained weights on the Hub, so we need to be logged in and add the `--push_to_hub` flag. In order to do that, you have to be a registered user on the 🤗 Hugging Face Hub, and you'll also need to use an access token for the code to work. For more information on access tokens, please refer to the [User Access Tokens](https://huggingface.co/docs/hub/security-tokens) guide.
-
-Run the following command to authenticate your token
-
-```bash
-huggingface-cli login
-```
-
-We also use [Weights and Biases](https://docs.wandb.ai/quickstart) logging by default, because it is really useful to monitor the training progress by regularly generating sample images during training. To install wandb, run 
-
-```bash
-pip install wandb
-```
-
-To disable wandb logging, remove the `--report_to=="wandb"` and `--validation_prompts="A robot pokemon, 4k photo"` flags from below examples
-
-#### Fine-tune decoder
-<br>
-
-<!-- accelerate_snippet_start -->
-```bash
-export DATASET_NAME="lambdalabs/pokemon-blip-captions"
-
-accelerate launch --mixed_precision="fp16"  train_text_to_image_decoder.py \
-  --dataset_name=$DATASET_NAME \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --gradient_checkpointing \
-  --max_train_steps=15000 \
-  --learning_rate=1e-05 \
-  --max_grad_norm=1 \
-  --checkpoints_total_limit=3 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --validation_prompts="A robot pokemon, 4k photo" \
-  --report_to="wandb" \
-  --push_to_hub \
-  --output_dir="kandi2-decoder-pokemon-model" 
-```
-<!-- accelerate_snippet_end -->
-
-
-To train on your own training files, prepare the dataset according to the format required by `datasets`. You can find the instructions for how to do that in the [ImageFolder with metadata](https://huggingface.co/docs/datasets/en/image_load#imagefolder-with-metadata) guide.
-If you wish to use custom loading logic, you should modify the script and we have left pointers for that in the training script.
-
-```bash
-export TRAIN_DIR="path_to_your_dataset"
-
-accelerate launch --mixed_precision="fp16" train_text_to_image_decoder.py \
-  --train_data_dir=$TRAIN_DIR \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --gradient_checkpointing \
-  --max_train_steps=15000 \
-  --learning_rate=1e-05 \
-  --max_grad_norm=1 \
-  --checkpoints_total_limit=3 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --validation_prompts="A robot pokemon, 4k photo" \
-  --report_to="wandb" \
-  --push_to_hub \
-  --output_dir="kandi22-decoder-pokemon-model" 
-```
-
-
-Once the training is finished the model will be saved in the `output_dir` specified in the command. In this example it's `kandi22-decoder-pokemon-model`. To load the fine-tuned model for inference just pass that path to `AutoPipelineForText2Image`
-
-```python
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipe = AutoPipelineForText2Image.from_pretrained(output_dir, torch_dtype=torch.float16)
-pipe.enable_model_cpu_offload()
-
-prompt='A robot pokemon, 4k photo'
-images = pipe(prompt=prompt).images
-images[0].save("robot-pokemon.png")
-```
-
-Checkpoints only save the unet, so to run inference from a checkpoint, just load the unet
-```python
-from diffusers import AutoPipelineForText2Image, UNet2DConditionModel
-
-model_path = "path_to_saved_model"
-
-unet = UNet2DConditionModel.from_pretrained(model_path + "/checkpoint-<N>/unet")
-
-pipe = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", unet=unet, torch_dtype=torch.float16)
-pipe.enable_model_cpu_offload()
-
-image = pipe(prompt="A robot pokemon, 4k photo").images[0]
-image.save("robot-pokemon.png")
-```
-
-#### Fine-tune prior 
-
-You can fine-tune the Kandinsky prior model with `train_text_to_image_prior.py` script. Note that we currently do not support `--gradient_checkpointing` for prior model fine-tuning.
-
-<br>
-
-<!-- accelerate_snippet_start -->
-```bash
-export DATASET_NAME="lambdalabs/pokemon-blip-captions"
-
-accelerate launch --mixed_precision="fp16"  train_text_to_image_prior.py \
-  --dataset_name=$DATASET_NAME \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --max_train_steps=15000 \
-  --learning_rate=1e-05 \
-  --max_grad_norm=1 \
-  --checkpoints_total_limit=3 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --validation_prompts="A robot pokemon, 4k photo" \
-  --report_to="wandb" \
-  --push_to_hub \
-  --output_dir="kandi2-prior-pokemon-model" 
-```
-<!-- accelerate_snippet_end -->
-
-
-To perform inference with the fine-tuned prior model, you will need to first create a prior pipeline by passing the `output_dir` to `DiffusionPipeline`. Then create a `KandinskyV22CombinedPipeline` from a pretrained or fine-tuned decoder checkpoint along with all the modules of the prior pipeline you just created. 
-
-```python
-from diffusers import AutoPipelineForText2Image, DiffusionPipeline
-import torch
-
-pipe_prior = DiffusionPipeline.from_pretrained(output_dir, torch_dtype=torch.float16)
-prior_components = {"prior_" + k: v for k,v in pipe_prior.components.items()}
-pipe = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", **prior_components, torch_dtype=torch.float16)
-
-pipe.enable_model_cpu_offload()
-prompt='A robot pokemon, 4k photo'
-images = pipe(prompt=prompt, negative_prompt=negative_prompt).images
-images[0]
-```
-
-If you want to use a fine-tuned decoder checkpoint along with your fine-tuned prior checkpoint, you can simply replace the "kandinsky-community/kandinsky-2-2-decoder" in above code with your custom model repo name. Note that in order to be able to create a `KandinskyV22CombinedPipeline`, your model repository need to have a prior tag. If you have created your model repo using our training script, the prior tag is automatically included. 
-
-#### Training with multiple GPUs
-
-`accelerate` allows for seamless multi-GPU training. Follow the instructions [here](https://huggingface.co/docs/accelerate/basic_tutorials/launch)
-for running distributed training with `accelerate`. Here is an example command:
-
-```bash
-export DATASET_NAME="lambdalabs/pokemon-blip-captions"
-
-accelerate launch --mixed_precision="fp16" --multi_gpu  train_text_to_image_decoder.py \
-  --dataset_name=$DATASET_NAME \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --gradient_accumulation_steps=4 \
-  --gradient_checkpointing \
-  --max_train_steps=15000 \
-  --learning_rate=1e-05 \
-  --max_grad_norm=1 \
-  --checkpoints_total_limit=3 \
-  --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --validation_prompts="A robot pokemon, 4k photo" \
-  --report_to="wandb" \
-  --push_to_hub \
-  --output_dir="kandi2-decoder-pokemon-model" 
-```
-
-
-#### Training with Min-SNR weighting
-
-We support training with the Min-SNR weighting strategy proposed in [Efficient Diffusion Training via Min-SNR Weighting Strategy](https://arxiv.org/abs/2303.09556) which helps achieve faster convergence
-by rebalancing the loss. Enable the `--snr_gamma` argument and set it to the recommended
-value of 5.0.
-
-
-## Training with LoRA
-
-Low-Rank Adaption of Large Language Models was first introduced by Microsoft in [LoRA: Low-Rank Adaptation of Large Language Models](https://arxiv.org/abs/2106.09685) by *Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen*.
-
-In a nutshell, LoRA allows adapting pretrained models by adding pairs of rank-decomposition matrices to existing weights and **only** training those newly added weights. This has a couple of advantages:
-
-- Previous pretrained weights are kept frozen so that model is not prone to [catastrophic forgetting](https://www.pnas.org/doi/10.1073/pnas.1611835114).
-- Rank-decomposition matrices have significantly fewer parameters than original model, which means that trained LoRA weights are easily portable.
-- LoRA attention layers allow to control to which extent the model is adapted toward new training images via a `scale` parameter.
-
-[cloneofsimo](https://github.com/cloneofsimo) was the first to try out LoRA training for Stable Diffusion in the popular [lora](https://github.com/cloneofsimo/lora) GitHub repository.
-
-With LoRA, it's possible to fine-tune Kandinsky 2.2 on a custom image-caption pair dataset
-on consumer GPUs like Tesla T4, Tesla V100.
-
-### Training
-
-First, you need to set up your development environment as explained in the [installation](#installing-the-dependencies). Make sure to set the `MODEL_NAME` and `DATASET_NAME` environment variables. Here, we will use [Kandinsky 2.2](https://huggingface.co/kandinsky-community/kandinsky-2-2-decoder) and the [Pokemons dataset](https://huggingface.co/datasets/lambdalabs/pokemon-blip-captions).  
-
-
-#### Train decoder 
-
-```bash
-export DATASET_NAME="lambdalabs/pokemon-blip-captions"
-
-accelerate launch --mixed_precision="fp16" train_text_to_image_decoder_lora.py \
-  --dataset_name=$DATASET_NAME --caption_column="text" \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --num_train_epochs=100 --checkpointing_steps=5000 \
-  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --seed=42 \
-  --rank=4 \
-  --gradient_checkpointing \
-  --output_dir="kandi22-decoder-pokemon-lora" \
-  --validation_prompt="cute dragon creature" --report_to="wandb" \
-  --push_to_hub \
-```
-
-#### Train prior
-
-```bash
-export DATASET_NAME="lambdalabs/pokemon-blip-captions"
-
-accelerate launch --mixed_precision="fp16" train_text_to_image_prior_lora.py \
-  --dataset_name=$DATASET_NAME --caption_column="text" \
-  --resolution=768 \
-  --train_batch_size=1 \
-  --num_train_epochs=100 --checkpointing_steps=5000 \
-  --learning_rate=1e-04 --lr_scheduler="constant" --lr_warmup_steps=0 \
-  --seed=42 \
-  --rank=4 \
-  --output_dir="kandi22-prior-pokemon-lora" \
-  --validation_prompt="cute dragon creature" --report_to="wandb" \
-  --push_to_hub \
-```
-
-**___Note: When using LoRA we can use a much higher learning rate compared to non-LoRA fine-tuning. Here we use *1e-4* instead of the usual *1e-5*. Also, by using LoRA, it's possible to run above scripts in consumer GPUs like T4 or V100.___**
-
-
-### Inference
-
-#### Inference using fine-tuned LoRA checkpoint for decoder
-
-Once you have trained a Kandinsky decoder model using the above command, inference can be done with the `AutoPipelineForText2Image` after loading the trained LoRA weights.  You need to pass the `output_dir` for loading the LoRA weights, which in this case is `kandi22-decoder-pokemon-lora`.
-
-
-```python
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipe = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16)
-pipe.unet.load_attn_procs(output_dir)
-pipe.enable_model_cpu_offload()
-
-prompt='A robot pokemon, 4k photo'
-image = pipe(prompt=prompt).images[0]
-image.save("robot_pokemon.png")
-```
-
-#### Inference using fine-tuned LoRA checkpoint for prior
-
-```python
-from diffusers import AutoPipelineForText2Image
-import torch
-
-pipe = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16)
-pipe.prior_prior.load_attn_procs(output_dir)
-pipe.enable_model_cpu_offload()
-
-prompt='A robot pokemon, 4k photo'
-image = pipe(prompt=prompt).images[0]
-image.save("robot_pokemon.png")
-image
-```
-
-### Training with xFormers:
-
-You can enable memory efficient attention by [installing xFormers](https://huggingface.co/docs/diffusers/main/en/optimization/xformers) and passing the `--enable_xformers_memory_efficient_attention` argument to the script.
-
-xFormers training is not available for fine-tuning the prior model.
-
-**Note**:
-
-According to [this issue](https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212), xFormers `v0.0.16` cannot be used for training in some GPUs. If you observe that problem, please install a development version as indicated in that comment.

examples/kandinsky2_2/text_to_image/requirements.txt

@@ -1,7 +0,0 @@
-accelerate>=0.16.0
-torchvision
-transformers>=4.25.1
-datasets
-ftfy
-tensorboard
-Jinja2

examples/kandinsky2_2/text_to_image/train_text_to_image_decoder.py

@@ -1,917 +0,0 @@
-#!/usr/bin/env python
-# coding=utf-8
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-
-import argparse
-import logging
-import math
-import os
-import shutil
-from pathlib import Path
-
-import accelerate
-import datasets
-import numpy as np
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-import transformers
-from accelerate import Accelerator
-from accelerate.logging import get_logger
-from accelerate.state import AcceleratorState
-from accelerate.utils import ProjectConfiguration, set_seed
-from datasets import load_dataset
-from huggingface_hub import create_repo, upload_folder
-from packaging import version
-from PIL import Image
-from tqdm import tqdm
-from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
-from transformers.utils import ContextManagers
-
-import diffusers
-from diffusers import AutoPipelineForText2Image, DDPMScheduler, UNet2DConditionModel, VQModel
-from diffusers.optimization import get_scheduler
-from diffusers.training_utils import EMAModel, compute_snr
-from diffusers.utils import check_min_version, is_wandb_available, make_image_grid
-from diffusers.utils.import_utils import is_xformers_available
-
-
-if is_wandb_available():
-    import wandb
-
-
-# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.21.0.dev0")
-
-logger = get_logger(__name__, log_level="INFO")
-
-
-def save_model_card(
-    args,
-    repo_id: str,
-    images=None,
-    repo_folder=None,
-):
-    img_str = ""
-    if len(images) > 0:
-        image_grid = make_image_grid(images, 1, len(args.validation_prompts))
-        image_grid.save(os.path.join(repo_folder, "val_imgs_grid.png"))
-        img_str += "![val_imgs_grid](./val_imgs_grid.png)\n"
-
-    yaml = f"""
----
-license: creativeml-openrail-m
-base_model: {args.pretrained_decoder_model_name_or_path}
-datasets:
-- {args.dataset_name}
-prior:
-- {args.pretrained_prior_model_name_or_path}
-tags:
-- kandinsky
-- text-to-image
-- diffusers
-inference: true
----
-    """
-    model_card = f"""
-# Finetuning - {repo_id}
-
-This pipeline was finetuned from **{args.pretrained_decoder_model_name_or_path}** on the **{args.dataset_name}** dataset. Below are some example images generated with the finetuned pipeline using the following prompts: {args.validation_prompts}: \n
-{img_str}
-
-## Pipeline usage
-
-You can use the pipeline like so:
-
-```python
-from diffusers import DiffusionPipeline
-import torch
-
-pipeline = AutoPipelineForText2Image.from_pretrained("{repo_id}", torch_dtype=torch.float16)
-prompt = "{args.validation_prompts[0]}"
-image = pipeline(prompt).images[0]
-image.save("my_image.png")
-```
-
-## Training info
-
-These are the key hyperparameters used during training:
-
-* Epochs: {args.num_train_epochs}
-* Learning rate: {args.learning_rate}
-* Batch size: {args.train_batch_size}
-* Gradient accumulation steps: {args.gradient_accumulation_steps}
-* Image resolution: {args.resolution}
-* Mixed-precision: {args.mixed_precision}
-
-"""
-    wandb_info = ""
-    if is_wandb_available():
-        wandb_run_url = None
-        if wandb.run is not None:
-            wandb_run_url = wandb.run.url
-
-    if wandb_run_url is not None:
-        wandb_info = f"""
-More information on all the CLI arguments and the environment are available on your [`wandb` run page]({wandb_run_url}).
-"""
-
-    model_card += wandb_info
-
-    with open(os.path.join(repo_folder, "README.md"), "w") as f:
-        f.write(yaml + model_card)
-
-
-def log_validation(vae, image_encoder, image_processor, unet, args, accelerator, weight_dtype, epoch):
-    logger.info("Running validation... ")
-
-    pipeline = AutoPipelineForText2Image.from_pretrained(
-        args.pretrained_decoder_model_name_or_path,
-        vae=accelerator.unwrap_model(vae),
-        prior_image_encoder=accelerator.unwrap_model(image_encoder),
-        prior_image_processor=image_processor,
-        unet=accelerator.unwrap_model(unet),
-        torch_dtype=weight_dtype,
-    )
-    pipeline = pipeline.to(accelerator.device)
-    pipeline.set_progress_bar_config(disable=True)
-
-    if args.enable_xformers_memory_efficient_attention:
-        pipeline.enable_xformers_memory_efficient_attention()
-
-    if args.seed is None:
-        generator = None
-    else:
-        generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
-
-    images = []
-    for i in range(len(args.validation_prompts)):
-        with torch.autocast("cuda"):
-            image = pipeline(args.validation_prompts[i], num_inference_steps=20, generator=generator).images[0]
-
-        images.append(image)
-
-    for tracker in accelerator.trackers:
-        if tracker.name == "tensorboard":
-            np_images = np.stack([np.asarray(img) for img in images])
-            tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
-        elif tracker.name == "wandb":
-            tracker.log(
-                {
-                    "validation": [
-                        wandb.Image(image, caption=f"{i}: {args.validation_prompts[i]}")
-                        for i, image in enumerate(images)
-                    ]
-                }
-            )
-        else:
-            logger.warn(f"image logging not implemented for {tracker.name}")
-
-    del pipeline
-    torch.cuda.empty_cache()
-
-    return images
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(description="Simple example of finetuning Kandinsky 2.2.")
-    parser.add_argument(
-        "--pretrained_decoder_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-decoder",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--pretrained_prior_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-prior",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--dataset_name",
-        type=str,
-        default=None,
-        help=(
-            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
-            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
-            " or to a folder containing files that 🤗 Datasets can understand."
-        ),
-    )
-    parser.add_argument(
-        "--dataset_config_name",
-        type=str,
-        default=None,
-        help="The config of the Dataset, leave as None if there's only one config.",
-    )
-    parser.add_argument(
-        "--train_data_dir",
-        type=str,
-        default=None,
-        help=(
-            "A folder containing the training data. Folder contents must follow the structure described in"
-            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
-            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
-        ),
-    )
-    parser.add_argument(
-        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
-    )
-    parser.add_argument(
-        "--max_train_samples",
-        type=int,
-        default=None,
-        help=(
-            "For debugging purposes or quicker training, truncate the number of training examples to this "
-            "value if set."
-        ),
-    )
-    parser.add_argument(
-        "--validation_prompts",
-        type=str,
-        default=None,
-        nargs="+",
-        help=("A set of prompts evaluated every `--validation_epochs` and logged to `--report_to`."),
-    )
-    parser.add_argument(
-        "--output_dir",
-        type=str,
-        default="kandi_2_2-model-finetuned",
-        help="The output directory where the model predictions and checkpoints will be written.",
-    )
-    parser.add_argument(
-        "--cache_dir",
-        type=str,
-        default=None,
-        help="The directory where the downloaded models and datasets will be stored.",
-    )
-    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
-    parser.add_argument(
-        "--resolution",
-        type=int,
-        default=512,
-        help=(
-            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
-            " resolution"
-        ),
-    )
-    parser.add_argument(
-        "--train_batch_size", type=int, default=1, help="Batch size (per device) for the training dataloader."
-    )
-    parser.add_argument("--num_train_epochs", type=int, default=100)
-    parser.add_argument(
-        "--max_train_steps",
-        type=int,
-        default=None,
-        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
-    )
-    parser.add_argument(
-        "--gradient_accumulation_steps",
-        type=int,
-        default=1,
-        help="Number of updates steps to accumulate before performing a backward/update pass.",
-    )
-    parser.add_argument(
-        "--gradient_checkpointing",
-        action="store_true",
-        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
-    )
-    parser.add_argument(
-        "--learning_rate",
-        type=float,
-        default=1e-4,
-        help="learning rate",
-    )
-    parser.add_argument(
-        "--lr_scheduler",
-        type=str,
-        default="constant",
-        help=(
-            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
-            ' "constant", "constant_with_warmup"]'
-        ),
-    )
-    parser.add_argument(
-        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
-    )
-    parser.add_argument(
-        "--snr_gamma",
-        type=float,
-        default=None,
-        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
-        "More details here: https://arxiv.org/abs/2303.09556.",
-    )
-    parser.add_argument(
-        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
-    )
-    parser.add_argument(
-        "--allow_tf32",
-        action="store_true",
-        help=(
-            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
-            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
-        ),
-    )
-    parser.add_argument("--use_ema", action="store_true", help="Whether to use EMA model.")
-    parser.add_argument(
-        "--dataloader_num_workers",
-        type=int,
-        default=0,
-        help=(
-            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
-        ),
-    )
-    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
-    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
-    parser.add_argument(
-        "--adam_weight_decay",
-        type=float,
-        default=0.0,
-        required=False,
-        help="weight decay_to_use",
-    )
-    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
-    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
-    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
-    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
-    parser.add_argument(
-        "--hub_model_id",
-        type=str,
-        default=None,
-        help="The name of the repository to keep in sync with the local `output_dir`.",
-    )
-    parser.add_argument(
-        "--logging_dir",
-        type=str,
-        default="logs",
-        help=(
-            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
-            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
-        ),
-    )
-    parser.add_argument(
-        "--mixed_precision",
-        type=str,
-        default=None,
-        choices=["no", "fp16", "bf16"],
-        help=(
-            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
-            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
-            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
-        ),
-    )
-    parser.add_argument(
-        "--report_to",
-        type=str,
-        default="tensorboard",
-        help=(
-            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
-            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
-        ),
-    )
-    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
-    parser.add_argument(
-        "--checkpointing_steps",
-        type=int,
-        default=500,
-        help=(
-            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
-            " training using `--resume_from_checkpoint`."
-        ),
-    )
-    parser.add_argument(
-        "--checkpoints_total_limit",
-        type=int,
-        default=None,
-        help=("Max number of checkpoints to store."),
-    )
-    parser.add_argument(
-        "--resume_from_checkpoint",
-        type=str,
-        default=None,
-        help=(
-            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
-            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
-        ),
-    )
-    parser.add_argument(
-        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
-    )
-    parser.add_argument(
-        "--validation_epochs",
-        type=int,
-        default=5,
-        help="Run validation every X epochs.",
-    )
-    parser.add_argument(
-        "--tracker_project_name",
-        type=str,
-        default="text2image-fine-tune",
-        help=(
-            "The `project_name` argument passed to Accelerator.init_trackers for"
-            " more information see https://huggingface.co/docs/accelerate/v0.17.0/en/package_reference/accelerator#accelerate.Accelerator"
-        ),
-    )
-
-    args = parser.parse_args()
-    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
-    if env_local_rank != -1 and env_local_rank != args.local_rank:
-        args.local_rank = env_local_rank
-
-    # Sanity checks
-    if args.dataset_name is None and args.train_data_dir is None:
-        raise ValueError("Need either a dataset name or a training folder.")
-
-    return args
-
-
-def main():
-    args = parse_args()
-    logging_dir = os.path.join(args.output_dir, args.logging_dir)
-    accelerator_project_config = ProjectConfiguration(
-        total_limit=args.checkpoints_total_limit, project_dir=args.output_dir, logging_dir=logging_dir
-    )
-    accelerator = Accelerator(
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        mixed_precision=args.mixed_precision,
-        log_with=args.report_to,
-        project_config=accelerator_project_config,
-    )
-
-    # Make one log on every process with the configuration for debugging.
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO,
-    )
-    logger.info(accelerator.state, main_process_only=False)
-    if accelerator.is_local_main_process:
-        datasets.utils.logging.set_verbosity_warning()
-        transformers.utils.logging.set_verbosity_warning()
-        diffusers.utils.logging.set_verbosity_info()
-    else:
-        datasets.utils.logging.set_verbosity_error()
-        transformers.utils.logging.set_verbosity_error()
-        diffusers.utils.logging.set_verbosity_error()
-
-    # If passed along, set the training seed now.
-    if args.seed is not None:
-        set_seed(args.seed)
-
-    # Handle the repository creation
-    if accelerator.is_main_process:
-        if args.output_dir is not None:
-            os.makedirs(args.output_dir, exist_ok=True)
-
-        if args.push_to_hub:
-            repo_id = create_repo(
-                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
-            ).repo_id
-    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="scheduler")
-    image_processor = CLIPImageProcessor.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_processor"
-    )
-
-    def deepspeed_zero_init_disabled_context_manager():
-        """
-        returns either a context list that includes one that will disable zero.Init or an empty context list
-        """
-        deepspeed_plugin = AcceleratorState().deepspeed_plugin if accelerate.state.is_initialized() else None
-        if deepspeed_plugin is None:
-            return []
-
-        return [deepspeed_plugin.zero3_init_context_manager(enable=False)]
-
-    weight_dtype = torch.float32
-    if accelerator.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-    with ContextManagers(deepspeed_zero_init_disabled_context_manager()):
-        vae = VQModel.from_pretrained(
-            args.pretrained_decoder_model_name_or_path, subfolder="movq", torch_dtype=weight_dtype
-        ).eval()
-        image_encoder = CLIPVisionModelWithProjection.from_pretrained(
-            args.pretrained_prior_model_name_or_path, subfolder="image_encoder", torch_dtype=weight_dtype
-        ).eval()
-    unet = UNet2DConditionModel.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="unet")
-
-    # Freeze vae and image_encoder
-    vae.requires_grad_(False)
-    image_encoder.requires_grad_(False)
-
-    # Set unet to trainable.
-    unet.train()
-
-    # Create EMA for the unet.
-    if args.use_ema:
-        ema_unet = UNet2DConditionModel.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="unet")
-        ema_unet = EMAModel(ema_unet.parameters(), model_cls=UNet2DConditionModel, model_config=ema_unet.config)
-        ema_unet.to(accelerator.device)
-    if args.enable_xformers_memory_efficient_attention:
-        if is_xformers_available():
-            import xformers
-
-            xformers_version = version.parse(xformers.__version__)
-            if xformers_version == version.parse("0.0.16"):
-                logger.warn(
-                    "xFormers 0.0.16 cannot be used for training in some GPUs. If you observe problems during training, please update xFormers to at least 0.0.17. See https://huggingface.co/docs/diffusers/main/en/optimization/xformers for more details."
-                )
-            unet.enable_xformers_memory_efficient_attention()
-        else:
-            raise ValueError("xformers is not available. Make sure it is installed correctly")
-
-    # `accelerate` 0.16.0 will have better support for customized saving
-    if version.parse(accelerate.__version__) >= version.parse("0.16.0"):
-        # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
-        def save_model_hook(models, weights, output_dir):
-            if args.use_ema:
-                ema_unet.save_pretrained(os.path.join(output_dir, "unet_ema"))
-
-            for i, model in enumerate(models):
-                model.save_pretrained(os.path.join(output_dir, "unet"))
-
-                # make sure to pop weight so that corresponding model is not saved again
-                weights.pop()
-
-        def load_model_hook(models, input_dir):
-            if args.use_ema:
-                load_model = EMAModel.from_pretrained(os.path.join(input_dir, "unet_ema"), UNet2DConditionModel)
-                ema_unet.load_state_dict(load_model.state_dict())
-                ema_unet.to(accelerator.device)
-                del load_model
-
-            for i in range(len(models)):
-                # pop models so that they are not loaded again
-                model = models.pop()
-
-                # load diffusers style into model
-                load_model = UNet2DConditionModel.from_pretrained(input_dir, subfolder="unet")
-                model.register_to_config(**load_model.config)
-
-                model.load_state_dict(load_model.state_dict())
-                del load_model
-
-        accelerator.register_save_state_pre_hook(save_model_hook)
-        accelerator.register_load_state_pre_hook(load_model_hook)
-
-    if args.gradient_checkpointing:
-        unet.enable_gradient_checkpointing()
-
-    # Enable TF32 for faster training on Ampere GPUs,
-    # cf https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices
-    if args.allow_tf32:
-        torch.backends.cuda.matmul.allow_tf32 = True
-
-    if args.use_8bit_adam:
-        try:
-            import bitsandbytes as bnb
-        except ImportError:
-            raise ImportError(
-                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
-            )
-
-        optimizer_cls = bnb.optim.AdamW8bit
-    else:
-        optimizer_cls = torch.optim.AdamW
-
-    optimizer = optimizer_cls(
-        unet.parameters(),
-        lr=args.learning_rate,
-        betas=(args.adam_beta1, args.adam_beta2),
-        weight_decay=args.adam_weight_decay,
-        eps=args.adam_epsilon,
-    )
-
-    # Get the datasets: you can either provide your own training and evaluation files (see below)
-    # or specify a Dataset from the hub (the dataset will be downloaded automatically from the datasets Hub).
-
-    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
-    # download the dataset.
-    if args.dataset_name is not None:
-        # Downloading and loading a dataset from the hub.
-        dataset = load_dataset(
-            args.dataset_name,
-            args.dataset_config_name,
-            cache_dir=args.cache_dir,
-        )
-    else:
-        data_files = {}
-        if args.train_data_dir is not None:
-            data_files["train"] = os.path.join(args.train_data_dir, "**")
-        dataset = load_dataset(
-            "imagefolder",
-            data_files=data_files,
-            cache_dir=args.cache_dir,
-        )
-        # See more about loading custom images at
-        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
-
-    # Preprocessing the datasets.
-    # We need to tokenize inputs and targets.
-    column_names = dataset["train"].column_names
-
-    image_column = args.image_column
-    if image_column not in column_names:
-        raise ValueError(f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}")
-
-    def center_crop(image):
-        width, height = image.size
-        new_size = min(width, height)
-        left = (width - new_size) / 2
-        top = (height - new_size) / 2
-        right = (width + new_size) / 2
-        bottom = (height + new_size) / 2
-        return image.crop((left, top, right, bottom))
-
-    def train_transforms(img):
-        img = center_crop(img)
-        img = img.resize((args.resolution, args.resolution), resample=Image.BICUBIC, reducing_gap=1)
-        img = np.array(img).astype(np.float32) / 127.5 - 1
-        img = torch.from_numpy(np.transpose(img, [2, 0, 1]))
-        return img
-
-    def preprocess_train(examples):
-        images = [image.convert("RGB") for image in examples[image_column]]
-        examples["pixel_values"] = [train_transforms(image) for image in images]
-        examples["clip_pixel_values"] = image_processor(images, return_tensors="pt").pixel_values
-        return examples
-
-    with accelerator.main_process_first():
-        if args.max_train_samples is not None:
-            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
-        # Set the training transforms
-        train_dataset = dataset["train"].with_transform(preprocess_train)
-
-    def collate_fn(examples):
-        pixel_values = torch.stack([example["pixel_values"] for example in examples])
-        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
-        clip_pixel_values = torch.stack([example["clip_pixel_values"] for example in examples])
-        clip_pixel_values = clip_pixel_values.to(memory_format=torch.contiguous_format).float()
-        return {"pixel_values": pixel_values, "clip_pixel_values": clip_pixel_values}
-
-    train_dataloader = torch.utils.data.DataLoader(
-        train_dataset,
-        shuffle=True,
-        collate_fn=collate_fn,
-        batch_size=args.train_batch_size,
-        num_workers=args.dataloader_num_workers,
-    )
-
-    # Scheduler and math around the number of training steps.
-    overrode_max_train_steps = False
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if args.max_train_steps is None:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        overrode_max_train_steps = True
-
-    lr_scheduler = get_scheduler(
-        args.lr_scheduler,
-        optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-    )
-    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        unet, optimizer, train_dataloader, lr_scheduler
-    )
-    # Move image_encode and vae to gpu and cast to weight_dtype
-    image_encoder.to(accelerator.device, dtype=weight_dtype)
-    vae.to(accelerator.device, dtype=weight_dtype)
-    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if overrode_max_train_steps:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-    # Afterwards we recalculate our number of training epochs
-    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-
-    # We need to initialize the trackers we use, and also store our configuration.
-    # The trackers initializes automatically on the main process.
-    if accelerator.is_main_process:
-        tracker_config = dict(vars(args))
-        tracker_config.pop("validation_prompts")
-        accelerator.init_trackers(args.tracker_project_name, tracker_config)
-
-    # Train!
-    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-
-    logger.info("***** Running training *****")
-    logger.info(f"  Num examples = {len(train_dataset)}")
-    logger.info(f"  Num Epochs = {args.num_train_epochs}")
-    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
-    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
-    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
-    logger.info(f"  Total optimization steps = {args.max_train_steps}")
-    global_step = 0
-    first_epoch = 0
-    if args.resume_from_checkpoint:
-        if args.resume_from_checkpoint != "latest":
-            path = os.path.basename(args.resume_from_checkpoint)
-        else:
-            # Get the most recent checkpoint
-            dirs = os.listdir(args.output_dir)
-            dirs = [d for d in dirs if d.startswith("checkpoint")]
-            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
-            path = dirs[-1] if len(dirs) > 0 else None
-
-        if path is None:
-            accelerator.print(
-                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
-            )
-            args.resume_from_checkpoint = None
-            initial_global_step = 0
-        else:
-            accelerator.print(f"Resuming from checkpoint {path}")
-            accelerator.load_state(os.path.join(args.output_dir, path))
-            global_step = int(path.split("-")[1])
-
-            initial_global_step = global_step
-            first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
-
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
-
-    for epoch in range(first_epoch, args.num_train_epochs):
-        train_loss = 0.0
-        for step, batch in enumerate(train_dataloader):
-            with accelerator.accumulate(unet):
-                # Convert images to latent space
-                images = batch["pixel_values"].to(weight_dtype)
-                clip_images = batch["clip_pixel_values"].to(weight_dtype)
-                latents = vae.encode(images).latents
-                image_embeds = image_encoder(clip_images).image_embeds
-                # Sample noise that we'll add to the latents
-                noise = torch.randn_like(latents)
-                bsz = latents.shape[0]
-                # Sample a random timestep for each image
-                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (bsz,), device=latents.device)
-                timesteps = timesteps.long()
-
-                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
-
-                target = noise
-
-                # Predict the noise residual and compute loss
-                added_cond_kwargs = {"image_embeds": image_embeds}
-
-                model_pred = unet(noisy_latents, timesteps, None, added_cond_kwargs=added_cond_kwargs).sample[:, :4]
-
-                if args.snr_gamma is None:
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-                else:
-                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
-                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
-                    # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective requires that we add one to SNR values before we divide by them.
-                        snr = snr + 1
-                    mse_loss_weights = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )
-
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
-                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
-                    loss = loss.mean()
-
-                # Gather the losses across all processes for logging (if we use distributed training).
-                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
-                train_loss += avg_loss.item() / args.gradient_accumulation_steps
-
-                # Backpropagate
-                accelerator.backward(loss)
-                if accelerator.sync_gradients:
-                    accelerator.clip_grad_norm_(unet.parameters(), args.max_grad_norm)
-                optimizer.step()
-                lr_scheduler.step()
-                optimizer.zero_grad()
-
-            # Checks if the accelerator has performed an optimization step behind the scenes
-            if accelerator.sync_gradients:
-                if args.use_ema:
-                    ema_unet.step(unet.parameters())
-                progress_bar.update(1)
-                global_step += 1
-                accelerator.log({"train_loss": train_loss}, step=global_step)
-                train_loss = 0.0
-
-                if global_step % args.checkpointing_steps == 0:
-                    if accelerator.is_main_process:
-                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
-                        if args.checkpoints_total_limit is not None:
-                            checkpoints = os.listdir(args.output_dir)
-                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
-                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
-
-                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
-                            if len(checkpoints) >= args.checkpoints_total_limit:
-                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
-                                removing_checkpoints = checkpoints[0:num_to_remove]
-
-                                logger.info(
-                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
-                                )
-                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
-
-                                for removing_checkpoint in removing_checkpoints:
-                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
-                                    shutil.rmtree(removing_checkpoint)
-
-                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
-                        accelerator.save_state(save_path)
-                        logger.info(f"Saved state to {save_path}")
-
-            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
-            progress_bar.set_postfix(**logs)
-
-            if global_step >= args.max_train_steps:
-                break
-
-        if accelerator.is_main_process:
-            if args.validation_prompts is not None and epoch % args.validation_epochs == 0:
-                if args.use_ema:
-                    # Store the UNet parameters temporarily and load the EMA parameters to perform inference.
-                    ema_unet.store(unet.parameters())
-                    ema_unet.copy_to(unet.parameters())
-                log_validation(
-                    vae,
-                    image_encoder,
-                    image_processor,
-                    unet,
-                    args,
-                    accelerator,
-                    weight_dtype,
-                    global_step,
-                )
-                if args.use_ema:
-                    # Switch back to the original UNet parameters.
-                    ema_unet.restore(unet.parameters())
-
-    # Create the pipeline using the trained modules and save it.
-    accelerator.wait_for_everyone()
-    if accelerator.is_main_process:
-        unet = accelerator.unwrap_model(unet)
-        if args.use_ema:
-            ema_unet.copy_to(unet.parameters())
-
-        pipeline = AutoPipelineForText2Image.from_pretrained(
-            args.pretrained_decoder_model_name_or_path,
-            vae=vae,
-            unet=unet,
-        )
-        pipeline.decoder_pipe.save_pretrained(args.output_dir)
-
-        # Run a final round of inference.
-        images = []
-        if args.validation_prompts is not None:
-            logger.info("Running inference for collecting generated images...")
-            pipeline = pipeline.to(accelerator.device)
-            pipeline.torch_dtype = weight_dtype
-            pipeline.set_progress_bar_config(disable=True)
-            pipeline.enable_model_cpu_offload()
-
-            if args.enable_xformers_memory_efficient_attention:
-                pipeline.enable_xformers_memory_efficient_attention()
-
-            if args.seed is None:
-                generator = None
-            else:
-                generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
-
-            for i in range(len(args.validation_prompts)):
-                with torch.autocast("cuda"):
-                    image = pipeline(args.validation_prompts[i], num_inference_steps=20, generator=generator).images[0]
-                images.append(image)
-
-        if args.push_to_hub:
-            save_model_card(args, repo_id, images, repo_folder=args.output_dir)
-            upload_folder(
-                repo_id=repo_id,
-                folder_path=args.output_dir,
-                commit_message="End of training",
-                ignore_patterns=["step_*", "epoch_*"],
-            )
-
-    accelerator.end_training()
-
-
-if __name__ == "__main__":
-    main()

examples/kandinsky2_2/text_to_image/train_text_to_image_lora_decoder.py

@@ -1,798 +0,0 @@
-# coding=utf-8
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Fine-tuning script for Kandinsky with support for LoRA."""
-
-import argparse
-import logging
-import math
-import os
-import shutil
-from pathlib import Path
-
-import datasets
-import numpy as np
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-import transformers
-from accelerate import Accelerator
-from accelerate.logging import get_logger
-from accelerate.utils import ProjectConfiguration, set_seed
-from datasets import load_dataset
-from huggingface_hub import create_repo, upload_folder
-from PIL import Image
-from tqdm import tqdm
-from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
-
-import diffusers
-from diffusers import AutoPipelineForText2Image, DDPMScheduler, UNet2DConditionModel, VQModel
-from diffusers.loaders import AttnProcsLayers
-from diffusers.models.attention_processor import LoRAAttnAddedKVProcessor
-from diffusers.optimization import get_scheduler
-from diffusers.training_utils import compute_snr
-from diffusers.utils import check_min_version, is_wandb_available
-
-
-# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.21.0.dev0")
-
-logger = get_logger(__name__, log_level="INFO")
-
-
-def save_model_card(repo_id: str, images=None, base_model=str, dataset_name=str, repo_folder=None):
-    img_str = ""
-    for i, image in enumerate(images):
-        image.save(os.path.join(repo_folder, f"image_{i}.png"))
-        img_str += f"![img_{i}](./image_{i}.png)\n"
-
-    yaml = f"""
----
-license: creativeml-openrail-m
-base_model: {base_model}
-tags:
-- kandinsky
-- text-to-image
-- diffusers
-- lora
-inference: true
----
-    """
-    model_card = f"""
-# LoRA text2image fine-tuning - {repo_id}
-These are LoRA adaption weights for {base_model}. The weights were fine-tuned on the {dataset_name} dataset. You can find some example images in the following. \n
-{img_str}
-"""
-    with open(os.path.join(repo_folder, "README.md"), "w") as f:
-        f.write(yaml + model_card)
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(description="Simple example of finetuning Kandinsky 2.2 with LoRA.")
-    parser.add_argument(
-        "--pretrained_decoder_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-decoder",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--pretrained_prior_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-prior",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--dataset_name",
-        type=str,
-        default=None,
-        help=(
-            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
-            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
-            " or to a folder containing files that 🤗 Datasets can understand."
-        ),
-    )
-    parser.add_argument(
-        "--dataset_config_name",
-        type=str,
-        default=None,
-        help="The config of the Dataset, leave as None if there's only one config.",
-    )
-    parser.add_argument(
-        "--train_data_dir",
-        type=str,
-        default=None,
-        help=(
-            "A folder containing the training data. Folder contents must follow the structure described in"
-            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
-            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
-        ),
-    )
-    parser.add_argument(
-        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
-    )
-    parser.add_argument(
-        "--validation_prompt", type=str, default=None, help="A prompt that is sampled during training for inference."
-    )
-    parser.add_argument(
-        "--num_validation_images",
-        type=int,
-        default=4,
-        help="Number of images that should be generated during validation with `validation_prompt`.",
-    )
-    parser.add_argument(
-        "--validation_epochs",
-        type=int,
-        default=1,
-        help=(
-            "Run fine-tuning validation every X epochs. The validation process consists of running the prompt"
-            " `args.validation_prompt` multiple times: `args.num_validation_images`."
-        ),
-    )
-    parser.add_argument(
-        "--max_train_samples",
-        type=int,
-        default=None,
-        help=(
-            "For debugging purposes or quicker training, truncate the number of training examples to this "
-            "value if set."
-        ),
-    )
-    parser.add_argument(
-        "--output_dir",
-        type=str,
-        default="kandi_2_2-model-finetuned-lora",
-        help="The output directory where the model predictions and checkpoints will be written.",
-    )
-    parser.add_argument(
-        "--cache_dir",
-        type=str,
-        default=None,
-        help="The directory where the downloaded models and datasets will be stored.",
-    )
-    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
-    parser.add_argument(
-        "--resolution",
-        type=int,
-        default=512,
-        help=(
-            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
-            " resolution"
-        ),
-    )
-    parser.add_argument(
-        "--train_batch_size", type=int, default=1, help="Batch size (per device) for the training dataloader."
-    )
-    parser.add_argument("--num_train_epochs", type=int, default=100)
-    parser.add_argument(
-        "--max_train_steps",
-        type=int,
-        default=None,
-        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
-    )
-    parser.add_argument(
-        "--gradient_accumulation_steps",
-        type=int,
-        default=1,
-        help="Number of updates steps to accumulate before performing a backward/update pass.",
-    )
-    parser.add_argument(
-        "--gradient_checkpointing",
-        action="store_true",
-        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
-    )
-    parser.add_argument(
-        "--learning_rate",
-        type=float,
-        default=1e-4,
-        help="Initial learning rate (after the potential warmup period) to use.",
-    )
-    parser.add_argument(
-        "--lr_scheduler",
-        type=str,
-        default="constant",
-        help=(
-            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
-            ' "constant", "constant_with_warmup"]'
-        ),
-    )
-    parser.add_argument(
-        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
-    )
-    parser.add_argument(
-        "--snr_gamma",
-        type=float,
-        default=None,
-        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
-        "More details here: https://arxiv.org/abs/2303.09556.",
-    )
-    parser.add_argument(
-        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
-    )
-    parser.add_argument(
-        "--allow_tf32",
-        action="store_true",
-        help=(
-            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
-            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
-        ),
-    )
-    parser.add_argument(
-        "--dataloader_num_workers",
-        type=int,
-        default=0,
-        help=(
-            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
-        ),
-    )
-    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
-    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
-    parser.add_argument("--adam_weight_decay", type=float, default=0.0, help="Weight decay to use.")
-    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
-    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
-    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
-    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
-    parser.add_argument(
-        "--hub_model_id",
-        type=str,
-        default=None,
-        help="The name of the repository to keep in sync with the local `output_dir`.",
-    )
-    parser.add_argument(
-        "--logging_dir",
-        type=str,
-        default="logs",
-        help=(
-            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
-            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
-        ),
-    )
-    parser.add_argument(
-        "--mixed_precision",
-        type=str,
-        default=None,
-        choices=["no", "fp16", "bf16"],
-        help=(
-            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
-            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
-            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
-        ),
-    )
-    parser.add_argument(
-        "--report_to",
-        type=str,
-        default="tensorboard",
-        help=(
-            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
-            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
-        ),
-    )
-    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
-    parser.add_argument(
-        "--checkpointing_steps",
-        type=int,
-        default=500,
-        help=(
-            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
-            " training using `--resume_from_checkpoint`."
-        ),
-    )
-    parser.add_argument(
-        "--checkpoints_total_limit",
-        type=int,
-        default=None,
-        help=("Max number of checkpoints to store."),
-    )
-    parser.add_argument(
-        "--resume_from_checkpoint",
-        type=str,
-        default=None,
-        help=(
-            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
-            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
-        ),
-    )
-    parser.add_argument(
-        "--rank",
-        type=int,
-        default=4,
-        help=("The dimension of the LoRA update matrices."),
-    )
-
-    args = parser.parse_args()
-    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
-    if env_local_rank != -1 and env_local_rank != args.local_rank:
-        args.local_rank = env_local_rank
-
-    # Sanity checks
-    if args.dataset_name is None and args.train_data_dir is None:
-        raise ValueError("Need either a dataset name or a training folder.")
-
-    return args
-
-
-def main():
-    args = parse_args()
-    logging_dir = Path(args.output_dir, args.logging_dir)
-    accelerator_project_config = ProjectConfiguration(
-        total_limit=args.checkpoints_total_limit, project_dir=args.output_dir, logging_dir=logging_dir
-    )
-    accelerator = Accelerator(
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        mixed_precision=args.mixed_precision,
-        log_with=args.report_to,
-        project_config=accelerator_project_config,
-    )
-    if args.report_to == "wandb":
-        if not is_wandb_available():
-            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
-        import wandb
-
-    # Make one log on every process with the configuration for debugging.
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO,
-    )
-    logger.info(accelerator.state, main_process_only=False)
-    if accelerator.is_local_main_process:
-        datasets.utils.logging.set_verbosity_warning()
-        transformers.utils.logging.set_verbosity_warning()
-        diffusers.utils.logging.set_verbosity_info()
-    else:
-        datasets.utils.logging.set_verbosity_error()
-        transformers.utils.logging.set_verbosity_error()
-        diffusers.utils.logging.set_verbosity_error()
-
-    # If passed along, set the training seed now.
-    if args.seed is not None:
-        set_seed(args.seed)
-
-    # Handle the repository creation
-    if accelerator.is_main_process:
-        if args.output_dir is not None:
-            os.makedirs(args.output_dir, exist_ok=True)
-
-        if args.push_to_hub:
-            repo_id = create_repo(
-                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
-            ).repo_id
-    # Load scheduler, tokenizer and models.
-    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="scheduler")
-    image_processor = CLIPImageProcessor.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_processor"
-    )
-    image_encoder = CLIPVisionModelWithProjection.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_encoder"
-    )
-
-    vae = VQModel.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="movq")
-
-    unet = UNet2DConditionModel.from_pretrained(args.pretrained_decoder_model_name_or_path, subfolder="unet")
-    # freeze parameters of models to save more memory
-    unet.requires_grad_(False)
-    vae.requires_grad_(False)
-
-    image_encoder.requires_grad_(False)
-
-    # For mixed precision training we cast all non-trainable weigths (vae, non-lora text_encoder and non-lora unet) to half-precision
-    # as these weights are only used for inference, keeping weights in full precision is not required.
-    weight_dtype = torch.float32
-    if accelerator.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-
-    # Move unet, vae and text_encoder to device and cast to weight_dtype
-    unet.to(accelerator.device, dtype=weight_dtype)
-    vae.to(accelerator.device, dtype=weight_dtype)
-    image_encoder.to(accelerator.device, dtype=weight_dtype)
-
-    lora_attn_procs = {}
-    for name in unet.attn_processors.keys():
-        cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
-        if name.startswith("mid_block"):
-            hidden_size = unet.config.block_out_channels[-1]
-        elif name.startswith("up_blocks"):
-            block_id = int(name[len("up_blocks.")])
-            hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
-        elif name.startswith("down_blocks"):
-            block_id = int(name[len("down_blocks.")])
-            hidden_size = unet.config.block_out_channels[block_id]
-
-        lora_attn_procs[name] = LoRAAttnAddedKVProcessor(
-            hidden_size=hidden_size,
-            cross_attention_dim=cross_attention_dim,
-            rank=args.rank,
-        )
-
-    unet.set_attn_processor(lora_attn_procs)
-
-    lora_layers = AttnProcsLayers(unet.attn_processors)
-
-    if args.allow_tf32:
-        torch.backends.cuda.matmul.allow_tf32 = True
-
-    if args.use_8bit_adam:
-        try:
-            import bitsandbytes as bnb
-        except ImportError:
-            raise ImportError(
-                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
-            )
-
-        optimizer_cls = bnb.optim.AdamW8bit
-    else:
-        optimizer_cls = torch.optim.AdamW
-
-    optimizer = optimizer_cls(
-        lora_layers.parameters(),
-        lr=args.learning_rate,
-        betas=(args.adam_beta1, args.adam_beta2),
-        weight_decay=args.adam_weight_decay,
-        eps=args.adam_epsilon,
-    )
-
-    # Get the datasets: you can either provide your own training and evaluation files (see below)
-    # or specify a Dataset from the hub (the dataset will be downloaded automatically from the datasets Hub).
-
-    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
-    # download the dataset.
-    if args.dataset_name is not None:
-        # Downloading and loading a dataset from the hub.
-        dataset = load_dataset(
-            args.dataset_name,
-            args.dataset_config_name,
-            cache_dir=args.cache_dir,
-        )
-    else:
-        data_files = {}
-        if args.train_data_dir is not None:
-            data_files["train"] = os.path.join(args.train_data_dir, "**")
-        dataset = load_dataset(
-            "imagefolder",
-            data_files=data_files,
-            cache_dir=args.cache_dir,
-        )
-        # See more about loading custom images at
-        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
-
-    # Preprocessing the datasets.
-    # We need to tokenize inputs and targets.
-    column_names = dataset["train"].column_names
-
-    image_column = args.image_column
-    if image_column not in column_names:
-        raise ValueError(f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}")
-
-    def center_crop(image):
-        width, height = image.size
-        new_size = min(width, height)
-        left = (width - new_size) / 2
-        top = (height - new_size) / 2
-        right = (width + new_size) / 2
-        bottom = (height + new_size) / 2
-        return image.crop((left, top, right, bottom))
-
-    def train_transforms(img):
-        img = center_crop(img)
-        img = img.resize((args.resolution, args.resolution), resample=Image.BICUBIC, reducing_gap=1)
-        img = np.array(img).astype(np.float32) / 127.5 - 1
-        img = torch.from_numpy(np.transpose(img, [2, 0, 1]))
-        return img
-
-    def preprocess_train(examples):
-        images = [image.convert("RGB") for image in examples[image_column]]
-        examples["pixel_values"] = [train_transforms(image) for image in images]
-        examples["clip_pixel_values"] = image_processor(images, return_tensors="pt").pixel_values
-        return examples
-
-    with accelerator.main_process_first():
-        if args.max_train_samples is not None:
-            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
-        # Set the training transforms
-        train_dataset = dataset["train"].with_transform(preprocess_train)
-
-    def collate_fn(examples):
-        pixel_values = torch.stack([example["pixel_values"] for example in examples])
-        pixel_values = pixel_values.to(memory_format=torch.contiguous_format).float()
-        clip_pixel_values = torch.stack([example["clip_pixel_values"] for example in examples])
-        clip_pixel_values = clip_pixel_values.to(memory_format=torch.contiguous_format).float()
-        return {"pixel_values": pixel_values, "clip_pixel_values": clip_pixel_values}
-
-    train_dataloader = torch.utils.data.DataLoader(
-        train_dataset,
-        shuffle=True,
-        collate_fn=collate_fn,
-        batch_size=args.train_batch_size,
-        num_workers=args.dataloader_num_workers,
-    )
-
-    # Scheduler and math around the number of training steps.
-    overrode_max_train_steps = False
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if args.max_train_steps is None:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        overrode_max_train_steps = True
-
-    lr_scheduler = get_scheduler(
-        args.lr_scheduler,
-        optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-    )
-    # Prepare everything with our `accelerator`.
-    lora_layers, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        lora_layers, optimizer, train_dataloader, lr_scheduler
-    )
-
-    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if overrode_max_train_steps:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-    # Afterwards we recalculate our number of training epochs
-    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-
-    # We need to initialize the trackers we use, and also store our configuration.
-    # The trackers initializes automatically on the main process.
-    if accelerator.is_main_process:
-        accelerator.init_trackers("text2image-fine-tune", config=vars(args))
-
-    # Train!
-    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-
-    logger.info("***** Running training *****")
-    logger.info(f"  Num examples = {len(train_dataset)}")
-    logger.info(f"  Num Epochs = {args.num_train_epochs}")
-    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
-    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
-    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
-    logger.info(f"  Total optimization steps = {args.max_train_steps}")
-    global_step = 0
-    first_epoch = 0
-
-    # Potentially load in the weights and states from a previous save
-    if args.resume_from_checkpoint:
-        if args.resume_from_checkpoint != "latest":
-            path = os.path.basename(args.resume_from_checkpoint)
-        else:
-            # Get the most recent checkpoint
-            dirs = os.listdir(args.output_dir)
-            dirs = [d for d in dirs if d.startswith("checkpoint")]
-            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
-            path = dirs[-1] if len(dirs) > 0 else None
-
-        if path is None:
-            accelerator.print(
-                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
-            )
-            args.resume_from_checkpoint = None
-            initial_global_step = 0
-        else:
-            accelerator.print(f"Resuming from checkpoint {path}")
-            accelerator.load_state(os.path.join(args.output_dir, path))
-            global_step = int(path.split("-")[1])
-
-            initial_global_step = global_step
-            first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
-
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
-
-    for epoch in range(first_epoch, args.num_train_epochs):
-        unet.train()
-        train_loss = 0.0
-        for step, batch in enumerate(train_dataloader):
-            with accelerator.accumulate(unet):
-                # Convert images to latent space
-                images = batch["pixel_values"].to(weight_dtype)
-                clip_images = batch["clip_pixel_values"].to(weight_dtype)
-                latents = vae.encode(images).latents
-                image_embeds = image_encoder(clip_images).image_embeds
-                # Sample noise that we'll add to the latents
-                noise = torch.randn_like(latents)
-                bsz = latents.shape[0]
-                # Sample a random timestep for each image
-                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (bsz,), device=latents.device)
-                timesteps = timesteps.long()
-
-                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
-
-                target = noise
-
-                # Predict the noise residual and compute loss
-                added_cond_kwargs = {"image_embeds": image_embeds}
-
-                model_pred = unet(noisy_latents, timesteps, None, added_cond_kwargs=added_cond_kwargs).sample[:, :4]
-
-                if args.snr_gamma is None:
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-                else:
-                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
-                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
-                    # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective requires that we add one to SNR values before we divide by them.
-                        snr = snr + 1
-                    mse_loss_weights = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )
-
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
-                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
-                    loss = loss.mean()
-
-                # Gather the losses across all processes for logging (if we use distributed training).
-                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
-                train_loss += avg_loss.item() / args.gradient_accumulation_steps
-
-                # Backpropagate
-                accelerator.backward(loss)
-                if accelerator.sync_gradients:
-                    params_to_clip = lora_layers.parameters()
-                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
-                optimizer.step()
-                lr_scheduler.step()
-                optimizer.zero_grad()
-
-            # Checks if the accelerator has performed an optimization step behind the scenes
-            if accelerator.sync_gradients:
-                progress_bar.update(1)
-                global_step += 1
-                accelerator.log({"train_loss": train_loss}, step=global_step)
-                train_loss = 0.0
-
-                if global_step % args.checkpointing_steps == 0:
-                    if accelerator.is_main_process:
-                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
-                        if args.checkpoints_total_limit is not None:
-                            checkpoints = os.listdir(args.output_dir)
-                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
-                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
-
-                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
-                            if len(checkpoints) >= args.checkpoints_total_limit:
-                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
-                                removing_checkpoints = checkpoints[0:num_to_remove]
-
-                                logger.info(
-                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
-                                )
-                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
-
-                                for removing_checkpoint in removing_checkpoints:
-                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
-                                    shutil.rmtree(removing_checkpoint)
-
-                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
-                        accelerator.save_state(save_path)
-                        logger.info(f"Saved state to {save_path}")
-
-            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
-            progress_bar.set_postfix(**logs)
-
-            if global_step >= args.max_train_steps:
-                break
-
-        if accelerator.is_main_process:
-            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
-                logger.info(
-                    f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
-                    f" {args.validation_prompt}."
-                )
-                # create pipeline
-                pipeline = AutoPipelineForText2Image.from_pretrained(
-                    args.pretrained_decoder_model_name_or_path,
-                    unet=accelerator.unwrap_model(unet),
-                    torch_dtype=weight_dtype,
-                )
-                pipeline = pipeline.to(accelerator.device)
-                pipeline.set_progress_bar_config(disable=True)
-
-                # run inference
-                generator = torch.Generator(device=accelerator.device)
-                if args.seed is not None:
-                    generator = generator.manual_seed(args.seed)
-                images = []
-                for _ in range(args.num_validation_images):
-                    images.append(
-                        pipeline(args.validation_prompt, num_inference_steps=30, generator=generator).images[0]
-                    )
-
-                for tracker in accelerator.trackers:
-                    if tracker.name == "tensorboard":
-                        np_images = np.stack([np.asarray(img) for img in images])
-                        tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
-                    if tracker.name == "wandb":
-                        tracker.log(
-                            {
-                                "validation": [
-                                    wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
-                                    for i, image in enumerate(images)
-                                ]
-                            }
-                        )
-
-                del pipeline
-                torch.cuda.empty_cache()
-
-    # Save the lora layers
-    accelerator.wait_for_everyone()
-    if accelerator.is_main_process:
-        unet = unet.to(torch.float32)
-        unet.save_attn_procs(args.output_dir)
-
-        if args.push_to_hub:
-            save_model_card(
-                repo_id,
-                images=images,
-                base_model=args.pretrained_decoder_model_name_or_path,
-                dataset_name=args.dataset_name,
-                repo_folder=args.output_dir,
-            )
-            upload_folder(
-                repo_id=repo_id,
-                folder_path=args.output_dir,
-                commit_message="End of training",
-                ignore_patterns=["step_*", "epoch_*"],
-            )
-
-    # Final inference
-    # Load previous pipeline
-    pipeline = AutoPipelineForText2Image.from_pretrained(
-        args.pretrained_decoder_model_name_or_path, torch_dtype=weight_dtype
-    )
-    pipeline = pipeline.to(accelerator.device)
-
-    # load attention processors
-    pipeline.unet.load_attn_procs(args.output_dir)
-
-    # run inference
-    generator = torch.Generator(device=accelerator.device)
-    if args.seed is not None:
-        generator = generator.manual_seed(args.seed)
-    images = []
-    for _ in range(args.num_validation_images):
-        images.append(pipeline(args.validation_prompt, num_inference_steps=30, generator=generator).images[0])
-
-    if accelerator.is_main_process:
-        for tracker in accelerator.trackers:
-            if len(images) != 0:
-                if tracker.name == "tensorboard":
-                    np_images = np.stack([np.asarray(img) for img in images])
-                    tracker.writer.add_images("test", np_images, epoch, dataformats="NHWC")
-                if tracker.name == "wandb":
-                    tracker.log(
-                        {
-                            "test": [
-                                wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
-                                for i, image in enumerate(images)
-                            ]
-                        }
-                    )
-
-    accelerator.end_training()
-
-
-if __name__ == "__main__":
-    main()

examples/kandinsky2_2/text_to_image/train_text_to_image_lora_prior.py

@@ -1,830 +0,0 @@
-# coding=utf-8
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-"""Fine-tuning script for Stable Diffusion for text2image with support for LoRA."""
-
-import argparse
-import logging
-import math
-import os
-import random
-import shutil
-from pathlib import Path
-
-import datasets
-import numpy as np
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-import transformers
-from accelerate import Accelerator
-from accelerate.logging import get_logger
-from accelerate.utils import ProjectConfiguration, set_seed
-from datasets import load_dataset
-from huggingface_hub import create_repo, upload_folder
-from tqdm import tqdm
-from transformers import CLIPImageProcessor, CLIPTextModelWithProjection, CLIPTokenizer, CLIPVisionModelWithProjection
-
-import diffusers
-from diffusers import AutoPipelineForText2Image, DDPMScheduler, PriorTransformer
-from diffusers.loaders import AttnProcsLayers
-from diffusers.models.attention_processor import LoRAAttnProcessor
-from diffusers.optimization import get_scheduler
-from diffusers.training_utils import compute_snr
-from diffusers.utils import check_min_version, is_wandb_available
-
-
-# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.21.0.dev0")
-
-logger = get_logger(__name__, log_level="INFO")
-
-
-def save_model_card(repo_id: str, images=None, base_model=str, dataset_name=str, repo_folder=None):
-    img_str = ""
-    for i, image in enumerate(images):
-        image.save(os.path.join(repo_folder, f"image_{i}.png"))
-        img_str += f"![img_{i}](./image_{i}.png)\n"
-
-    yaml = f"""
----
-license: creativeml-openrail-m
-base_model: {base_model}
-tags:
-- kandinsky
-- text-to-image
-- diffusers
-- lora
-inference: true
----
-    """
-    model_card = f"""
-# LoRA text2image fine-tuning - {repo_id}
-These are LoRA adaption weights for {base_model}. The weights were fine-tuned on the {dataset_name} dataset. You can find some example images in the following. \n
-{img_str}
-"""
-    with open(os.path.join(repo_folder, "README.md"), "w") as f:
-        f.write(yaml + model_card)
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(description="Simple example of finetuning Kandinsky 2.2.")
-    parser.add_argument(
-        "--pretrained_decoder_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-decoder",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--pretrained_prior_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-prior",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--dataset_name",
-        type=str,
-        default=None,
-        help=(
-            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
-            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
-            " or to a folder containing files that 🤗 Datasets can understand."
-        ),
-    )
-    parser.add_argument(
-        "--dataset_config_name",
-        type=str,
-        default=None,
-        help="The config of the Dataset, leave as None if there's only one config.",
-    )
-    parser.add_argument(
-        "--train_data_dir",
-        type=str,
-        default=None,
-        help=(
-            "A folder containing the training data. Folder contents must follow the structure described in"
-            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
-            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
-        ),
-    )
-    parser.add_argument(
-        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
-    )
-    parser.add_argument(
-        "--caption_column",
-        type=str,
-        default="text",
-        help="The column of the dataset containing a caption or a list of captions.",
-    )
-    parser.add_argument(
-        "--validation_prompt", type=str, default=None, help="A prompt that is sampled during training for inference."
-    )
-    parser.add_argument(
-        "--num_validation_images",
-        type=int,
-        default=4,
-        help="Number of images that should be generated during validation with `validation_prompt`.",
-    )
-    parser.add_argument(
-        "--validation_epochs",
-        type=int,
-        default=1,
-        help=(
-            "Run fine-tuning validation every X epochs. The validation process consists of running the prompt"
-            " `args.validation_prompt` multiple times: `args.num_validation_images`."
-        ),
-    )
-    parser.add_argument(
-        "--max_train_samples",
-        type=int,
-        default=None,
-        help=(
-            "For debugging purposes or quicker training, truncate the number of training examples to this "
-            "value if set."
-        ),
-    )
-    parser.add_argument(
-        "--output_dir",
-        type=str,
-        default="kandi_2_2-model-finetuned-lora",
-        help="The output directory where the model predictions and checkpoints will be written.",
-    )
-    parser.add_argument(
-        "--cache_dir",
-        type=str,
-        default=None,
-        help="The directory where the downloaded models and datasets will be stored.",
-    )
-    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
-    parser.add_argument(
-        "--resolution",
-        type=int,
-        default=512,
-        help=(
-            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
-            " resolution"
-        ),
-    )
-    parser.add_argument(
-        "--train_batch_size", type=int, default=1, help="Batch size (per device) for the training dataloader."
-    )
-    parser.add_argument("--num_train_epochs", type=int, default=100)
-    parser.add_argument(
-        "--max_train_steps",
-        type=int,
-        default=None,
-        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
-    )
-    parser.add_argument(
-        "--gradient_accumulation_steps",
-        type=int,
-        default=1,
-        help="Number of updates steps to accumulate before performing a backward/update pass.",
-    )
-    parser.add_argument(
-        "--learning_rate",
-        type=float,
-        default=1e-4,
-        help="learning rate",
-    )
-    parser.add_argument(
-        "--lr_scheduler",
-        type=str,
-        default="constant",
-        help=(
-            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
-            ' "constant", "constant_with_warmup"]'
-        ),
-    )
-    parser.add_argument(
-        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
-    )
-    parser.add_argument(
-        "--snr_gamma",
-        type=float,
-        default=None,
-        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
-        "More details here: https://arxiv.org/abs/2303.09556.",
-    )
-    parser.add_argument(
-        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
-    )
-    parser.add_argument(
-        "--allow_tf32",
-        action="store_true",
-        help=(
-            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
-            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
-        ),
-    )
-    parser.add_argument(
-        "--dataloader_num_workers",
-        type=int,
-        default=0,
-        help=(
-            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
-        ),
-    )
-    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
-    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
-    parser.add_argument(
-        "--adam_weight_decay",
-        type=float,
-        default=0.0,
-        required=False,
-        help="weight decay_to_use",
-    )
-    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
-    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
-    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
-    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
-    parser.add_argument(
-        "--hub_model_id",
-        type=str,
-        default=None,
-        help="The name of the repository to keep in sync with the local `output_dir`.",
-    )
-    parser.add_argument(
-        "--logging_dir",
-        type=str,
-        default="logs",
-        help=(
-            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
-            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
-        ),
-    )
-    parser.add_argument(
-        "--mixed_precision",
-        type=str,
-        default=None,
-        choices=["no", "fp16", "bf16"],
-        help=(
-            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
-            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
-            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
-        ),
-    )
-    parser.add_argument(
-        "--report_to",
-        type=str,
-        default="tensorboard",
-        help=(
-            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
-            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
-        ),
-    )
-    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
-    parser.add_argument(
-        "--checkpointing_steps",
-        type=int,
-        default=500,
-        help=(
-            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
-            " training using `--resume_from_checkpoint`."
-        ),
-    )
-    parser.add_argument(
-        "--checkpoints_total_limit",
-        type=int,
-        default=None,
-        help=("Max number of checkpoints to store."),
-    )
-    parser.add_argument(
-        "--resume_from_checkpoint",
-        type=str,
-        default=None,
-        help=(
-            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
-            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
-        ),
-    )
-    parser.add_argument(
-        "--rank",
-        type=int,
-        default=4,
-        help=("The dimension of the LoRA update matrices."),
-    )
-
-    args = parser.parse_args()
-    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
-    if env_local_rank != -1 and env_local_rank != args.local_rank:
-        args.local_rank = env_local_rank
-
-    # Sanity checks
-    if args.dataset_name is None and args.train_data_dir is None:
-        raise ValueError("Need either a dataset name or a training folder.")
-
-    return args
-
-
-DATASET_NAME_MAPPING = {
-    "lambdalabs/pokemon-blip-captions": ("image", "text"),
-}
-
-
-def main():
-    args = parse_args()
-    logging_dir = Path(args.output_dir, args.logging_dir)
-
-    accelerator_project_config = ProjectConfiguration(
-        total_limit=args.checkpoints_total_limit, project_dir=args.output_dir, logging_dir=logging_dir
-    )
-    accelerator = Accelerator(
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        mixed_precision=args.mixed_precision,
-        log_with=args.report_to,
-        project_config=accelerator_project_config,
-    )
-    if args.report_to == "wandb":
-        if not is_wandb_available():
-            raise ImportError("Make sure to install wandb if you want to use it for logging during training.")
-        import wandb
-
-    # Make one log on every process with the configuration for debugging.
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO,
-    )
-    logger.info(accelerator.state, main_process_only=False)
-    if accelerator.is_local_main_process:
-        datasets.utils.logging.set_verbosity_warning()
-        transformers.utils.logging.set_verbosity_warning()
-        diffusers.utils.logging.set_verbosity_info()
-    else:
-        datasets.utils.logging.set_verbosity_error()
-        transformers.utils.logging.set_verbosity_error()
-        diffusers.utils.logging.set_verbosity_error()
-
-    # If passed along, set the training seed now.
-    if args.seed is not None:
-        set_seed(args.seed)
-
-    # Handle the repository creation
-    if accelerator.is_main_process:
-        if args.output_dir is not None:
-            os.makedirs(args.output_dir, exist_ok=True)
-
-        if args.push_to_hub:
-            repo_id = create_repo(
-                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
-            ).repo_id
-    # Load scheduler, image_processor, tokenizer and models.
-    noise_scheduler = DDPMScheduler(beta_schedule="squaredcos_cap_v2", prediction_type="sample")
-    image_processor = CLIPImageProcessor.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_processor"
-    )
-    tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder="tokenizer")
-    image_encoder = CLIPVisionModelWithProjection.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_encoder"
-    )
-    text_encoder = CLIPTextModelWithProjection.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="text_encoder"
-    )
-    prior = PriorTransformer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder="prior")
-    # freeze parameters of models to save more memory
-    image_encoder.requires_grad_(False)
-    prior.requires_grad_(False)
-    text_encoder.requires_grad_(False)
-    weight_dtype = torch.float32
-    if accelerator.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-
-    # Move image_encoder, text_encoder and prior to device and cast to weight_dtype
-    prior.to(accelerator.device, dtype=weight_dtype)
-    image_encoder.to(accelerator.device, dtype=weight_dtype)
-    text_encoder.to(accelerator.device, dtype=weight_dtype)
-    lora_attn_procs = {}
-    for name in prior.attn_processors.keys():
-        lora_attn_procs[name] = LoRAAttnProcessor(hidden_size=2048, rank=args.rank)
-
-    prior.set_attn_processor(lora_attn_procs)
-    lora_layers = AttnProcsLayers(prior.attn_processors)
-
-    if args.allow_tf32:
-        torch.backends.cuda.matmul.allow_tf32 = True
-
-    if args.use_8bit_adam:
-        try:
-            import bitsandbytes as bnb
-        except ImportError:
-            raise ImportError(
-                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
-            )
-
-        optimizer_cls = bnb.optim.AdamW8bit
-    else:
-        optimizer_cls = torch.optim.AdamW
-
-    optimizer = optimizer_cls(
-        lora_layers.parameters(),
-        lr=args.learning_rate,
-        betas=(args.adam_beta1, args.adam_beta2),
-        weight_decay=args.adam_weight_decay,
-        eps=args.adam_epsilon,
-    )
-
-    # Get the datasets: you can either provide your own training and evaluation files (see below)
-    # or specify a Dataset from the hub (the dataset will be downloaded automatically from the datasets Hub).
-
-    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
-    # download the dataset.
-    if args.dataset_name is not None:
-        # Downloading and loading a dataset from the hub.
-        dataset = load_dataset(
-            args.dataset_name,
-            args.dataset_config_name,
-            cache_dir=args.cache_dir,
-        )
-    else:
-        data_files = {}
-        if args.train_data_dir is not None:
-            data_files["train"] = os.path.join(args.train_data_dir, "**")
-        dataset = load_dataset(
-            "imagefolder",
-            data_files=data_files,
-            cache_dir=args.cache_dir,
-        )
-        # See more about loading custom images at
-        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
-
-    # Preprocessing the datasets.
-    # We need to tokenize inputs and targets.
-    column_names = dataset["train"].column_names
-
-    # 6. Get the column names for input/target.
-    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
-    if args.image_column is None:
-        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
-    else:
-        image_column = args.image_column
-        if image_column not in column_names:
-            raise ValueError(
-                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
-            )
-    if args.caption_column is None:
-        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
-    else:
-        caption_column = args.caption_column
-        if caption_column not in column_names:
-            raise ValueError(
-                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
-            )
-
-    # Preprocessing the datasets.
-    # We need to tokenize input captions and transform the images.
-    def tokenize_captions(examples, is_train=True):
-        captions = []
-        for caption in examples[caption_column]:
-            if isinstance(caption, str):
-                captions.append(caption)
-            elif isinstance(caption, (list, np.ndarray)):
-                # take a random caption if there are multiple
-                captions.append(random.choice(caption) if is_train else caption[0])
-            else:
-                raise ValueError(
-                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
-                )
-        inputs = tokenizer(
-            captions, max_length=tokenizer.model_max_length, padding="max_length", truncation=True, return_tensors="pt"
-        )
-        text_input_ids = inputs.input_ids
-        text_mask = inputs.attention_mask.bool()
-        return text_input_ids, text_mask
-
-    def preprocess_train(examples):
-        images = [image.convert("RGB") for image in examples[image_column]]
-        examples["clip_pixel_values"] = image_processor(images, return_tensors="pt").pixel_values
-        examples["text_input_ids"], examples["text_mask"] = tokenize_captions(examples)
-        return examples
-
-    with accelerator.main_process_first():
-        if args.max_train_samples is not None:
-            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
-        # Set the training transforms
-        train_dataset = dataset["train"].with_transform(preprocess_train)
-
-    def collate_fn(examples):
-        clip_pixel_values = torch.stack([example["clip_pixel_values"] for example in examples])
-        clip_pixel_values = clip_pixel_values.to(memory_format=torch.contiguous_format).float()
-        text_input_ids = torch.stack([example["text_input_ids"] for example in examples])
-        text_mask = torch.stack([example["text_mask"] for example in examples])
-        return {"clip_pixel_values": clip_pixel_values, "text_input_ids": text_input_ids, "text_mask": text_mask}
-
-    # DataLoaders creation:
-    train_dataloader = torch.utils.data.DataLoader(
-        train_dataset,
-        shuffle=True,
-        collate_fn=collate_fn,
-        batch_size=args.train_batch_size,
-        num_workers=args.dataloader_num_workers,
-    )
-
-    # Scheduler and math around the number of training steps.
-    overrode_max_train_steps = False
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if args.max_train_steps is None:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        overrode_max_train_steps = True
-
-    lr_scheduler = get_scheduler(
-        args.lr_scheduler,
-        optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-    )
-    clip_mean = prior.clip_mean.clone()
-    clip_std = prior.clip_std.clone()
-    lora_layers, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        lora_layers, optimizer, train_dataloader, lr_scheduler
-    )
-
-    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if overrode_max_train_steps:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-    # Afterwards we recalculate our number of training epochs
-    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-
-    # We need to initialize the trackers we use, and also store our configuration.
-    # The trackers initializes automatically on the main process.
-    if accelerator.is_main_process:
-        accelerator.init_trackers("text2image-fine-tune", config=vars(args))
-
-    # Train!
-    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-
-    logger.info("***** Running training *****")
-    logger.info(f"  Num examples = {len(train_dataset)}")
-    logger.info(f"  Num Epochs = {args.num_train_epochs}")
-    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
-    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
-    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
-    logger.info(f"  Total optimization steps = {args.max_train_steps}")
-    global_step = 0
-    first_epoch = 0
-
-    # Potentially load in the weights and states from a previous save
-    if args.resume_from_checkpoint:
-        if args.resume_from_checkpoint != "latest":
-            path = os.path.basename(args.resume_from_checkpoint)
-        else:
-            # Get the most recent checkpoint
-            dirs = os.listdir(args.output_dir)
-            dirs = [d for d in dirs if d.startswith("checkpoint")]
-            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
-            path = dirs[-1] if len(dirs) > 0 else None
-
-        if path is None:
-            accelerator.print(
-                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
-            )
-            args.resume_from_checkpoint = None
-            initial_global_step = 0
-        else:
-            accelerator.print(f"Resuming from checkpoint {path}")
-            accelerator.load_state(os.path.join(args.output_dir, path))
-            global_step = int(path.split("-")[1])
-
-            initial_global_step = global_step
-            first_epoch = global_step // num_update_steps_per_epoch
-
-    else:
-        initial_global_step = 0
-
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
-
-    clip_mean = clip_mean.to(weight_dtype).to(accelerator.device)
-    clip_std = clip_std.to(weight_dtype).to(accelerator.device)
-
-    for epoch in range(first_epoch, args.num_train_epochs):
-        prior.train()
-        train_loss = 0.0
-        for step, batch in enumerate(train_dataloader):
-            with accelerator.accumulate(prior):
-                # Convert images to latent space
-                text_input_ids, text_mask, clip_images = (
-                    batch["text_input_ids"],
-                    batch["text_mask"],
-                    batch["clip_pixel_values"].to(weight_dtype),
-                )
-                with torch.no_grad():
-                    text_encoder_output = text_encoder(text_input_ids)
-                    prompt_embeds = text_encoder_output.text_embeds
-                    text_encoder_hidden_states = text_encoder_output.last_hidden_state
-
-                    image_embeds = image_encoder(clip_images).image_embeds
-                    # Sample noise that we'll add to the image_embeds
-                    noise = torch.randn_like(image_embeds)
-                    bsz = image_embeds.shape[0]
-                    # Sample a random timestep for each image
-                    timesteps = torch.randint(
-                        0, noise_scheduler.config.num_train_timesteps, (bsz,), device=image_embeds.device
-                    )
-                    timesteps = timesteps.long()
-                    image_embeds = (image_embeds - clip_mean) / clip_std
-                    noisy_latents = noise_scheduler.add_noise(image_embeds, noise, timesteps)
-
-                    target = image_embeds
-
-                # Predict the noise residual and compute loss
-                model_pred = prior(
-                    noisy_latents,
-                    timestep=timesteps,
-                    proj_embedding=prompt_embeds,
-                    encoder_hidden_states=text_encoder_hidden_states,
-                    attention_mask=text_mask,
-                ).predicted_image_embedding
-
-                if args.snr_gamma is None:
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-                else:
-                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
-                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
-                    # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective requires that we add one to SNR values before we divide by them.
-                        snr = snr + 1
-                    mse_loss_weights = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )
-
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
-                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
-                    loss = loss.mean()
-
-                # Gather the losses across all processes for logging (if we use distributed training).
-                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
-                train_loss += avg_loss.item() / args.gradient_accumulation_steps
-
-                # Backpropagate
-                accelerator.backward(loss)
-                if accelerator.sync_gradients:
-                    accelerator.clip_grad_norm_(prior.parameters(), args.max_grad_norm)
-                optimizer.step()
-                lr_scheduler.step()
-                optimizer.zero_grad()
-
-            # Checks if the accelerator has performed an optimization step behind the scenes
-            if accelerator.sync_gradients:
-                progress_bar.update(1)
-                global_step += 1
-                accelerator.log({"train_loss": train_loss}, step=global_step)
-                train_loss = 0.0
-
-                if global_step % args.checkpointing_steps == 0:
-                    if accelerator.is_main_process:
-                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
-                        if args.checkpoints_total_limit is not None:
-                            checkpoints = os.listdir(args.output_dir)
-                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
-                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
-
-                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
-                            if len(checkpoints) >= args.checkpoints_total_limit:
-                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
-                                removing_checkpoints = checkpoints[0:num_to_remove]
-
-                                logger.info(
-                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
-                                )
-                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
-
-                                for removing_checkpoint in removing_checkpoints:
-                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
-                                    shutil.rmtree(removing_checkpoint)
-
-                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
-                        accelerator.save_state(save_path)
-                        logger.info(f"Saved state to {save_path}")
-
-            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
-            progress_bar.set_postfix(**logs)
-
-            if global_step >= args.max_train_steps:
-                break
-
-        if accelerator.is_main_process:
-            if args.validation_prompt is not None and epoch % args.validation_epochs == 0:
-                logger.info(
-                    f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
-                    f" {args.validation_prompt}."
-                )
-                # create pipeline
-                pipeline = AutoPipelineForText2Image.from_pretrained(
-                    args.pretrained_decoder_model_name_or_path,
-                    prior_prior=accelerator.unwrap_model(prior),
-                    torch_dtype=weight_dtype,
-                )
-                pipeline = pipeline.to(accelerator.device)
-                pipeline.set_progress_bar_config(disable=True)
-
-                # run inference
-                generator = torch.Generator(device=accelerator.device)
-                if args.seed is not None:
-                    generator = generator.manual_seed(args.seed)
-                images = []
-                for _ in range(args.num_validation_images):
-                    images.append(
-                        pipeline(args.validation_prompt, num_inference_steps=30, generator=generator).images[0]
-                    )
-
-                for tracker in accelerator.trackers:
-                    if tracker.name == "tensorboard":
-                        np_images = np.stack([np.asarray(img) for img in images])
-                        tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
-                    if tracker.name == "wandb":
-                        tracker.log(
-                            {
-                                "validation": [
-                                    wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
-                                    for i, image in enumerate(images)
-                                ]
-                            }
-                        )
-
-                del pipeline
-                torch.cuda.empty_cache()
-
-    # Save the lora layers
-    accelerator.wait_for_everyone()
-    if accelerator.is_main_process:
-        prior = prior.to(torch.float32)
-        prior.save_attn_procs(args.output_dir)
-
-        if args.push_to_hub:
-            save_model_card(
-                repo_id,
-                images=images,
-                base_model=args.pretrained_prior_model_name_or_path,
-                dataset_name=args.dataset_name,
-                repo_folder=args.output_dir,
-            )
-            upload_folder(
-                repo_id=repo_id,
-                folder_path=args.output_dir,
-                commit_message="End of training",
-                ignore_patterns=["step_*", "epoch_*"],
-            )
-
-    # Final inference
-    # Load previous pipeline
-    pipeline = AutoPipelineForText2Image.from_pretrained(
-        args.pretrained_decoder_model_name_or_path, torch_dtype=weight_dtype
-    )
-    pipeline = pipeline.to(accelerator.device)
-
-    # load attention processors
-    pipeline.prior_prior.load_attn_procs(args.output_dir)
-
-    # run inference
-    generator = torch.Generator(device=accelerator.device)
-    if args.seed is not None:
-        generator = generator.manual_seed(args.seed)
-    images = []
-    for _ in range(args.num_validation_images):
-        images.append(pipeline(args.validation_prompt, num_inference_steps=30, generator=generator).images[0])
-
-    if accelerator.is_main_process:
-        for tracker in accelerator.trackers:
-            if len(images) != 0:
-                if tracker.name == "tensorboard":
-                    np_images = np.stack([np.asarray(img) for img in images])
-                    tracker.writer.add_images("test", np_images, epoch, dataformats="NHWC")
-                if tracker.name == "wandb":
-                    tracker.log(
-                        {
-                            "test": [
-                                wandb.Image(image, caption=f"{i}: {args.validation_prompt}")
-                                for i, image in enumerate(images)
-                            ]
-                        }
-                    )
-
-    accelerator.end_training()
-
-
-if __name__ == "__main__":
-    main()

examples/kandinsky2_2/text_to_image/train_text_to_image_prior.py

@@ -1,945 +0,0 @@
-#!/usr/bin/env python
-# coding=utf-8
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-
-import argparse
-import logging
-import math
-import os
-import random
-import shutil
-from pathlib import Path
-
-import accelerate
-import datasets
-import numpy as np
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-import transformers
-from accelerate import Accelerator
-from accelerate.logging import get_logger
-from accelerate.state import AcceleratorState
-from accelerate.utils import ProjectConfiguration, set_seed
-from datasets import load_dataset
-from huggingface_hub import create_repo, upload_folder
-from packaging import version
-from tqdm import tqdm
-from transformers import CLIPImageProcessor, CLIPTextModelWithProjection, CLIPTokenizer, CLIPVisionModelWithProjection
-from transformers.utils import ContextManagers
-
-import diffusers
-from diffusers import AutoPipelineForText2Image, DDPMScheduler, PriorTransformer
-from diffusers.optimization import get_scheduler
-from diffusers.training_utils import EMAModel, compute_snr
-from diffusers.utils import check_min_version, is_wandb_available, make_image_grid
-
-
-if is_wandb_available():
-    import wandb
-
-
-# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.21.0.dev0")
-
-logger = get_logger(__name__, log_level="INFO")
-
-DATASET_NAME_MAPPING = {
-    "lambdalabs/pokemon-blip-captions": ("image", "text"),
-}
-
-
-def save_model_card(
-    args,
-    repo_id: str,
-    images=None,
-    repo_folder=None,
-):
-    img_str = ""
-    if len(images) > 0:
-        image_grid = make_image_grid(images, 1, len(args.validation_prompts))
-        image_grid.save(os.path.join(repo_folder, "val_imgs_grid.png"))
-        img_str += "![val_imgs_grid](./val_imgs_grid.png)\n"
-
-    yaml = f"""
----
-license: creativeml-openrail-m
-base_model: {args.pretrained_prior_model_name_or_path}
-datasets:
-- {args.dataset_name}
-tags:
-- kandinsky
-- text-to-image
-- diffusers
-inference: true
----
-    """
-    model_card = f"""
-# Finetuning - {repo_id}
-
-This pipeline was finetuned from **{args.pretrained_prior_model_name_or_path}** on the **{args.dataset_name}** dataset. Below are some example images generated with the finetuned pipeline using the following prompts: {args.validation_prompts}: \n
-{img_str}
-
-## Pipeline usage
-
-You can use the pipeline like so:
-
-```python
-from diffusers import DiffusionPipeline
-import torch
-
-pipe_prior = DiffusionPipeline.from_pretrained("{repo_id}", torch_dtype=torch.float16)
-pipe_t2i = DiffusionPipeline.from_pretrained("{args.pretrained_decoder_model_name_or_path}", torch_dtype=torch.float16)
-prompt = "{args.validation_prompts[0]}"
-image_embeds, negative_image_embeds = pipe_prior(prompt, guidance_scale=1.0).to_tuple()
-image = pipe_t2i(image_embeds=image_embeds, negative_image_embeds=negative_image_embeds).images[0]
-image.save("my_image.png")
-```
-
-## Training info
-
-These are the key hyperparameters used during training:
-
-* Epochs: {args.num_train_epochs}
-* Learning rate: {args.learning_rate}
-* Batch size: {args.train_batch_size}
-* Gradient accumulation steps: {args.gradient_accumulation_steps}
-* Image resolution: {args.resolution}
-* Mixed-precision: {args.mixed_precision}
-
-"""
-    wandb_info = ""
-    if is_wandb_available():
-        wandb_run_url = None
-        if wandb.run is not None:
-            wandb_run_url = wandb.run.url
-
-    if wandb_run_url is not None:
-        wandb_info = f"""
-More information on all the CLI arguments and the environment are available on your [`wandb` run page]({wandb_run_url}).
-"""
-
-    model_card += wandb_info
-
-    with open(os.path.join(repo_folder, "README.md"), "w") as f:
-        f.write(yaml + model_card)
-
-
-def log_validation(
-    image_encoder, image_processor, text_encoder, tokenizer, prior, args, accelerator, weight_dtype, epoch
-):
-    logger.info("Running validation... ")
-
-    pipeline = AutoPipelineForText2Image.from_pretrained(
-        args.pretrained_decoder_model_name_or_path,
-        prior_image_encoder=accelerator.unwrap_model(image_encoder),
-        prior_image_processor=image_processor,
-        prior_text_encoder=accelerator.unwrap_model(text_encoder),
-        prior_tokenizer=tokenizer,
-        prior_prior=accelerator.unwrap_model(prior),
-        torch_dtype=weight_dtype,
-    )
-    pipeline = pipeline.to(accelerator.device)
-    pipeline.set_progress_bar_config(disable=True)
-
-    if args.seed is None:
-        generator = None
-    else:
-        generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
-
-    images = []
-    for i in range(len(args.validation_prompts)):
-        with torch.autocast("cuda"):
-            image = pipeline(args.validation_prompts[i], num_inference_steps=20, generator=generator).images[0]
-
-        images.append(image)
-
-    for tracker in accelerator.trackers:
-        if tracker.name == "tensorboard":
-            np_images = np.stack([np.asarray(img) for img in images])
-            tracker.writer.add_images("validation", np_images, epoch, dataformats="NHWC")
-        elif tracker.name == "wandb":
-            tracker.log(
-                {
-                    "validation": [
-                        wandb.Image(image, caption=f"{i}: {args.validation_prompts[i]}")
-                        for i, image in enumerate(images)
-                    ]
-                }
-            )
-        else:
-            logger.warn(f"image logging not implemented for {tracker.name}")
-
-    del pipeline
-    torch.cuda.empty_cache()
-
-    return images
-
-
-def parse_args():
-    parser = argparse.ArgumentParser(description="Simple example of finetuning Kandinsky 2.2.")
-    parser.add_argument(
-        "--pretrained_decoder_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-decoder",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--pretrained_prior_model_name_or_path",
-        type=str,
-        default="kandinsky-community/kandinsky-2-2-prior",
-        required=False,
-        help="Path to pretrained model or model identifier from huggingface.co/models.",
-    )
-    parser.add_argument(
-        "--dataset_name",
-        type=str,
-        default=None,
-        help=(
-            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
-            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
-            " or to a folder containing files that 🤗 Datasets can understand."
-        ),
-    )
-    parser.add_argument(
-        "--dataset_config_name",
-        type=str,
-        default=None,
-        help="The config of the Dataset, leave as None if there's only one config.",
-    )
-    parser.add_argument(
-        "--train_data_dir",
-        type=str,
-        default=None,
-        help=(
-            "A folder containing the training data. Folder contents must follow the structure described in"
-            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
-            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
-        ),
-    )
-    parser.add_argument(
-        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
-    )
-    parser.add_argument(
-        "--caption_column",
-        type=str,
-        default="text",
-        help="The column of the dataset containing a caption or a list of captions.",
-    )
-    parser.add_argument(
-        "--max_train_samples",
-        type=int,
-        default=None,
-        help=(
-            "For debugging purposes or quicker training, truncate the number of training examples to this "
-            "value if set."
-        ),
-    )
-    parser.add_argument(
-        "--validation_prompts",
-        type=str,
-        default=None,
-        nargs="+",
-        help=("A set of prompts evaluated every `--validation_epochs` and logged to `--report_to`."),
-    )
-    parser.add_argument(
-        "--output_dir",
-        type=str,
-        default="kandi_2_2-model-finetuned",
-        help="The output directory where the model predictions and checkpoints will be written.",
-    )
-    parser.add_argument(
-        "--cache_dir",
-        type=str,
-        default=None,
-        help="The directory where the downloaded models and datasets will be stored.",
-    )
-    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
-    parser.add_argument(
-        "--resolution",
-        type=int,
-        default=512,
-        help=(
-            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
-            " resolution"
-        ),
-    )
-    parser.add_argument(
-        "--train_batch_size", type=int, default=1, help="Batch size (per device) for the training dataloader."
-    )
-    parser.add_argument("--num_train_epochs", type=int, default=100)
-    parser.add_argument(
-        "--max_train_steps",
-        type=int,
-        default=None,
-        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
-    )
-    parser.add_argument(
-        "--gradient_accumulation_steps",
-        type=int,
-        default=1,
-        help="Number of updates steps to accumulate before performing a backward/update pass.",
-    )
-    parser.add_argument(
-        "--learning_rate",
-        type=float,
-        default=1e-4,
-        help="learning rate",
-    )
-    parser.add_argument(
-        "--lr_scheduler",
-        type=str,
-        default="constant",
-        help=(
-            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
-            ' "constant", "constant_with_warmup"]'
-        ),
-    )
-    parser.add_argument(
-        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
-    )
-    parser.add_argument(
-        "--snr_gamma",
-        type=float,
-        default=None,
-        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
-        "More details here: https://arxiv.org/abs/2303.09556.",
-    )
-    parser.add_argument(
-        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
-    )
-    parser.add_argument(
-        "--allow_tf32",
-        action="store_true",
-        help=(
-            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
-            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
-        ),
-    )
-    parser.add_argument("--use_ema", action="store_true", help="Whether to use EMA model.")
-    parser.add_argument(
-        "--dataloader_num_workers",
-        type=int,
-        default=0,
-        help=(
-            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
-        ),
-    )
-    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
-    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
-    parser.add_argument(
-        "--adam_weight_decay",
-        type=float,
-        default=0.0,
-        required=False,
-        help="weight decay_to_use",
-    )
-    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
-    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
-    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
-    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
-    parser.add_argument(
-        "--hub_model_id",
-        type=str,
-        default=None,
-        help="The name of the repository to keep in sync with the local `output_dir`.",
-    )
-    parser.add_argument(
-        "--logging_dir",
-        type=str,
-        default="logs",
-        help=(
-            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
-            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
-        ),
-    )
-    parser.add_argument(
-        "--mixed_precision",
-        type=str,
-        default=None,
-        choices=["no", "fp16", "bf16"],
-        help=(
-            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
-            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
-            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
-        ),
-    )
-    parser.add_argument(
-        "--report_to",
-        type=str,
-        default="tensorboard",
-        help=(
-            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
-            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
-        ),
-    )
-    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
-    parser.add_argument(
-        "--checkpointing_steps",
-        type=int,
-        default=500,
-        help=(
-            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
-            " training using `--resume_from_checkpoint`."
-        ),
-    )
-    parser.add_argument(
-        "--checkpoints_total_limit",
-        type=int,
-        default=None,
-        help=("Max number of checkpoints to store."),
-    )
-    parser.add_argument(
-        "--resume_from_checkpoint",
-        type=str,
-        default=None,
-        help=(
-            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
-            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
-        ),
-    )
-    parser.add_argument(
-        "--validation_epochs",
-        type=int,
-        default=5,
-        help="Run validation every X epochs.",
-    )
-    parser.add_argument(
-        "--tracker_project_name",
-        type=str,
-        default="text2image-fine-tune",
-        help=(
-            "The `project_name` argument passed to Accelerator.init_trackers for"
-            " more information see https://huggingface.co/docs/accelerate/v0.17.0/en/package_reference/accelerator#accelerate.Accelerator"
-        ),
-    )
-
-    args = parser.parse_args()
-    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
-    if env_local_rank != -1 and env_local_rank != args.local_rank:
-        args.local_rank = env_local_rank
-
-    # Sanity checks
-    if args.dataset_name is None and args.train_data_dir is None:
-        raise ValueError("Need either a dataset name or a training folder.")
-
-    return args
-
-
-def main():
-    args = parse_args()
-    logging_dir = os.path.join(args.output_dir, args.logging_dir)
-    accelerator_project_config = ProjectConfiguration(
-        total_limit=args.checkpoints_total_limit, project_dir=args.output_dir, logging_dir=logging_dir
-    )
-    accelerator = Accelerator(
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        mixed_precision=args.mixed_precision,
-        log_with=args.report_to,
-        project_config=accelerator_project_config,
-    )
-
-    # Make one log on every process with the configuration for debugging.
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO,
-    )
-    logger.info(accelerator.state, main_process_only=False)
-    if accelerator.is_local_main_process:
-        datasets.utils.logging.set_verbosity_warning()
-        transformers.utils.logging.set_verbosity_warning()
-        diffusers.utils.logging.set_verbosity_info()
-    else:
-        datasets.utils.logging.set_verbosity_error()
-        transformers.utils.logging.set_verbosity_error()
-        diffusers.utils.logging.set_verbosity_error()
-
-    # If passed along, set the training seed now.
-    if args.seed is not None:
-        set_seed(args.seed)
-
-    # Handle the repository creation
-    if accelerator.is_main_process:
-        if args.output_dir is not None:
-            os.makedirs(args.output_dir, exist_ok=True)
-
-        if args.push_to_hub:
-            repo_id = create_repo(
-                repo_id=args.hub_model_id or Path(args.output_dir).name, exist_ok=True, token=args.hub_token
-            ).repo_id
-
-    # Load scheduler, image_processor, tokenizer and models.
-    noise_scheduler = DDPMScheduler(beta_schedule="squaredcos_cap_v2", prediction_type="sample")
-    image_processor = CLIPImageProcessor.from_pretrained(
-        args.pretrained_prior_model_name_or_path, subfolder="image_processor"
-    )
-    tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder="tokenizer")
-
-    def deepspeed_zero_init_disabled_context_manager():
-        """
-        returns either a context list that includes one that will disable zero.Init or an empty context list
-        """
-        deepspeed_plugin = AcceleratorState().deepspeed_plugin if accelerate.state.is_initialized() else None
-        if deepspeed_plugin is None:
-            return []
-
-        return [deepspeed_plugin.zero3_init_context_manager(enable=False)]
-
-    weight_dtype = torch.float32
-    if accelerator.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-    with ContextManagers(deepspeed_zero_init_disabled_context_manager()):
-        image_encoder = CLIPVisionModelWithProjection.from_pretrained(
-            args.pretrained_prior_model_name_or_path, subfolder="image_encoder", torch_dtype=weight_dtype
-        ).eval()
-        text_encoder = CLIPTextModelWithProjection.from_pretrained(
-            args.pretrained_prior_model_name_or_path, subfolder="text_encoder", torch_dtype=weight_dtype
-        ).eval()
-
-    prior = PriorTransformer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder="prior")
-
-    # Freeze text_encoder and image_encoder
-    text_encoder.requires_grad_(False)
-    image_encoder.requires_grad_(False)
-
-    # Set prior to trainable.
-    prior.train()
-
-    # Create EMA for the prior.
-    if args.use_ema:
-        ema_prior = PriorTransformer.from_pretrained(args.pretrained_prior_model_name_or_path, subfolder="prior")
-        ema_prior = EMAModel(ema_prior.parameters(), model_cls=PriorTransformer, model_config=ema_prior.config)
-        ema_prior.to(accelerator.device)
-
-    # `accelerate` 0.16.0 will have better support for customized saving
-    if version.parse(accelerate.__version__) >= version.parse("0.16.0"):
-        # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
-        def save_model_hook(models, weights, output_dir):
-            if args.use_ema:
-                ema_prior.save_pretrained(os.path.join(output_dir, "prior_ema"))
-
-            for i, model in enumerate(models):
-                model.save_pretrained(os.path.join(output_dir, "prior"))
-
-                # make sure to pop weight so that corresponding model is not saved again
-                weights.pop()
-
-        def load_model_hook(models, input_dir):
-            if args.use_ema:
-                load_model = EMAModel.from_pretrained(os.path.join(input_dir, "prior_ema"), PriorTransformer)
-                ema_prior.load_state_dict(load_model.state_dict())
-                ema_prior.to(accelerator.device)
-                del load_model
-
-            for i in range(len(models)):
-                # pop models so that they are not loaded again
-                model = models.pop()
-
-                # load diffusers style into model
-                load_model = PriorTransformer.from_pretrained(input_dir, subfolder="prior")
-                model.register_to_config(**load_model.config)
-
-                model.load_state_dict(load_model.state_dict())
-                del load_model
-
-        accelerator.register_save_state_pre_hook(save_model_hook)
-        accelerator.register_load_state_pre_hook(load_model_hook)
-
-    if args.allow_tf32:
-        torch.backends.cuda.matmul.allow_tf32 = True
-
-    if args.use_8bit_adam:
-        try:
-            import bitsandbytes as bnb
-        except ImportError:
-            raise ImportError(
-                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
-            )
-
-        optimizer_cls = bnb.optim.AdamW8bit
-    else:
-        optimizer_cls = torch.optim.AdamW
-    optimizer = optimizer_cls(
-        prior.parameters(),
-        lr=args.learning_rate,
-        betas=(args.adam_beta1, args.adam_beta2),
-        weight_decay=args.adam_weight_decay,
-        eps=args.adam_epsilon,
-    )
-
-    # Get the datasets: you can either provide your own training and evaluation files (see below)
-    # or specify a Dataset from the hub (the dataset will be downloaded automatically from the datasets Hub).
-
-    # In distributed training, the load_dataset function guarantees that only one local process can concurrently
-    # download the dataset.
-    if args.dataset_name is not None:
-        # Downloading and loading a dataset from the hub.
-        dataset = load_dataset(
-            args.dataset_name,
-            args.dataset_config_name,
-            cache_dir=args.cache_dir,
-        )
-    else:
-        data_files = {}
-        if args.train_data_dir is not None:
-            data_files["train"] = os.path.join(args.train_data_dir, "**")
-        dataset = load_dataset(
-            "imagefolder",
-            data_files=data_files,
-            cache_dir=args.cache_dir,
-        )
-        # See more about loading custom images at
-        # https://huggingface.co/docs/datasets/v2.4.0/en/image_load#imagefolder
-
-    # Preprocessing the datasets.
-    # We need to tokenize inputs and targets.
-    column_names = dataset["train"].column_names
-
-    # 6. Get the column names for input/target.
-    dataset_columns = DATASET_NAME_MAPPING.get(args.dataset_name, None)
-    if args.image_column is None:
-        image_column = dataset_columns[0] if dataset_columns is not None else column_names[0]
-    else:
-        image_column = args.image_column
-        if image_column not in column_names:
-            raise ValueError(
-                f"--image_column' value '{args.image_column}' needs to be one of: {', '.join(column_names)}"
-            )
-    if args.caption_column is None:
-        caption_column = dataset_columns[1] if dataset_columns is not None else column_names[1]
-    else:
-        caption_column = args.caption_column
-        if caption_column not in column_names:
-            raise ValueError(
-                f"--caption_column' value '{args.caption_column}' needs to be one of: {', '.join(column_names)}"
-            )
-
-    # Preprocessing the datasets.
-    # We need to tokenize input captions and transform the images.
-    def tokenize_captions(examples, is_train=True):
-        captions = []
-        for caption in examples[caption_column]:
-            if isinstance(caption, str):
-                captions.append(caption)
-            elif isinstance(caption, (list, np.ndarray)):
-                # take a random caption if there are multiple
-                captions.append(random.choice(caption) if is_train else caption[0])
-            else:
-                raise ValueError(
-                    f"Caption column `{caption_column}` should contain either strings or lists of strings."
-                )
-        inputs = tokenizer(
-            captions, max_length=tokenizer.model_max_length, padding="max_length", truncation=True, return_tensors="pt"
-        )
-        text_input_ids = inputs.input_ids
-        text_mask = inputs.attention_mask.bool()
-        return text_input_ids, text_mask
-
-    def preprocess_train(examples):
-        images = [image.convert("RGB") for image in examples[image_column]]
-        examples["clip_pixel_values"] = image_processor(images, return_tensors="pt").pixel_values
-        examples["text_input_ids"], examples["text_mask"] = tokenize_captions(examples)
-        return examples
-
-    with accelerator.main_process_first():
-        if args.max_train_samples is not None:
-            dataset["train"] = dataset["train"].shuffle(seed=args.seed).select(range(args.max_train_samples))
-        # Set the training transforms
-        train_dataset = dataset["train"].with_transform(preprocess_train)
-
-    def collate_fn(examples):
-        clip_pixel_values = torch.stack([example["clip_pixel_values"] for example in examples])
-        clip_pixel_values = clip_pixel_values.to(memory_format=torch.contiguous_format).float()
-        text_input_ids = torch.stack([example["text_input_ids"] for example in examples])
-        text_mask = torch.stack([example["text_mask"] for example in examples])
-        return {"clip_pixel_values": clip_pixel_values, "text_input_ids": text_input_ids, "text_mask": text_mask}
-
-    # DataLoaders creation:
-    train_dataloader = torch.utils.data.DataLoader(
-        train_dataset,
-        shuffle=True,
-        collate_fn=collate_fn,
-        batch_size=args.train_batch_size,
-        num_workers=args.dataloader_num_workers,
-    )
-
-    # Scheduler and math around the number of training steps.
-    overrode_max_train_steps = False
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if args.max_train_steps is None:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-        overrode_max_train_steps = True
-
-    lr_scheduler = get_scheduler(
-        args.lr_scheduler,
-        optimizer=optimizer,
-        num_warmup_steps=args.lr_warmup_steps * args.gradient_accumulation_steps,
-        num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-    )
-
-    clip_mean = prior.clip_mean.clone()
-    clip_std = prior.clip_std.clone()
-
-    prior, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        prior, optimizer, train_dataloader, lr_scheduler
-    )
-
-    image_encoder.to(accelerator.device, dtype=weight_dtype)
-    text_encoder.to(accelerator.device, dtype=weight_dtype)
-    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-    if overrode_max_train_steps:
-        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
-    # Afterwards we recalculate our number of training epochs
-    args.num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-
-    # We need to initialize the trackers we use, and also store our configuration.
-    # The trackers initializes automatically on the main process.
-    if accelerator.is_main_process:
-        tracker_config = dict(vars(args))
-        tracker_config.pop("validation_prompts")
-        accelerator.init_trackers(args.tracker_project_name, tracker_config)
-
-    # Train!
-    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-
-    logger.info("***** Running training *****")
-    logger.info(f"  Num examples = {len(train_dataset)}")
-    logger.info(f"  Num Epochs = {args.num_train_epochs}")
-    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
-    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
-    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
-    logger.info(f"  Total optimization steps = {args.max_train_steps}")
-    global_step = 0
-    first_epoch = 0
-
-    # Potentially load in the weights and states from a previous save
-    if args.resume_from_checkpoint:
-        if args.resume_from_checkpoint != "latest":
-            path = os.path.basename(args.resume_from_checkpoint)
-        else:
-            # Get the most recent checkpoint
-            dirs = os.listdir(args.output_dir)
-            dirs = [d for d in dirs if d.startswith("checkpoint")]
-            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
-            path = dirs[-1] if len(dirs) > 0 else None
-
-        if path is None:
-            accelerator.print(
-                f"Checkpoint '{args.resume_from_checkpoint}' does not exist. Starting a new training run."
-            )
-            args.resume_from_checkpoint = None
-            initial_global_step = 0
-        else:
-            accelerator.print(f"Resuming from checkpoint {path}")
-            accelerator.load_state(os.path.join(args.output_dir, path))
-            global_step = int(path.split("-")[1])
-
-            initial_global_step = global_step
-            first_epoch = global_step // num_update_steps_per_epoch
-    else:
-        initial_global_step = 0
-
-    progress_bar = tqdm(
-        range(0, args.max_train_steps),
-        initial=initial_global_step,
-        desc="Steps",
-        # Only show the progress bar once on each machine.
-        disable=not accelerator.is_local_main_process,
-    )
-
-    clip_mean = clip_mean.to(weight_dtype).to(accelerator.device)
-    clip_std = clip_std.to(weight_dtype).to(accelerator.device)
-
-    for epoch in range(first_epoch, args.num_train_epochs):
-        train_loss = 0.0
-        for step, batch in enumerate(train_dataloader):
-            with accelerator.accumulate(prior):
-                # Convert images to latent space
-                text_input_ids, text_mask, clip_images = (
-                    batch["text_input_ids"],
-                    batch["text_mask"],
-                    batch["clip_pixel_values"].to(weight_dtype),
-                )
-                with torch.no_grad():
-                    text_encoder_output = text_encoder(text_input_ids)
-                    prompt_embeds = text_encoder_output.text_embeds
-                    text_encoder_hidden_states = text_encoder_output.last_hidden_state
-
-                    image_embeds = image_encoder(clip_images).image_embeds
-                    # Sample noise that we'll add to the image_embeds
-                    noise = torch.randn_like(image_embeds)
-                    bsz = image_embeds.shape[0]
-                    # Sample a random timestep for each image
-                    timesteps = torch.randint(
-                        0, noise_scheduler.config.num_train_timesteps, (bsz,), device=image_embeds.device
-                    )
-                    timesteps = timesteps.long()
-                    image_embeds = (image_embeds - clip_mean) / clip_std
-                    noisy_latents = noise_scheduler.add_noise(image_embeds, noise, timesteps)
-
-                    target = image_embeds
-
-                # Predict the noise residual and compute loss
-                model_pred = prior(
-                    noisy_latents,
-                    timestep=timesteps,
-                    proj_embedding=prompt_embeds,
-                    encoder_hidden_states=text_encoder_hidden_states,
-                    attention_mask=text_mask,
-                ).predicted_image_embedding
-
-                if args.snr_gamma is None:
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-                else:
-                    # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
-                    # Since we predict the noise instead of x_0, the original formulation is slightly changed.
-                    # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective requires that we add one to SNR values before we divide by them.
-                        snr = snr + 1
-                    mse_loss_weights = (
-                        torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
-                    )
-
-                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
-                    loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
-                    loss = loss.mean()
-
-                # Gather the losses across all processes for logging (if we use distributed training).
-                avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
-                train_loss += avg_loss.item() / args.gradient_accumulation_steps
-
-                # Backpropagate
-                accelerator.backward(loss)
-                if accelerator.sync_gradients:
-                    accelerator.clip_grad_norm_(prior.parameters(), args.max_grad_norm)
-                optimizer.step()
-                lr_scheduler.step()
-                optimizer.zero_grad()
-
-            # Checks if the accelerator has performed an optimization step behind the scenes
-            if accelerator.sync_gradients:
-                if args.use_ema:
-                    ema_prior.step(prior.parameters())
-                progress_bar.update(1)
-                global_step += 1
-                accelerator.log({"train_loss": train_loss}, step=global_step)
-                train_loss = 0.0
-
-                if global_step % args.checkpointing_steps == 0:
-                    if accelerator.is_main_process:
-                        # _before_ saving state, check if this save would set us over the `checkpoints_total_limit`
-                        if args.checkpoints_total_limit is not None:
-                            checkpoints = os.listdir(args.output_dir)
-                            checkpoints = [d for d in checkpoints if d.startswith("checkpoint")]
-                            checkpoints = sorted(checkpoints, key=lambda x: int(x.split("-")[1]))
-
-                            # before we save the new checkpoint, we need to have at _most_ `checkpoints_total_limit - 1` checkpoints
-                            if len(checkpoints) >= args.checkpoints_total_limit:
-                                num_to_remove = len(checkpoints) - args.checkpoints_total_limit + 1
-                                removing_checkpoints = checkpoints[0:num_to_remove]
-
-                                logger.info(
-                                    f"{len(checkpoints)} checkpoints already exist, removing {len(removing_checkpoints)} checkpoints"
-                                )
-                                logger.info(f"removing checkpoints: {', '.join(removing_checkpoints)}")
-
-                                for removing_checkpoint in removing_checkpoints:
-                                    removing_checkpoint = os.path.join(args.output_dir, removing_checkpoint)
-                                    shutil.rmtree(removing_checkpoint)
-
-                        save_path = os.path.join(args.output_dir, f"checkpoint-{global_step}")
-                        accelerator.save_state(save_path)
-                        logger.info(f"Saved state to {save_path}")
-
-            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
-            progress_bar.set_postfix(**logs)
-
-            if global_step >= args.max_train_steps:
-                break
-
-        if accelerator.is_main_process:
-            if args.validation_prompts is not None and epoch % args.validation_epochs == 0:
-                if args.use_ema:
-                    # Store the UNet parameters temporarily and load the EMA parameters to perform inference.
-                    ema_prior.store(prior.parameters())
-                    ema_prior.copy_to(prior.parameters())
-                log_validation(
-                    image_encoder,
-                    image_processor,
-                    text_encoder,
-                    tokenizer,
-                    prior,
-                    args,
-                    accelerator,
-                    weight_dtype,
-                    global_step,
-                )
-                if args.use_ema:
-                    # Switch back to the original UNet parameters.
-                    ema_prior.restore(prior.parameters())
-
-    # Create the pipeline using the trained modules and save it.
-    accelerator.wait_for_everyone()
-    if accelerator.is_main_process:
-        prior = accelerator.unwrap_model(prior)
-        if args.use_ema:
-            ema_prior.copy_to(prior.parameters())
-
-        pipeline = AutoPipelineForText2Image.from_pretrained(
-            args.pretrained_decoder_model_name_or_path,
-            prior_image_encoder=image_encoder,
-            prior_text_encoder=text_encoder,
-            prior_prior=prior,
-        )
-        pipeline.prior_pipe.save_pretrained(args.output_dir)
-
-        # Run a final round of inference.
-        images = []
-        if args.validation_prompts is not None:
-            logger.info("Running inference for collecting generated images...")
-            pipeline = pipeline.to(accelerator.device)
-            pipeline.torch_dtype = weight_dtype
-            pipeline.set_progress_bar_config(disable=True)
-
-            if args.seed is None:
-                generator = None
-            else:
-                generator = torch.Generator(device=accelerator.device).manual_seed(args.seed)
-
-            for i in range(len(args.validation_prompts)):
-                with torch.autocast("cuda"):
-                    image = pipeline(args.validation_prompts[i], num_inference_steps=20, generator=generator).images[0]
-                images.append(image)
-
-        if args.push_to_hub:
-            save_model_card(args, repo_id, images, repo_folder=args.output_dir)
-            upload_folder(
-                repo_id=repo_id,
-                folder_path=args.output_dir,
-                commit_message="End of training",
-                ignore_patterns=["step_*", "epoch_*"],
-            )
-
-    accelerator.end_training()
-
-
-if __name__ == "__main__":
-    main()

examples/research_projects/onnxruntime/text_to_image/train_text_to_image.py

@@ -44,7 +44,7 @@ from transformers.utils import ContextManagers
 import diffusers
 from diffusers import AutoencoderKL, DDPMScheduler, StableDiffusionPipeline, UNet2DConditionModel
 from diffusers.optimization import get_scheduler
-from diffusers.training_utils import EMAModel, compute_snr
+from diffusers.training_utils import EMAModel
 from diffusers.utils import check_min_version, deprecate, is_wandb_available
 from diffusers.utils.import_utils import is_xformers_available
 
@@ -524,6 +524,30 @@ def main():
         else:
             raise ValueError("xformers is not available. Make sure it is installed correctly")
 
+    def compute_snr(timesteps):
+        """
+        Computes SNR as per https://github.com/TiankaiHang/Min-SNR-Diffusion-Training/blob/521b624bd70c67cee4bdf49225915f5945a872e3/guided_diffusion/gaussian_diffusion.py#L847-L849
+        """
+        alphas_cumprod = noise_scheduler.alphas_cumprod
+        sqrt_alphas_cumprod = alphas_cumprod**0.5
+        sqrt_one_minus_alphas_cumprod = (1.0 - alphas_cumprod) ** 0.5
+
+        # Expand the tensors.
+        # Adapted from https://github.com/TiankaiHang/Min-SNR-Diffusion-Training/blob/521b624bd70c67cee4bdf49225915f5945a872e3/guided_diffusion/gaussian_diffusion.py#L1026
+        sqrt_alphas_cumprod = sqrt_alphas_cumprod.to(device=timesteps.device)[timesteps].float()
+        while len(sqrt_alphas_cumprod.shape) < len(timesteps.shape):
+            sqrt_alphas_cumprod = sqrt_alphas_cumprod[..., None]
+        alpha = sqrt_alphas_cumprod.expand(timesteps.shape)
+
+        sqrt_one_minus_alphas_cumprod = sqrt_one_minus_alphas_cumprod.to(device=timesteps.device)[timesteps].float()
+        while len(sqrt_one_minus_alphas_cumprod.shape) < len(timesteps.shape):
+            sqrt_one_minus_alphas_cumprod = sqrt_one_minus_alphas_cumprod[..., None]
+        sigma = sqrt_one_minus_alphas_cumprod.expand(timesteps.shape)
+
+        # Compute SNR.
+        snr = (alpha / sigma) ** 2
+        return snr
+
     # `accelerate` 0.16.0 will have better support for customized saving
     if version.parse(accelerate.__version__) >= version.parse("0.16.0"):
         # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
@@ -847,14 +871,13 @@ def main():
                     # Compute loss-weights as per Section 3.4 of https://arxiv.org/abs/2303.09556.
                     # Since we predict the noise instead of x_0, the original formulation is slightly changed.
                     # This is discussed in Section 4.2 of the same paper.
-                    snr = compute_snr(noise_scheduler, timesteps)
-                    if noise_scheduler.config.prediction_type == "v_prediction":
-                        # Velocity objective requires that we add one to SNR values before we divide by them.
-                        snr = snr + 1
+                    snr = compute_snr(timesteps)
                     mse_loss_weights = (
                         torch.stack([snr, args.snr_gamma * torch.ones_like(timesteps)], dim=1).min(dim=1)[0] / snr
                     )
-
+                    # We first calculate the original loss. Then we mean over the non-batch dimensions and
+                    # rebalance the sample-wise losses with their respective loss weights.
+                    # Finally, we take the mean of the rebalanced loss.
                     loss = F.mse_loss(model_pred.float(), target.float(), reduction="none")
                     loss = loss.mean(dim=list(range(1, len(loss.shape)))) * mse_loss_weights
                     loss = loss.mean()