Merge branch 'main' into toctree

Author: sayakpaul

.github/workflows/pr_tests_gpu.yml

@@ -13,6 +13,7 @@ on:
       - "src/diffusers/loaders/peft.py"
       - "tests/pipelines/test_pipelines_common.py"
       - "tests/models/test_modeling_common.py"
+      - "examples/**/*.py"
   workflow_dispatch:
 
 concurrency:

docs/source/en/_toctree.yml

@@ -370,6 +370,8 @@
         title: SanaTransformer2DModel
       - local: api/models/sd3_transformer2d
         title: SD3Transformer2DModel
+      - local: api/models/skyreels_v2_transformer_3d
+        title: SkyReelsV2Transformer3DModel
       - local: api/models/stable_audio_transformer
         title: StableAudioDiTModel
       - local: api/models/transformer2d
@@ -562,6 +564,8 @@
       title: Semantic Guidance
     - local: api/pipelines/shap_e
       title: Shap-E
+    - local: api/pipelines/skyreels_v2
+      title: SkyReels-V2
     - local: api/pipelines/stable_audio
       title: Stable Audio
     - local: api/pipelines/stable_cascade

docs/source/en/api/loaders/lora.md

@@ -26,6 +26,7 @@ LoRA is a fast and lightweight training method that inserts and trains a signifi
 - [`HunyuanVideoLoraLoaderMixin`] provides similar functions for [HunyuanVideo](https://huggingface.co/docs/diffusers/main/en/api/pipelines/hunyuan_video).
 - [`Lumina2LoraLoaderMixin`] provides similar functions for [Lumina2](https://huggingface.co/docs/diffusers/main/en/api/pipelines/lumina2).
 - [`WanLoraLoaderMixin`] provides similar functions for [Wan](https://huggingface.co/docs/diffusers/main/en/api/pipelines/wan).
+- [`SkyReelsV2LoraLoaderMixin`] provides similar functions for [SkyReels-V2](https://huggingface.co/docs/diffusers/main/en/api/pipelines/skyreels_v2).
 - [`CogView4LoraLoaderMixin`] provides similar functions for [CogView4](https://huggingface.co/docs/diffusers/main/en/api/pipelines/cogview4).
 - [`AmusedLoraLoaderMixin`] is for the [`AmusedPipeline`].
 - [`HiDreamImageLoraLoaderMixin`] provides similar functions for [HiDream Image](https://huggingface.co/docs/diffusers/main/en/api/pipelines/hidream)
@@ -92,6 +93,10 @@ To learn more about how to load LoRA weights, see the [LoRA](../../using-diffuse
 
 [[autodoc]] loaders.lora_pipeline.WanLoraLoaderMixin
 
+## SkyReelsV2LoraLoaderMixin
+
+[[autodoc]] loaders.lora_pipeline.SkyReelsV2LoraLoaderMixin
+
 ## AmusedLoraLoaderMixin
 
 [[autodoc]] loaders.lora_pipeline.AmusedLoraLoaderMixin
@@ -100,6 +105,6 @@ To learn more about how to load LoRA weights, see the [LoRA](../../using-diffuse
 
 [[autodoc]] loaders.lora_pipeline.HiDreamImageLoraLoaderMixin
 
-## WanLoraLoaderMixin
+## LoraBaseMixin
 
-[[autodoc]] loaders.lora_pipeline.WanLoraLoaderMixin
+[[autodoc]] loaders.lora_base.LoraBaseMixin

docs/source/en/api/models/skyreels_v2_transformer_3d.md

@@ -0,0 +1,30 @@
+<!-- Copyright 2024 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. -->
+
+# SkyReelsV2Transformer3DModel
+
+A Diffusion Transformer model for 3D video-like data was introduced in [SkyReels-V2](https://github.com/SkyworkAI/SkyReels-V2) by the Skywork AI.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import SkyReelsV2Transformer3DModel
+
+transformer = SkyReelsV2Transformer3DModel.from_pretrained("Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers", subfolder="transformer", torch_dtype=torch.bfloat16)
+```
+
+## SkyReelsV2Transformer3DModel
+
+[[autodoc]] SkyReelsV2Transformer3DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput

docs/source/en/api/pipelines/skyreels_v2.md

@@ -58,6 +58,7 @@
     compute_density_for_timestep_sampling,
     compute_loss_weighting_for_sd3,
     free_memory,
+    offload_models,
 )
 from diffusers.utils import (
     check_min_version,
@@ -1364,43 +1365,34 @@ def compute_text_embeddings(prompt, text_encoding_pipeline):
     # provided (i.e. the --instance_prompt is used for all images), we encode the instance prompt once to avoid
     # the redundant encoding.
     if not train_dataset.custom_instance_prompts:
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to(accelerator.device)
-        (
-            instance_prompt_hidden_states_t5,
-            instance_prompt_hidden_states_llama3,
-            instance_pooled_prompt_embeds,
-            _,
-            _,
-            _,
-        ) = compute_text_embeddings(args.instance_prompt, text_encoding_pipeline)
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to("cpu")
+        with offload_models(text_encoding_pipeline, device=accelerator.device, offload=args.offload):
+            (
+                instance_prompt_hidden_states_t5,
+                instance_prompt_hidden_states_llama3,
+                instance_pooled_prompt_embeds,
+                _,
+                _,
+                _,
+            ) = compute_text_embeddings(args.instance_prompt, text_encoding_pipeline)
 
     # Handle class prompt for prior-preservation.
     if args.with_prior_preservation:
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to(accelerator.device)
-        (class_prompt_hidden_states_t5, class_prompt_hidden_states_llama3, class_pooled_prompt_embeds, _, _, _) = (
-            compute_text_embeddings(args.class_prompt, text_encoding_pipeline)
-        )
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to("cpu")
+        with offload_models(text_encoding_pipeline, device=accelerator.device, offload=args.offload):
+            (class_prompt_hidden_states_t5, class_prompt_hidden_states_llama3, class_pooled_prompt_embeds, _, _, _) = (
+                compute_text_embeddings(args.class_prompt, text_encoding_pipeline)
+            )
 
     validation_embeddings = {}
     if args.validation_prompt is not None:
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to(accelerator.device)
-        (
-            validation_embeddings["prompt_embeds_t5"],
-            validation_embeddings["prompt_embeds_llama3"],
-            validation_embeddings["pooled_prompt_embeds"],
-            validation_embeddings["negative_prompt_embeds_t5"],
-            validation_embeddings["negative_prompt_embeds_llama3"],
-            validation_embeddings["negative_pooled_prompt_embeds"],
-        ) = compute_text_embeddings(args.validation_prompt, text_encoding_pipeline)
-        if args.offload:
-            text_encoding_pipeline = text_encoding_pipeline.to("cpu")
+        with offload_models(text_encoding_pipeline, device=accelerator.device, offload=args.offload):
+            (
+                validation_embeddings["prompt_embeds_t5"],
+                validation_embeddings["prompt_embeds_llama3"],
+                validation_embeddings["pooled_prompt_embeds"],
+                validation_embeddings["negative_prompt_embeds_t5"],
+                validation_embeddings["negative_prompt_embeds_llama3"],
+                validation_embeddings["negative_pooled_prompt_embeds"],
+            ) = compute_text_embeddings(args.validation_prompt, text_encoding_pipeline)
 
     # If custom instance prompts are NOT provided (i.e. the instance prompt is used for all images),
     # pack the statically computed variables appropriately here. This is so that we don't
@@ -1581,12 +1573,10 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
                 if args.cache_latents:
                     model_input = latents_cache[step].sample()
                 else:
-                    if args.offload:
-                        vae = vae.to(accelerator.device)
-                    pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
+                    with offload_models(vae, device=accelerator.device, offload=args.offload):
+                        pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
                     model_input = vae.encode(pixel_values).latent_dist.sample()
-                    if args.offload:
-                        vae = vae.to("cpu")
+
                 model_input = (model_input - vae_config_shift_factor) * vae_config_scaling_factor
                 model_input = model_input.to(dtype=weight_dtype)