[core] Allegro T2V (#9736)

* update

* refactor transformer part 1

* refactor part 2

* refactor part 3

* make style

* refactor part 4; modeling tests

* make style

* refactor part 5

* refactor part 6

* gradient checkpointing

* pipeline tests (broken atm)

* update

* add coauthor

Co-Authored-By: Huan Yang <[email protected]>

* refactor part 7

* add docs

* make style

* add coauthor

Co-Authored-By: YiYi Xu <[email protected]>

* make fix-copies

* undo unrelated change

* revert changes to embeddings, normalization, transformer

* refactor part 8

* make style

* refactor part 9

* make style

* fix

* apply suggestions from review

* Apply suggestions from code review

Co-authored-by: Steven Liu <[email protected]>

* update example

* remove attention mask for self-attention

* update

* copied from

* update

* update

---------

Co-authored-by: Huan Yang <[email protected]>
Co-authored-by: YiYi Xu <[email protected]>
Co-authored-by: Steven Liu <[email protected]>

Author: 4 people

docs/source/en/_toctree.yml

@@ -252,6 +252,8 @@
         title: SparseControlNetModel
       title: ControlNets
     - sections:
+      - local: api/models/allegro_transformer3d
+        title: AllegroTransformer3DModel
       - local: api/models/aura_flow_transformer2d
         title: AuraFlowTransformer2DModel
       - local: api/models/cogvideox_transformer3d
@@ -300,6 +302,8 @@
     - sections:
       - local: api/models/autoencoderkl
         title: AutoencoderKL
+      - local: api/models/autoencoderkl_allegro
+        title: AutoencoderKLAllegro
       - local: api/models/autoencoderkl_cogvideox
         title: AutoencoderKLCogVideoX
       - local: api/models/asymmetricautoencoderkl
@@ -318,6 +322,8 @@
     sections:
     - local: api/pipelines/overview
       title: Overview
+    - local: api/pipelines/allegro
+      title: Allegro
     - local: api/pipelines/amused
       title: aMUSEd
     - local: api/pipelines/animatediff

docs/source/en/api/models/allegro_transformer3d.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. -->
+
+# AllegroTransformer3DModel
+
+A Diffusion Transformer model for 3D data from [Allegro](https://github.com/rhymes-ai/Allegro) was introduced in [Allegro: Open the Black Box of Commercial-Level Video Generation Model](https://huggingface.co/papers/2410.15458) by RhymesAI.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import AllegroTransformer3DModel
+
+vae = AllegroTransformer3DModel.from_pretrained("rhymes-ai/Allegro", subfolder="transformer", torch_dtype=torch.bfloat16).to("cuda")
+```
+
+## AllegroTransformer3DModel
+
+[[autodoc]] AllegroTransformer3DModel
+
+## Transformer2DModelOutput
+
+[[autodoc]] models.modeling_outputs.Transformer2DModelOutput

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

@@ -0,0 +1,37 @@
+<!-- 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. -->
+
+# AutoencoderKLAllegro
+
+The 3D variational autoencoder (VAE) model with KL loss used in [Allegro](https://github.com/rhymes-ai/Allegro) was introduced in [Allegro: Open the Black Box of Commercial-Level Video Generation Model](https://huggingface.co/papers/2410.15458) by RhymesAI.
+
+The model can be loaded with the following code snippet.
+
+```python
+from diffusers import AutoencoderKLAllegro
+
+vae = AutoencoderKLCogVideoX.from_pretrained("rhymes-ai/Allegro", subfolder="vae", torch_dtype=torch.float32).to("cuda")
+```
+
+## AutoencoderKLAllegro
+
+[[autodoc]] AutoencoderKLAllegro
+    - decode
+    - encode
+    - all
+
+## AutoencoderKLOutput
+
+[[autodoc]] models.autoencoders.autoencoder_kl.AutoencoderKLOutput
+
+## DecoderOutput
+
+[[autodoc]] models.autoencoders.vae.DecoderOutput

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

@@ -0,0 +1,34 @@
+<!-- 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. -->
+
+# Allegro
+
+[Allegro: Open the Black Box of Commercial-Level Video Generation Model](https://huggingface.co/papers/2410.15458) from RhymesAI, by Yuan Zhou, Qiuyue Wang, Yuxuan Cai, Huan Yang.
+
+The abstract from the paper is:
+
+*Significant advancements have been made in the field of video generation, with the open-source community contributing a wealth of research papers and tools for training high-quality models. However, despite these efforts, the available information and resources remain insufficient for achieving commercial-level performance. In this report, we open the black box and introduce Allegro, an advanced video generation model that excels in both quality and temporal consistency. We also highlight the current limitations in the field and present a comprehensive methodology for training high-performance, commercial-level video generation models, addressing key aspects such as data, model architecture, training pipeline, and evaluation. Our user study shows that Allegro surpasses existing open-source models and most commercial models, ranking just behind Hailuo and Kling. Code: https://github.com/rhymes-ai/Allegro , Model: https://huggingface.co/rhymes-ai/Allegro , Gallery: https://rhymes.ai/allegro_gallery .*
+
+<Tip>
+
+Make sure to check out the Schedulers [guide](../../using-diffusers/schedulers.md) to learn how to explore the tradeoff between scheduler speed and quality, and see the [reuse components across pipelines](../../using-diffusers/loading.md#reuse-a-pipeline) section to learn how to efficiently load the same components into multiple pipelines.
+
+</Tip>
+
+## AllegroPipeline
+
+[[autodoc]] AllegroPipeline
+  - all
+  - __call__
+
+## AllegroPipelineOutput
+
+[[autodoc]] pipelines.allegro.pipeline_output.AllegroPipelineOutput

src/diffusers/__init__.py

@@ -77,9 +77,11 @@
 else:
     _import_structure["models"].extend(
         [
+            "AllegroTransformer3DModel",
             "AsymmetricAutoencoderKL",
             "AuraFlowTransformer2DModel",
             "AutoencoderKL",
+            "AutoencoderKLAllegro",
             "AutoencoderKLCogVideoX",
             "AutoencoderKLTemporalDecoder",
             "AutoencoderOobleck",
@@ -237,6 +239,7 @@
 else:
     _import_structure["pipelines"].extend(
         [
+            "AllegroPipeline",
             "AltDiffusionImg2ImgPipeline",
             "AltDiffusionPipeline",
             "AmusedImg2ImgPipeline",
@@ -556,9 +559,11 @@
         from .utils.dummy_pt_objects import *  # noqa F403
     else:
         from .models import (
+            AllegroTransformer3DModel,
             AsymmetricAutoencoderKL,
             AuraFlowTransformer2DModel,
             AutoencoderKL,
+            AutoencoderKLAllegro,
             AutoencoderKLCogVideoX,
             AutoencoderKLTemporalDecoder,
             AutoencoderOobleck,
@@ -697,6 +702,7 @@
         from .utils.dummy_torch_and_transformers_objects import *  # noqa F403
     else:
         from .pipelines import (
+            AllegroPipeline,
             AltDiffusionImg2ImgPipeline,
             AltDiffusionPipeline,
             AmusedImg2ImgPipeline,

src/diffusers/models/__init__.py

@@ -28,6 +28,7 @@
     _import_structure["adapter"] = ["MultiAdapter", "T2IAdapter"]
     _import_structure["autoencoders.autoencoder_asym_kl"] = ["AsymmetricAutoencoderKL"]
     _import_structure["autoencoders.autoencoder_kl"] = ["AutoencoderKL"]
+    _import_structure["autoencoders.autoencoder_kl_allegro"] = ["AutoencoderKLAllegro"]
     _import_structure["autoencoders.autoencoder_kl_cogvideox"] = ["AutoencoderKLCogVideoX"]
     _import_structure["autoencoders.autoencoder_kl_temporal_decoder"] = ["AutoencoderKLTemporalDecoder"]
     _import_structure["autoencoders.autoencoder_oobleck"] = ["AutoencoderOobleck"]
@@ -54,6 +55,7 @@
     _import_structure["transformers.stable_audio_transformer"] = ["StableAudioDiTModel"]
     _import_structure["transformers.t5_film_transformer"] = ["T5FilmDecoder"]
     _import_structure["transformers.transformer_2d"] = ["Transformer2DModel"]
+    _import_structure["transformers.transformer_allegro"] = ["AllegroTransformer3DModel"]
     _import_structure["transformers.transformer_cogview3plus"] = ["CogView3PlusTransformer2DModel"]
     _import_structure["transformers.transformer_flux"] = ["FluxTransformer2DModel"]
     _import_structure["transformers.transformer_sd3"] = ["SD3Transformer2DModel"]
@@ -81,6 +83,7 @@
         from .autoencoders import (
             AsymmetricAutoencoderKL,
             AutoencoderKL,
+            AutoencoderKLAllegro,
             AutoencoderKLCogVideoX,
             AutoencoderKLTemporalDecoder,
             AutoencoderOobleck,
@@ -97,6 +100,7 @@
         from .embeddings import ImageProjection
         from .modeling_utils import ModelMixin
         from .transformers import (
+            AllegroTransformer3DModel,
             AuraFlowTransformer2DModel,
             CogVideoXTransformer3DModel,
             CogView3PlusTransformer2DModel,

src/diffusers/models/attention_processor.py

@@ -1521,6 +1521,100 @@ def __call__(
         return hidden_states, encoder_hidden_states
 
 
+class AllegroAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
+    used in the Allegro model. It applies a normalization layer and rotary embedding on the query and key vector.
+    """
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError(
+                "AllegroAttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
+            )
+
+    def __call__(
+        self,
+        attn: Attention,
+        hidden_states: torch.Tensor,
+        encoder_hidden_states: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        temb: Optional[torch.Tensor] = None,
+        image_rotary_emb: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # Apply RoPE if needed
+        if image_rotary_emb is not None and not attn.is_cross_attention:
+            from .embeddings import apply_rotary_emb_allegro
+
+            query = apply_rotary_emb_allegro(query, image_rotary_emb[0], image_rotary_emb[1])
+            key = apply_rotary_emb_allegro(key, image_rotary_emb[0], image_rotary_emb[1])
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
 class AuraFlowAttnProcessor2_0:
     """Attention processor used typically in processing Aura Flow."""
 

src/diffusers/models/autoencoders/__init__.py

@@ -1,5 +1,6 @@
 from .autoencoder_asym_kl import AsymmetricAutoencoderKL
 from .autoencoder_kl import AutoencoderKL
+from .autoencoder_kl_allegro import AutoencoderKLAllegro
 from .autoencoder_kl_cogvideox import AutoencoderKLCogVideoX
 from .autoencoder_kl_temporal_decoder import AutoencoderKLTemporalDecoder
 from .autoencoder_oobleck import AutoencoderOobleck