Merge branch 'main' into main

Author: friedrich

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

@@ -50,7 +50,7 @@ from diffusers.utils import export_to_video
 pipeline_quant_config = PipelineQuantizationConfig(
   quant_backend="torchao",
   quant_kwargs={"quant_type": "int8wo"},
-  components_to_quantize=["transformer"]
+  components_to_quantize="transformer"
 )
 
 # fp8 layerwise weight-casting

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

@@ -54,7 +54,7 @@ pipeline_quant_config = PipelineQuantizationConfig(
       "bnb_4bit_quant_type": "nf4",
       "bnb_4bit_compute_dtype": torch.bfloat16
       },
-    components_to_quantize=["transformer"]
+    components_to_quantize="transformer"
 )
 
 pipeline = HunyuanVideoPipeline.from_pretrained(
@@ -91,7 +91,7 @@ pipeline_quant_config = PipelineQuantizationConfig(
       "bnb_4bit_quant_type": "nf4",
       "bnb_4bit_compute_dtype": torch.bfloat16
       },
-    components_to_quantize=["transformer"]
+    components_to_quantize="transformer"
 )
 
 pipeline = HunyuanVideoPipeline.from_pretrained(
@@ -139,7 +139,7 @@ export_to_video(video, "output.mp4", fps=15)
         "bnb_4bit_quant_type": "nf4",
         "bnb_4bit_compute_dtype": torch.bfloat16
         },
-      components_to_quantize=["transformer"]
+      components_to_quantize="transformer"
   )
 
   pipeline = HunyuanVideoPipeline.from_pretrained(

docs/source/en/optimization/memory.md

@@ -291,13 +291,53 @@ Group offloading moves groups of internal layers ([torch.nn.ModuleList](https://
 > [!WARNING]
 > Group offloading may not work with all models if the forward implementation contains weight-dependent device casting of inputs because it may clash with group offloading's device casting mechanism.
 
-Call [`~ModelMixin.enable_group_offload`] to enable it for standard Diffusers model components that inherit from [`ModelMixin`]. For other model components that don't inherit from [`ModelMixin`], such as a generic [torch.nn.Module](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), use [`~hooks.apply_group_offloading`] instead.
-
-The `offload_type` parameter can be set to `block_level` or `leaf_level`.
+Enable group offloading by configuring the `offload_type` parameter to `block_level` or `leaf_level`.
 
 - `block_level` offloads groups of layers based on the `num_blocks_per_group` parameter. For example, if `num_blocks_per_group=2` on a model with 40 layers, 2 layers are onloaded and offloaded at a time (20 total onloads/offloads). This drastically reduces memory requirements.
 - `leaf_level` offloads individual layers at the lowest level and is equivalent to [CPU offloading](#cpu-offloading). But it can be made faster if you use streams without giving up inference speed.
 
+Group offloading is supported for entire pipelines or individual models. Applying group offloading to the entire pipeline is the easiest option while selectively applying it to individual models gives users more flexibility to use different offloading techniques for different models.
+
+<hfoptions id="group-offloading">
+<hfoption id="pipeline">
+
+Call [`~DiffusionPipeline.enable_group_offload`] on a pipeline.
+
+```py
+import torch
+from diffusers import CogVideoXPipeline
+from diffusers.hooks import apply_group_offloading
+from diffusers.utils import export_to_video
+
+onload_device = torch.device("cuda")
+offload_device = torch.device("cpu")
+
+pipeline = CogVideoXPipeline.from_pretrained("THUDM/CogVideoX-5b", torch_dtype=torch.bfloat16)
+pipeline.enable_group_offload(
+    onload_device=onload_device,
+    offload_device=offload_device,
+    offload_type="leaf_level",
+    use_stream=True
+)
+
+prompt = (
+    "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. "
+    "The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other "
+    "pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, "
+    "casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. "
+    "The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical "
+    "atmosphere of this unique musical performance."
+)
+video = pipeline(prompt=prompt, guidance_scale=6, num_inference_steps=50).frames[0]
+print(f"Max memory reserved: {torch.cuda.max_memory_allocated() / 1024**3:.2f} GB")
+export_to_video(video, "output.mp4", fps=8)
+```
+
+</hfoption>
+<hfoption id="model">
+
+Call [`~ModelMixin.enable_group_offload`] on standard Diffusers model components that inherit from [`ModelMixin`]. For other model components that don't inherit from [`ModelMixin`], such as a generic [torch.nn.Module](https://pytorch.org/docs/stable/generated/torch.nn.Module.html), use [`~hooks.apply_group_offloading`] instead.
+
 ```py
 import torch
 from diffusers import CogVideoXPipeline
@@ -328,6 +368,9 @@ print(f"Max memory reserved: {torch.cuda.max_memory_allocated() / 1024**3:.2f} G
 export_to_video(video, "output.mp4", fps=8)
 ```
 
+</hfoption>
+</hfoptions>
+
 #### CUDA stream
 
 The `use_stream` parameter can be activated for CUDA devices that support asynchronous data transfer streams to reduce overall execution time compared to [CPU offloading](#cpu-offloading). It overlaps data transfer and computation by using layer prefetching. The next layer to be executed is loaded onto the GPU while the current layer is still being executed. It can increase CPU memory significantly so ensure you have 2x the amount of memory as the model size.

docs/source/en/quantization/overview.md

@@ -34,7 +34,9 @@ Initialize [`~quantizers.PipelineQuantizationConfig`] with the following paramet
 > [!TIP]
 > These `quant_kwargs` arguments are different for each backend. Refer to the [Quantization API](../api/quantization) docs to view the arguments for each backend.
 
-- `components_to_quantize` specifies which components of the pipeline to quantize. Typically, you should quantize the most compute intensive components like the transformer. The text encoder is another component to consider quantizing if a pipeline has more than one such as [`FluxPipeline`]. The example below quantizes the T5 text encoder in [`FluxPipeline`] while keeping the CLIP model intact.
+- `components_to_quantize` specifies which component(s) of the pipeline to quantize. Typically, you should quantize the most compute intensive components like the transformer. The text encoder is another component to consider quantizing if a pipeline has more than one such as [`FluxPipeline`]. The example below quantizes the T5 text encoder in [`FluxPipeline`] while keeping the CLIP model intact.
+
+   `components_to_quantize` accepts either a list for multiple models or a string for a single model.
 
 The example below loads the bitsandbytes backend with the following arguments from [`~quantizers.quantization_config.BitsAndBytesConfig`], `load_in_4bit`, `bnb_4bit_quant_type`, and `bnb_4bit_compute_dtype`.
 
@@ -62,6 +64,7 @@ pipe = DiffusionPipeline.from_pretrained(
 image = pipe("photo of a cute dog").images[0]
 ```
 
+
 ### Advanced quantization
 
 The `quant_mapping` argument provides more options for how to quantize each individual component in a pipeline, like combining different quantization backends.

docs/source/en/using-diffusers/text-img2vid.md

@@ -98,7 +98,7 @@ pipeline_quant_config = PipelineQuantizationConfig(
     "bnb_4bit_quant_type": "nf4",
     "bnb_4bit_compute_dtype": torch.bfloat16
     },
-  components_to_quantize=["transformer"]
+  components_to_quantize="transformer"
 )
 
 pipeline = HunyuanVideoPipeline.from_pretrained(

examples/community/pipeline_faithdiff_stable_diffusion_xl.py

@@ -1705,6 +1705,12 @@ def enable_vae_tiling(self):
         compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
         processing larger images.
         """
+        depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
+        deprecate(
+            "enable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_tiling()
         self.unet.denoise_encoder.enable_tiling()
 
@@ -1713,6 +1719,12 @@ def disable_vae_tiling(self):
         Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
+        deprecate(
+            "disable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_tiling()
         self.unet.denoise_encoder.disable_tiling()
 

examples/community/pipeline_flux_kontext_multiple_images.py

@@ -35,6 +35,7 @@
 from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
 from diffusers.utils import (
     USE_PEFT_BACKEND,
+    deprecate,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
@@ -643,6 +644,12 @@ def enable_vae_tiling(self):
         compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
         processing larger images.
         """
+        depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
+        deprecate(
+            "enable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_tiling()
 
     # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.disable_vae_tiling
@@ -651,6 +658,12 @@ def disable_vae_tiling(self):
         Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
+        deprecate(
+            "disable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_tiling()
 
     def preprocess_image(self, image: PipelineImageInput, _auto_resize: bool, multiple_of: int) -> torch.Tensor:

examples/community/pipeline_flux_rf_inversion.py

@@ -30,6 +30,7 @@
 from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
 from diffusers.utils import (
     USE_PEFT_BACKEND,
+    deprecate,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
@@ -526,13 +527,25 @@ def enable_vae_slicing(self):
         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.
         """
+        depr_message = f"Calling `enable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_slicing()`."
+        deprecate(
+            "enable_vae_slicing",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_slicing()
 
     def disable_vae_slicing(self):
         r"""
         Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_slicing()`."
+        deprecate(
+            "disable_vae_slicing",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_slicing()
 
     def enable_vae_tiling(self):
@@ -541,13 +554,25 @@ def enable_vae_tiling(self):
         compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
         processing larger images.
         """
+        depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
+        deprecate(
+            "enable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_tiling()
 
     def disable_vae_tiling(self):
         r"""
         Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
+        deprecate(
+            "disable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_tiling()
 
     def prepare_latents_inversion(

examples/community/pipeline_flux_semantic_guidance.py

@@ -35,6 +35,7 @@
 from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
 from diffusers.utils import (
     USE_PEFT_BACKEND,
+    deprecate,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
@@ -702,6 +703,12 @@ def enable_vae_tiling(self):
         compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
         processing larger images.
         """
+        depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
+        deprecate(
+            "enable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_tiling()
 
     # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.disable_vae_tiling
@@ -710,6 +717,12 @@ def disable_vae_tiling(self):
         Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
+        deprecate(
+            "disable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_tiling()
 
     # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_latents

examples/community/pipeline_flux_with_cfg.py

@@ -28,6 +28,7 @@
 from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
 from diffusers.utils import (
     USE_PEFT_BACKEND,
+    deprecate,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
@@ -503,13 +504,25 @@ def enable_vae_slicing(self):
         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.
         """
+        depr_message = f"Calling `enable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_slicing()`."
+        deprecate(
+            "enable_vae_slicing",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_slicing()
 
     def disable_vae_slicing(self):
         r"""
         Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_slicing()`."
+        deprecate(
+            "disable_vae_slicing",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_slicing()
 
     def enable_vae_tiling(self):
@@ -518,13 +531,25 @@ def enable_vae_tiling(self):
         compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
         processing larger images.
         """
+        depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
+        deprecate(
+            "enable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.enable_tiling()
 
     def disable_vae_tiling(self):
         r"""
         Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
         computing decoding in one step.
         """
+        depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
+        deprecate(
+            "disable_vae_tiling",
+            "0.40.0",
+            depr_message,
+        )
         self.vae.disable_tiling()
 
     def prepare_latents(