Merge branch 'main' into fix-torchao-groupoffloading

up

.github/workflows/claude_review.yml

@@ -32,6 +32,9 @@ jobs:
       )
     runs-on: ubuntu-latest
     steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 1
       - uses: anthropics/claude-code-action@v1
         with:
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}

src/diffusers/hooks/group_offloading.py

@@ -22,7 +22,7 @@ from typing import Set
 import safetensors.torch
 import torch
 
-from ..utils import get_logger, is_accelerate_available
+from ..utils import get_logger, is_accelerate_available, is_torchao_available
 from ._common import _GO_LC_SUPPORTED_PYTORCH_LAYERS
 from .hooks import HookRegistry, ModelHook
 
@@ -35,6 +35,54 @@ if is_accelerate_available():
 logger = get_logger(__name__)  # pylint: disable=invalid-name
 
 
+def _is_torchao_tensor(tensor: torch.Tensor) -> bool:
+    if not is_torchao_available():
+        return False
+    from torchao.utils import TorchAOBaseTensor
+
+    return isinstance(tensor, TorchAOBaseTensor)
+
+
+def _get_torchao_inner_tensor_names(tensor: torch.Tensor) -> list[str]:
+    """Get names of all internal tensor data attributes from a TorchAO tensor."""
+    cls = type(tensor)
+    names = list(getattr(cls, "tensor_data_names", []))
+    for attr_name in getattr(cls, "optional_tensor_data_names", []):
+        if getattr(tensor, attr_name, None) is not None:
+            names.append(attr_name)
+    return names
+
+
+def _swap_torchao_tensor(param: torch.Tensor, source: torch.Tensor) -> None:
+    """Move a TorchAO parameter to the device of `source` via `swap_tensors`.
+
+    `param.data = source` does not work for `_make_wrapper_subclass` tensors because the `.data` setter only replaces
+    the outer wrapper storage while leaving the subclass's internal attributes (e.g. `.qdata`, `.scale`) on the
+    original device. `swap_tensors` swaps the full tensor contents in-place, preserving the parameter's identity so
+    that any dict keyed by `id(param)` remains valid.
+
+    Refer to https://github.com/huggingface/diffusers/pull/13276#discussion_r2944471548 for the full discussion.
+    """
+    torch.utils.swap_tensors(param, source)
+
+
+def _restore_torchao_tensor(param: torch.Tensor, source: torch.Tensor) -> None:
+    """Restore internal tensor data of a TorchAO parameter from `source` without mutating `source`.
+
+    Unlike `_swap_torchao_tensor` this copies attribute references one-by-one via `setattr` so that `source` is **not**
+    modified. Use this when `source` is a cached tensor that must remain unchanged (e.g. a pinned CPU copy in
+    `cpu_param_dict`).
+    """
+    for attr_name in _get_torchao_inner_tensor_names(source):
+        setattr(param, attr_name, getattr(source, attr_name))
+
+
+def _record_stream_torchao_tensor(param: torch.Tensor, stream) -> None:
+    """Record stream for all internal tensors of a TorchAO parameter."""
+    for attr_name in _get_torchao_inner_tensor_names(param):
+        getattr(param, attr_name).record_stream(stream)
+
+
 # fmt: off
 _GROUP_OFFLOADING = "group_offloading"
 _LAYER_EXECUTION_TRACKER = "layer_execution_tracker"
@@ -157,9 +205,16 @@ class ModuleGroup:
             pinned_dict = None
 
     def _transfer_tensor_to_device(self, tensor, source_tensor, default_stream):
-        tensor.data = source_tensor.to(self.onload_device, non_blocking=self.non_blocking)
+        moved = source_tensor.to(self.onload_device, non_blocking=self.non_blocking)
+        if _is_torchao_tensor(tensor):
+            _swap_torchao_tensor(tensor, moved)
+        else:
+            tensor.data = moved
         if self.record_stream:
-            tensor.data.record_stream(default_stream)
+            if _is_torchao_tensor(tensor):
+                _record_stream_torchao_tensor(tensor, default_stream)
+            else:
+                tensor.data.record_stream(default_stream)
 
     def _process_tensors_from_modules(self, pinned_memory=None, default_stream=None):
         for group_module in self.modules:
@@ -245,18 +300,35 @@ class ModuleGroup:
 
             for group_module in self.modules:
                 for param in group_module.parameters():
-                    param.data = self.cpu_param_dict[param]
+                    if _is_torchao_tensor(param):
+                        _restore_torchao_tensor(param, self.cpu_param_dict[param])
+                    else:
+                        param.data = self.cpu_param_dict[param]
             for param in self.parameters:
-                param.data = self.cpu_param_dict[param]
+                if _is_torchao_tensor(param):
+                    _restore_torchao_tensor(param, self.cpu_param_dict[param])
+                else:
+                    param.data = self.cpu_param_dict[param]
             for buffer in self.buffers:
-                buffer.data = self.cpu_param_dict[buffer]
+                if _is_torchao_tensor(buffer):
+                    _restore_torchao_tensor(buffer, self.cpu_param_dict[buffer])
+                else:
+                    buffer.data = self.cpu_param_dict[buffer]
         else:
             for group_module in self.modules:
                 group_module.to(self.offload_device, non_blocking=False)
             for param in self.parameters:
-                param.data = param.data.to(self.offload_device, non_blocking=False)
+                if _is_torchao_tensor(param):
+                    moved = param.data.to(self.offload_device, non_blocking=False)
+                    _swap_torchao_tensor(param, moved)
+                else:
+                    param.data = param.data.to(self.offload_device, non_blocking=False)
             for buffer in self.buffers:
-                buffer.data = buffer.data.to(self.offload_device, non_blocking=False)
+                if _is_torchao_tensor(buffer):
+                    moved = buffer.data.to(self.offload_device, non_blocking=False)
+                    _swap_torchao_tensor(buffer, moved)
+                else:
+                    buffer.data = buffer.data.to(self.offload_device, non_blocking=False)
 
     @torch.compiler.disable()
     def onload_(self):

tests/models/transformers/test_models_transformer_cosmos.py

@@ -12,46 +12,60 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import unittest
+
 import torch
 
 from diffusers import CosmosTransformer3DModel
-from diffusers.utils.torch_utils import randn_tensor
 
 from ...testing_utils import enable_full_determinism, torch_device
-from ..testing_utils import (
-    BaseModelTesterConfig,
-    MemoryTesterMixin,
-    ModelTesterMixin,
-    TrainingTesterMixin,
-)
+from ..test_modeling_common import ModelTesterMixin
 
 
 enable_full_determinism()
 
 
-class CosmosTransformerTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return CosmosTransformer3DModel
+class CosmosTransformer3DModelTests(ModelTesterMixin, unittest.TestCase):
+    model_class = CosmosTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
 
     @property
-    def output_shape(self) -> tuple[int, ...]:
-        return (4, 1, 16, 16)
+    def dummy_input(self):
+        batch_size = 1
+        num_channels = 4
+        num_frames = 1
+        height = 16
+        width = 16
+        text_embed_dim = 16
+        sequence_length = 12
+        fps = 30
 
-    @property
-    def input_shape(self) -> tuple[int, ...]:
-        return (4, 1, 16, 16)
+        hidden_states = torch.randn((batch_size, num_channels, num_frames, height, width)).to(torch_device)
+        timestep = torch.randint(0, 1000, size=(batch_size,)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, text_embed_dim)).to(torch_device)
+        attention_mask = torch.ones((batch_size, sequence_length)).to(torch_device)
+        padding_mask = torch.zeros(batch_size, 1, height, width).to(torch_device)
 
-    @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self) -> dict[str, int | list | tuple | float | bool | str]:
         return {
+            "hidden_states": hidden_states,
+            "timestep": timestep,
+            "encoder_hidden_states": encoder_hidden_states,
+            "attention_mask": attention_mask,
+            "fps": fps,
+            "padding_mask": padding_mask,
+        }
+
+    @property
+    def input_shape(self):
+        return (4, 1, 16, 16)
+
+    @property
+    def output_shape(self):
+        return (4, 1, 16, 16)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "in_channels": 4,
             "out_channels": 4,
             "num_attention_heads": 2,
@@ -66,68 +80,57 @@ class CosmosTransformerTesterConfig(BaseModelTesterConfig):
             "concat_padding_mask": True,
             "extra_pos_embed_type": "learnable",
         }
-
-    def get_dummy_inputs(self, batch_size: int = 1) -> dict[str, torch.Tensor]:
-        num_channels = 4
-        num_frames = 1
-        height = 16
-        width = 16
-        text_embed_dim = 16
-        sequence_length = 12
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_channels, num_frames, height, width), generator=self.generator, device=torch_device
-            ),
-            "timestep": torch.randint(0, 1000, size=(batch_size,), generator=self.generator).to(torch_device),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, text_embed_dim), generator=self.generator, device=torch_device
-            ),
-            "attention_mask": torch.ones((batch_size, sequence_length)).to(torch_device),
-            "fps": 30,
-            "padding_mask": torch.zeros(batch_size, 1, height, width).to(torch_device),
-        }
-
-
-class TestCosmosTransformer(CosmosTransformerTesterConfig, ModelTesterMixin):
-    """Core model tests for Cosmos Transformer."""
-
-
-class TestCosmosTransformerMemory(CosmosTransformerTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for Cosmos Transformer."""
-
-
-class TestCosmosTransformerTraining(CosmosTransformerTesterConfig, TrainingTesterMixin):
-    """Training tests for Cosmos Transformer."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"CosmosTransformer3DModel"}
         super().test_gradient_checkpointing_is_applied(expected_set=expected_set)
 
 
-class CosmosTransformerVideoToWorldTesterConfig(BaseModelTesterConfig):
-    @property
-    def model_class(self):
-        return CosmosTransformer3DModel
+class CosmosTransformer3DModelVideoToWorldTests(ModelTesterMixin, unittest.TestCase):
+    model_class = CosmosTransformer3DModel
+    main_input_name = "hidden_states"
+    uses_custom_attn_processor = True
 
     @property
-    def output_shape(self) -> tuple[int, ...]:
-        return (4, 1, 16, 16)
+    def dummy_input(self):
+        batch_size = 1
+        num_channels = 4
+        num_frames = 1
+        height = 16
+        width = 16
+        text_embed_dim = 16
+        sequence_length = 12
+        fps = 30
 
-    @property
-    def input_shape(self) -> tuple[int, ...]:
-        return (4, 1, 16, 16)
+        hidden_states = torch.randn((batch_size, num_channels, num_frames, height, width)).to(torch_device)
+        timestep = torch.randint(0, 1000, size=(batch_size,)).to(torch_device)
+        encoder_hidden_states = torch.randn((batch_size, sequence_length, text_embed_dim)).to(torch_device)
+        attention_mask = torch.ones((batch_size, sequence_length)).to(torch_device)
+        condition_mask = torch.ones(batch_size, 1, num_frames, height, width).to(torch_device)
+        padding_mask = torch.zeros(batch_size, 1, height, width).to(torch_device)
 
-    @property
-    def main_input_name(self) -> str:
-        return "hidden_states"
-
-    @property
-    def generator(self):
-        return torch.Generator("cpu").manual_seed(0)
-
-    def get_init_dict(self) -> dict[str, int | list | tuple | float | bool | str]:
         return {
+            "hidden_states": hidden_states,
+            "timestep": timestep,
+            "encoder_hidden_states": encoder_hidden_states,
+            "attention_mask": attention_mask,
+            "fps": fps,
+            "condition_mask": condition_mask,
+            "padding_mask": padding_mask,
+        }
+
+    @property
+    def input_shape(self):
+        return (4, 1, 16, 16)
+
+    @property
+    def output_shape(self):
+        return (4, 1, 16, 16)
+
+    def prepare_init_args_and_inputs_for_common(self):
+        init_dict = {
             "in_channels": 4 + 1,
             "out_channels": 4,
             "num_attention_heads": 2,
@@ -142,40 +145,8 @@ class CosmosTransformerVideoToWorldTesterConfig(BaseModelTesterConfig):
             "concat_padding_mask": True,
             "extra_pos_embed_type": "learnable",
         }
-
-    def get_dummy_inputs(self, batch_size: int = 1) -> dict[str, torch.Tensor]:
-        num_channels = 4
-        num_frames = 1
-        height = 16
-        width = 16
-        text_embed_dim = 16
-        sequence_length = 12
-
-        return {
-            "hidden_states": randn_tensor(
-                (batch_size, num_channels, num_frames, height, width), generator=self.generator, device=torch_device
-            ),
-            "timestep": torch.randint(0, 1000, size=(batch_size,), generator=self.generator).to(torch_device),
-            "encoder_hidden_states": randn_tensor(
-                (batch_size, sequence_length, text_embed_dim), generator=self.generator, device=torch_device
-            ),
-            "attention_mask": torch.ones((batch_size, sequence_length)).to(torch_device),
-            "fps": 30,
-            "condition_mask": torch.ones(batch_size, 1, num_frames, height, width).to(torch_device),
-            "padding_mask": torch.zeros(batch_size, 1, height, width).to(torch_device),
-        }
-
-
-class TestCosmosTransformerVideoToWorld(CosmosTransformerVideoToWorldTesterConfig, ModelTesterMixin):
-    """Core model tests for Cosmos Transformer (Video-to-World)."""
-
-
-class TestCosmosTransformerVideoToWorldMemory(CosmosTransformerVideoToWorldTesterConfig, MemoryTesterMixin):
-    """Memory optimization tests for Cosmos Transformer (Video-to-World)."""
-
-
-class TestCosmosTransformerVideoToWorldTraining(CosmosTransformerVideoToWorldTesterConfig, TrainingTesterMixin):
-    """Training tests for Cosmos Transformer (Video-to-World)."""
+        inputs_dict = self.dummy_input
+        return init_dict, inputs_dict
 
     def test_gradient_checkpointing_is_applied(self):
         expected_set = {"CosmosTransformer3DModel"}