andreas/ComfyUI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Enable Runtime Selection of Attention Functions (#9639)

Browse Source 
* Looking into a @wrap_attn decorator to look for 'optimized_attention_override' entry in transformer_options

* Created logging code for this branch so that it can be used to track down all the code paths where transformer_options would need to be added

* Fix memory usage issue with inspect

* Made WAN attention receive transformer_options, test node added to wan to test out attention override later

* Added **kwargs to all attention functions so transformer_options could potentially be passed through

* Make sure wrap_attn doesn't make itself recurse infinitely, attempt to load SageAttention and FlashAttention if not enabled so that they can be marked as available or not, create registry for available attention

* Turn off attention logging for now, make AttentionOverrideTestNode have a dropdown with available attention (this is a test node only)

* Make flux work with optimized_attention_override

* Add logs to verify optimized_attention_override is passed all the way into attention function

* Make Qwen work with optimized_attention_override

* Made hidream work with optimized_attention_override

* Made wan patches_replace work with optimized_attention_override

* Made SD3 work with optimized_attention_override

* Made HunyuanVideo work with optimized_attention_override

* Made Mochi work with optimized_attention_override

* Made LTX work with optimized_attention_override

* Made StableAudio work with optimized_attention_override

* Made optimized_attention_override work with ACE Step

* Made Hunyuan3D work with optimized_attention_override

* Make CosmosPredict2 work with optimized_attention_override

* Made CosmosVideo work with optimized_attention_override

* Made Omnigen 2 work with optimized_attention_override

* Made StableCascade work with optimized_attention_override

* Made AuraFlow work with optimized_attention_override

* Made Lumina work with optimized_attention_override

* Made Chroma work with optimized_attention_override

* Made SVD work with optimized_attention_override

* Fix WanI2VCrossAttention so that it expects to receive transformer_options

* Fixed Wan2.1 Fun Camera transformer_options passthrough

* Fixed WAN 2.1 VACE transformer_options passthrough

* Add optimized to get_attention_function

* Disable attention logs for now

* Remove attention logging code

* Remove _register_core_attention_functions, as we wouldn't want someone to call that, just in case

* Satisfy ruff

* Remove AttentionOverrideTest node, that's something to cook up for later
This commit is contained in:
Jedrzej Kosinski
2025-09-12 15:07:38 -07:00
committed by GitHub
parent b149e2e1e3
commit d7f40442f9
26 changed files with 316 additions and 179 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
comfy/ldm/omnigen/omnigen2.py
View File

@@ -120,7 +120,7 @@ class Attention(nn.Module):
nn.Dropout(0.0)
)
def forward(self, hidden_states: torch.Tensor, encoder_hidden_states: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, image_rotary_emb: Optional[torch.Tensor] = None) -> torch.Tensor:
def forward(self, hidden_states: torch.Tensor, encoder_hidden_states: torch.Tensor, attention_mask: Optional[torch.Tensor] = None, image_rotary_emb: Optional[torch.Tensor] = None, transformer_options={}) -> torch.Tensor:
batch_size, sequence_length, _ = hidden_states.shape
query = self.to_q(hidden_states)
@@ -146,7 +146,7 @@ class Attention(nn.Module):
key = key.repeat_interleave(self.heads // self.kv_heads, dim=1)
value = value.repeat_interleave(self.heads // self.kv_heads, dim=1)
hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True)
hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options)
hidden_states = self.to_out[0](hidden_states)
return hidden_states
@@ -182,16 +182,16 @@ class OmniGen2TransformerBlock(nn.Module):
self.norm2 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
self.ffn_norm2 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
def forward(self, hidden_states: torch.Tensor, attention_mask: torch.Tensor, image_rotary_emb: torch.Tensor, temb: Optional[torch.Tensor] = None) -> torch.Tensor:
def forward(self, hidden_states: torch.Tensor, attention_mask: torch.Tensor, image_rotary_emb: torch.Tensor, temb: Optional[torch.Tensor] = None, transformer_options={}) -> torch.Tensor:
if self.modulation:
norm_hidden_states, gate_msa, scale_mlp, gate_mlp = self.norm1(hidden_states, temb)
attn_output = self.attn(norm_hidden_states, norm_hidden_states, attention_mask, image_rotary_emb)
attn_output = self.attn(norm_hidden_states, norm_hidden_states, attention_mask, image_rotary_emb, transformer_options=transformer_options)
hidden_states = hidden_states + gate_msa.unsqueeze(1).tanh() * self.norm2(attn_output)
mlp_output = self.feed_forward(self.ffn_norm1(hidden_states) * (1 + scale_mlp.unsqueeze(1)))
hidden_states = hidden_states + gate_mlp.unsqueeze(1).tanh() * self.ffn_norm2(mlp_output)
else:
norm_hidden_states = self.norm1(hidden_states)
attn_output = self.attn(norm_hidden_states, norm_hidden_states, attention_mask, image_rotary_emb)
attn_output = self.attn(norm_hidden_states, norm_hidden_states, attention_mask, image_rotary_emb, transformer_options=transformer_options)
hidden_states = hidden_states + self.norm2(attn_output)
mlp_output = self.feed_forward(self.ffn_norm1(hidden_states))
hidden_states = hidden_states + self.ffn_norm2(mlp_output)
@@ -390,7 +390,7 @@ class OmniGen2Transformer2DModel(nn.Module):
ref_img_sizes, img_sizes,
)
def img_patch_embed_and_refine(self, hidden_states, ref_image_hidden_states, padded_img_mask, padded_ref_img_mask, noise_rotary_emb, ref_img_rotary_emb, l_effective_ref_img_len, l_effective_img_len, temb):
def img_patch_embed_and_refine(self, hidden_states, ref_image_hidden_states, padded_img_mask, padded_ref_img_mask, noise_rotary_emb, ref_img_rotary_emb, l_effective_ref_img_len, l_effective_img_len, temb, transformer_options={}):
batch_size = len(hidden_states)
hidden_states = self.x_embedder(hidden_states)
@@ -405,17 +405,17 @@ class OmniGen2Transformer2DModel(nn.Module):
shift += ref_img_len
for layer in self.noise_refiner:
hidden_states = layer(hidden_states, padded_img_mask, noise_rotary_emb, temb)
hidden_states = layer(hidden_states, padded_img_mask, noise_rotary_emb, temb, transformer_options=transformer_options)
if ref_image_hidden_states is not None:
for layer in self.ref_image_refiner:
ref_image_hidden_states = layer(ref_image_hidden_states, padded_ref_img_mask, ref_img_rotary_emb, temb)
ref_image_hidden_states = layer(ref_image_hidden_states, padded_ref_img_mask, ref_img_rotary_emb, temb, transformer_options=transformer_options)
hidden_states = torch.cat([ref_image_hidden_states, hidden_states], dim=1)
return hidden_states
def forward(self, x, timesteps, context, num_tokens, ref_latents=None, attention_mask=None, **kwargs):
def forward(self, x, timesteps, context, num_tokens, ref_latents=None, attention_mask=None, transformer_options={}, **kwargs):
B, C, H, W = x.shape
hidden_states = comfy.ldm.common_dit.pad_to_patch_size(x, (self.patch_size, self.patch_size))
_, _, H_padded, W_padded = hidden_states.shape
@@ -444,7 +444,7 @@ class OmniGen2Transformer2DModel(nn.Module):
)
for layer in self.context_refiner:
text_hidden_states = layer(text_hidden_states, text_attention_mask, context_rotary_emb)
text_hidden_states = layer(text_hidden_states, text_attention_mask, context_rotary_emb, transformer_options=transformer_options)
img_len = hidden_states.shape[1]
combined_img_hidden_states = self.img_patch_embed_and_refine(
@@ -453,13 +453,14 @@ class OmniGen2Transformer2DModel(nn.Module):
noise_rotary_emb, ref_img_rotary_emb,
l_effective_ref_img_len, l_effective_img_len,
temb,
transformer_options=transformer_options,
)
hidden_states = torch.cat([text_hidden_states, combined_img_hidden_states], dim=1)
attention_mask = None
for layer in self.layers:
hidden_states = layer(hidden_states, attention_mask, rotary_emb, temb)
hidden_states = layer(hidden_states, attention_mask, rotary_emb, temb, transformer_options=transformer_options)
hidden_states = self.norm_out(hidden_states, temb)