Use T5Tokenizer instead of MT5Tokenizer

Given that the `MT5Tokenizer` in `transformers` is just a "re-export" of
`T5Tokenizer` as per
https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/models/mt5/tokenization_mt5.py
)on latest available stable Transformers i.e., v4.57.3), this commit
updates the imports to point to `T5Tokenizer` instead, so that those
still work with Transformers v5.0.0rc0 onwards.

examples/community/pipeline_hunyuandit_differential_img2img.py

@@ -21,8 +21,8 @@ from transformers import (
     BertModel,
     BertTokenizer,
     CLIPImageProcessor,
-    MT5Tokenizer,
     T5EncoderModel,
+    T5Tokenizer,
 )
 
 from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback
@@ -260,7 +260,7 @@ class HunyuanDiTDifferentialImg2ImgPipeline(DiffusionPipeline):
             The HunyuanDiT model designed by Tencent Hunyuan.
         text_encoder_2 (`T5EncoderModel`):
             The mT5 embedder. Specifically, it is 't5-v1_1-xxl'.
-        tokenizer_2 (`MT5Tokenizer`):
+        tokenizer_2 (`T5Tokenizer`):
             The tokenizer for the mT5 embedder.
         scheduler ([`DDPMScheduler`]):
             A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.
@@ -295,7 +295,7 @@ class HunyuanDiTDifferentialImg2ImgPipeline(DiffusionPipeline):
         feature_extractor: CLIPImageProcessor,
         requires_safety_checker: bool = True,
         text_encoder_2=T5EncoderModel,
-        tokenizer_2=MT5Tokenizer,
+        tokenizer_2=T5Tokenizer,
     ):
         super().__init__()
 

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage.py

@@ -27,7 +27,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
+from .vae import DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -410,7 +410,7 @@ class HunyuanImageDecoder2D(nn.Module):
         return h
 
 
-class AutoencoderKLHunyuanImage(ModelMixin, AutoencoderMixin, ConfigMixin, FromOriginalModelMixin):
+class AutoencoderKLHunyuanImage(ModelMixin, ConfigMixin, FromOriginalModelMixin):
     r"""
     A VAE model for 2D images with spatial tiling support.
 
@@ -486,6 +486,27 @@ class AutoencoderKLHunyuanImage(ModelMixin, AutoencoderMixin, ConfigMixin, FromO
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
         self.tile_latent_min_size = self.tile_sample_min_size // self.config.spatial_compression_ratio
 
+    def disable_tiling(self) -> None:
+        r"""
+        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_tiling = False
+
+    def enable_slicing(self) -> None:
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.use_slicing = True
+
+    def disable_slicing(self) -> None:
+        r"""
+        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_slicing = False
+
     def _encode(self, x: torch.Tensor):
 
         batch_size, num_channels, height, width = x.shape

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanimage_refiner.py

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
+from .vae import DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -584,7 +584,7 @@ class HunyuanImageRefinerDecoder3D(nn.Module):
         return hidden_states
 
 
-class AutoencoderKLHunyuanImageRefiner(ModelMixin, AutoencoderMixin, ConfigMixin):
+class AutoencoderKLHunyuanImageRefiner(ModelMixin, ConfigMixin):
     r"""
     A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
     HunyuanImage-2.1 Refiner.
@@ -685,6 +685,27 @@ class AutoencoderKLHunyuanImageRefiner(ModelMixin, AutoencoderMixin, ConfigMixin
         self.tile_sample_stride_width = tile_sample_stride_width or self.tile_sample_stride_width
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
 
+    def disable_tiling(self) -> None:
+        r"""
+        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_tiling = False
+
+    def enable_slicing(self) -> None:
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.use_slicing = True
+
+    def disable_slicing(self) -> None:
+        r"""
+        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_slicing = False
+
     def _encode(self, x: torch.Tensor) -> torch.Tensor:
         _, _, _, height, width = x.shape
 

src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py

@@ -26,7 +26,7 @@ from ...utils.accelerate_utils import apply_forward_hook
 from ..activations import get_activation
 from ..modeling_outputs import AutoencoderKLOutput
 from ..modeling_utils import ModelMixin
-from .vae import AutoencoderMixin, DecoderOutput, DiagonalGaussianDistribution
+from .vae import DecoderOutput, DiagonalGaussianDistribution
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -625,7 +625,7 @@ class HunyuanVideo15Decoder3D(nn.Module):
         return hidden_states
 
 
-class AutoencoderKLHunyuanVideo15(ModelMixin, AutoencoderMixin, ConfigMixin):
+class AutoencoderKLHunyuanVideo15(ModelMixin, ConfigMixin):
     r"""
     A VAE model with KL loss for encoding videos into latents and decoding latent representations into videos. Used for
     HunyuanVideo-1.5.
@@ -723,6 +723,27 @@ class AutoencoderKLHunyuanVideo15(ModelMixin, AutoencoderMixin, ConfigMixin):
         self.tile_latent_min_width = tile_latent_min_width or self.tile_latent_min_width
         self.tile_overlap_factor = tile_overlap_factor or self.tile_overlap_factor
 
+    def disable_tiling(self) -> None:
+        r"""
+        Disable tiled VAE decoding. If `enable_tiling` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_tiling = False
+
+    def enable_slicing(self) -> None:
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.use_slicing = True
+
+    def disable_slicing(self) -> None:
+        r"""
+        Disable sliced VAE decoding. If `enable_slicing` was previously enabled, this method will go back to computing
+        decoding in one step.
+        """
+        self.use_slicing = False
+
     def _encode(self, x: torch.Tensor) -> torch.Tensor:
         _, _, _, height, width = x.shape
 

src/diffusers/pipelines/controlnet_hunyuandit/pipeline_hunyuandit_controlnet.py

@@ -17,7 +17,7 @@ from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
-from transformers import BertModel, BertTokenizer, CLIPImageProcessor, MT5Tokenizer, T5EncoderModel
+from transformers import BertModel, BertTokenizer, CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
 
@@ -185,7 +185,7 @@ class HunyuanDiTControlNetPipeline(DiffusionPipeline):
             The HunyuanDiT model designed by Tencent Hunyuan.
         text_encoder_2 (`T5EncoderModel`):
             The mT5 embedder. Specifically, it is 't5-v1_1-xxl'.
-        tokenizer_2 (`MT5Tokenizer`):
+        tokenizer_2 (`T5Tokenizer`):
             The tokenizer for the mT5 embedder.
         scheduler ([`DDPMScheduler`]):
             A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.
@@ -229,7 +229,7 @@ class HunyuanDiTControlNetPipeline(DiffusionPipeline):
             HunyuanDiT2DMultiControlNetModel,
         ],
         text_encoder_2: Optional[T5EncoderModel] = None,
-        tokenizer_2: Optional[MT5Tokenizer] = None,
+        tokenizer_2: Optional[T5Tokenizer] = None,
         requires_safety_checker: bool = True,
     ):
         super().__init__()

src/diffusers/pipelines/hunyuandit/pipeline_hunyuandit.py

@@ -17,7 +17,7 @@ from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
-from transformers import BertModel, BertTokenizer, CLIPImageProcessor, MT5Tokenizer, T5EncoderModel
+from transformers import BertModel, BertTokenizer, CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
 
@@ -169,7 +169,7 @@ class HunyuanDiTPipeline(DiffusionPipeline):
             The HunyuanDiT model designed by Tencent Hunyuan.
         text_encoder_2 (`T5EncoderModel`):
             The mT5 embedder. Specifically, it is 't5-v1_1-xxl'.
-        tokenizer_2 (`MT5Tokenizer`):
+        tokenizer_2 (`T5Tokenizer`):
             The tokenizer for the mT5 embedder.
         scheduler ([`DDPMScheduler`]):
             A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.
@@ -204,7 +204,7 @@ class HunyuanDiTPipeline(DiffusionPipeline):
         feature_extractor: CLIPImageProcessor,
         requires_safety_checker: bool = True,
         text_encoder_2: Optional[T5EncoderModel] = None,
-        tokenizer_2: Optional[MT5Tokenizer] = None,
+        tokenizer_2: Optional[T5Tokenizer] = None,
     ):
         super().__init__()
 

src/diffusers/pipelines/pag/pipeline_pag_hunyuandit.py

@@ -17,7 +17,7 @@ from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import numpy as np
 import torch
-from transformers import BertModel, BertTokenizer, CLIPImageProcessor, MT5Tokenizer, T5EncoderModel
+from transformers import BertModel, BertTokenizer, CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
 
@@ -173,7 +173,7 @@ class HunyuanDiTPAGPipeline(DiffusionPipeline, PAGMixin):
             The HunyuanDiT model designed by Tencent Hunyuan.
         text_encoder_2 (`T5EncoderModel`):
             The mT5 embedder. Specifically, it is 't5-v1_1-xxl'.
-        tokenizer_2 (`MT5Tokenizer`):
+        tokenizer_2 (`T5Tokenizer`):
             The tokenizer for the mT5 embedder.
         scheduler ([`DDPMScheduler`]):
             A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.
@@ -208,7 +208,7 @@ class HunyuanDiTPAGPipeline(DiffusionPipeline, PAGMixin):
         feature_extractor: Optional[CLIPImageProcessor] = None,
         requires_safety_checker: bool = True,
         text_encoder_2: Optional[T5EncoderModel] = None,
-        tokenizer_2: Optional[MT5Tokenizer] = None,
+        tokenizer_2: Optional[T5Tokenizer] = None,
         pag_applied_layers: Union[str, List[str]] = "blocks.1",  # "blocks.16.attn1", "blocks.16", "16", 16
     ):
         super().__init__()