PAG variant for HunyuanDiT, PAG refactor (#8936)

* copy hunyuandit pipeline

* pag variant of hunyuan dit

* add tests

* update docs

* make style

* make fix-copies

* Update src/diffusers/pipelines/pag/pag_utils.py

* remove incorrect copied from

* remove pag hunyuan attn procs to resolve conflicts

* add pag attn procs again

* new implementation for pag_utils

* revert pag changes

* add pag refactor back; update pixart sigma

* update pixart pag tests

* apply suggestions from review

Co-Authored-By: yixu310@gmail.com

* make style

* update docs, fix tests

* fix tests

* fix test_components_function since list not accepted as valid __init__ param

* apply patch to fix broken tests

Co-Authored-By: Sayak Paul <spsayakpaul@gmail.com>

* make style

* fix hunyuan tests

---------

Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

Author: a-r-r-o-w

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

@@ -20,11 +20,29 @@ The abstract from the paper is:
 
 *Recent studies have demonstrated that diffusion models are capable of generating high-quality samples, but their quality heavily depends on sampling guidance techniques, such as classifier guidance (CG) and classifier-free guidance (CFG). These techniques are often not applicable in unconditional generation or in various downstream tasks such as image restoration. In this paper, we propose a novel sampling guidance, called Perturbed-Attention Guidance (PAG), which improves diffusion sample quality across both unconditional and conditional settings, achieving this without requiring additional training or the integration of external modules. PAG is designed to progressively enhance the structure of samples throughout the denoising process. It involves generating intermediate samples with degraded structure by substituting selected self-attention maps in diffusion U-Net with an identity matrix, by considering the self-attention mechanisms' ability to capture structural information, and guiding the denoising process away from these degraded samples. In both ADM and Stable Diffusion, PAG surprisingly improves sample quality in conditional and even unconditional scenarios. Moreover, PAG significantly improves the baseline performance in various downstream tasks where existing guidances such as CG or CFG cannot be fully utilized, including ControlNet with empty prompts and image restoration such as inpainting and deblurring.*
 
+PAG can be used by specifying the `pag_applied_layers` as a parameter when instantiating a PAG pipeline. It can be a single string or a list of strings. Each string can be a unique layer identifier or a regular expression to identify one or more layers. 
+
+- Full identifier as a normal string: `down_blocks.2.attentions.0.transformer_blocks.0.attn1.processor`
+- Full identifier as a RegEx: `down_blocks.2.(attentions|motion_modules).0.transformer_blocks.0.attn1.processor`
+- Partial identifier as a RegEx: `down_blocks.2`, or `attn1`
+- List of identifiers (can be combo of strings and ReGex): `["blocks.1", "blocks.(14|20)", r"down_blocks\.(2,3)"]`
+
+<Tip warning={true}>
+
+Since RegEx is supported as a way for matching layer identifiers, it is crucial to use it correctly otherwise there might be unexpected behaviour. The recommended way to use PAG is by specifying layers as `blocks.{layer_index}` and `blocks.({layer_index_1|layer_index_2|...})`. Using it in any other way, while doable, may bypass our basic validation checks and give you unexpected results.
+
+</Tip>
+
 ## AnimateDiffPAGPipeline
 [[autodoc]] AnimateDiffPAGPipeline
   - all
   - __call__
 
+## HunyuanDiTPAGPipeline
+[[autodoc]] HunyuanDiTPAGPipeline
+  - all
+  - __call__
+
 ## StableDiffusionPAGPipeline
 [[autodoc]] StableDiffusionPAGPipeline
 	- all
@@ -59,4 +77,4 @@ The abstract from the paper is:
 ## PixArtSigmaPAGPipeline
 [[autodoc]] PixArtSigmaPAGPipeline
 	- all
-	- __call__
+	- __call__

src/diffusers/__init__.py

@@ -252,6 +252,7 @@
             "CycleDiffusionPipeline",
             "FluxPipeline",
             "HunyuanDiTControlNetPipeline",
+            "HunyuanDiTPAGPipeline",
             "HunyuanDiTPipeline",
             "I2VGenXLPipeline",
             "IFImg2ImgPipeline",
@@ -675,6 +676,7 @@
             CycleDiffusionPipeline,
             FluxPipeline,
             HunyuanDiTControlNetPipeline,
+            HunyuanDiTPAGPipeline,
             HunyuanDiTPipeline,
             I2VGenXLPipeline,
             IFImg2ImgPipeline,

src/diffusers/models/attention_processor.py

@@ -2147,6 +2147,253 @@ def __call__(
         return hidden_states
 
 
+class PAGHunyuanAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
+    used in the HunyuanDiT model. It applies a normalization layer and rotary embedding on query and key vector. This
+    variant of the processor employs [Pertubed Attention Guidance](https://arxiv.org/abs/2403.17377).
+    """
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError(
+                "PAGHunyuanAttnProcessor2_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:
+        from .embeddings import apply_rotary_emb
+
+        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)
+
+        # chunk
+        hidden_states_org, hidden_states_ptb = hidden_states.chunk(2)
+
+        # 1. Original Path
+        batch_size, sequence_length, _ = (
+            hidden_states_org.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_org = attn.group_norm(hidden_states_org.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states_org)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states_org
+        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)
+
+        if attn.norm_q is not None:
+            query = attn.norm_q(query)
+        if attn.norm_k is not None:
+            key = attn.norm_k(key)
+
+        # Apply RoPE if needed
+        if image_rotary_emb is not None:
+            query = apply_rotary_emb(query, image_rotary_emb)
+            if not attn.is_cross_attention:
+                key = apply_rotary_emb(key, image_rotary_emb)
+
+        # 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_org = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states_org = hidden_states_org.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states_org = hidden_states_org.to(query.dtype)
+
+        # linear proj
+        hidden_states_org = attn.to_out[0](hidden_states_org)
+        # dropout
+        hidden_states_org = attn.to_out[1](hidden_states_org)
+
+        if input_ndim == 4:
+            hidden_states_org = hidden_states_org.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        # 2. Perturbed Path
+        if attn.group_norm is not None:
+            hidden_states_ptb = attn.group_norm(hidden_states_ptb.transpose(1, 2)).transpose(1, 2)
+
+        hidden_states_ptb = attn.to_v(hidden_states_ptb)
+        hidden_states_ptb = hidden_states_ptb.to(query.dtype)
+
+        # linear proj
+        hidden_states_ptb = attn.to_out[0](hidden_states_ptb)
+        # dropout
+        hidden_states_ptb = attn.to_out[1](hidden_states_ptb)
+
+        if input_ndim == 4:
+            hidden_states_ptb = hidden_states_ptb.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        # cat
+        hidden_states = torch.cat([hidden_states_org, hidden_states_ptb])
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class PAGCFGHunyuanAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
+    used in the HunyuanDiT model. It applies a normalization layer and rotary embedding on query and key vector. This
+    variant of the processor employs [Pertubed Attention Guidance](https://arxiv.org/abs/2403.17377).
+    """
+
+    def __init__(self):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError(
+                "PAGCFGHunyuanAttnProcessor2_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:
+        from .embeddings import apply_rotary_emb
+
+        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)
+
+        # chunk
+        hidden_states_uncond, hidden_states_org, hidden_states_ptb = hidden_states.chunk(3)
+        hidden_states_org = torch.cat([hidden_states_uncond, hidden_states_org])
+
+        # 1. Original Path
+        batch_size, sequence_length, _ = (
+            hidden_states_org.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_org = attn.group_norm(hidden_states_org.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states_org)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states_org
+        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)
+
+        if attn.norm_q is not None:
+            query = attn.norm_q(query)
+        if attn.norm_k is not None:
+            key = attn.norm_k(key)
+
+        # Apply RoPE if needed
+        if image_rotary_emb is not None:
+            query = apply_rotary_emb(query, image_rotary_emb)
+            if not attn.is_cross_attention:
+                key = apply_rotary_emb(key, image_rotary_emb)
+
+        # 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_org = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states_org = hidden_states_org.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states_org = hidden_states_org.to(query.dtype)
+
+        # linear proj
+        hidden_states_org = attn.to_out[0](hidden_states_org)
+        # dropout
+        hidden_states_org = attn.to_out[1](hidden_states_org)
+
+        if input_ndim == 4:
+            hidden_states_org = hidden_states_org.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        # 2. Perturbed Path
+        if attn.group_norm is not None:
+            hidden_states_ptb = attn.group_norm(hidden_states_ptb.transpose(1, 2)).transpose(1, 2)
+
+        hidden_states_ptb = attn.to_v(hidden_states_ptb)
+        hidden_states_ptb = hidden_states_ptb.to(query.dtype)
+
+        # linear proj
+        hidden_states_ptb = attn.to_out[0](hidden_states_ptb)
+        # dropout
+        hidden_states_ptb = attn.to_out[1](hidden_states_ptb)
+
+        if input_ndim == 4:
+            hidden_states_ptb = hidden_states_ptb.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        # cat
+        hidden_states = torch.cat([hidden_states_org, hidden_states_ptb])
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
 class LuminaAttnProcessor2_0:
     r"""
     Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0). This is
@@ -3468,4 +3715,6 @@ def __init__(self):
     CustomDiffusionAttnProcessor2_0,
     PAGCFGIdentitySelfAttnProcessor2_0,
     PAGIdentitySelfAttnProcessor2_0,
+    PAGCFGHunyuanAttnProcessor2_0,
+    PAGHunyuanAttnProcessor2_0,
 ]

src/diffusers/pipelines/__init__.py

@@ -145,6 +145,7 @@
     _import_structure["pag"].extend(
         [
             "AnimateDiffPAGPipeline",
+            "HunyuanDiTPAGPipeline",
             "StableDiffusionPAGPipeline",
             "StableDiffusionControlNetPAGPipeline",
             "StableDiffusionXLPAGPipeline",
@@ -532,6 +533,7 @@
         from .musicldm import MusicLDMPipeline
         from .pag import (
             AnimateDiffPAGPipeline,
+            HunyuanDiTPAGPipeline,
             PixArtSigmaPAGPipeline,
             StableDiffusionControlNetPAGPipeline,
             StableDiffusionPAGPipeline,

src/diffusers/pipelines/auto_pipeline.py

@@ -50,6 +50,7 @@
 from .kolors import KolorsImg2ImgPipeline, KolorsPipeline
 from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
 from .pag import (
+    HunyuanDiTPAGPipeline,
     PixArtSigmaPAGPipeline,
     StableDiffusionControlNetPAGPipeline,
     StableDiffusionPAGPipeline,
@@ -85,6 +86,7 @@
         ("stable-diffusion-3", StableDiffusion3Pipeline),
         ("if", IFPipeline),
         ("hunyuan", HunyuanDiTPipeline),
+        ("hunyuan-pag", HunyuanDiTPAGPipeline),
         ("kandinsky", KandinskyCombinedPipeline),
         ("kandinsky22", KandinskyV22CombinedPipeline),
         ("kandinsky3", Kandinsky3Pipeline),

src/diffusers/pipelines/pag/__init__.py

@@ -24,6 +24,7 @@
 else:
     _import_structure["pipeline_pag_controlnet_sd"] = ["StableDiffusionControlNetPAGPipeline"]
     _import_structure["pipeline_pag_controlnet_sd_xl"] = ["StableDiffusionXLControlNetPAGPipeline"]
+    _import_structure["pipeline_pag_hunyuandit"] = ["HunyuanDiTPAGPipeline"]
     _import_structure["pipeline_pag_pixart_sigma"] = ["PixArtSigmaPAGPipeline"]
     _import_structure["pipeline_pag_sd"] = ["StableDiffusionPAGPipeline"]
     _import_structure["pipeline_pag_sd_animatediff"] = ["AnimateDiffPAGPipeline"]
@@ -41,6 +42,7 @@
     else:
         from .pipeline_pag_controlnet_sd import StableDiffusionControlNetPAGPipeline
         from .pipeline_pag_controlnet_sd_xl import StableDiffusionXLControlNetPAGPipeline
+        from .pipeline_pag_hunyuandit import HunyuanDiTPAGPipeline
         from .pipeline_pag_pixart_sigma import PixArtSigmaPAGPipeline
         from .pipeline_pag_sd import StableDiffusionPAGPipeline
         from .pipeline_pag_sd_animatediff import AnimateDiffPAGPipeline