xinyun/ComfyUI
1
0
Fork 0
mirror of https://git.datalinker.icu/comfyanonymous/ComfyUI synced 2026-09-13 05:07:08 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use optimized attention

Browse Source
This commit is contained in:
City 2024-12-14 21:48:31 +01:00
parent e270142f9a
commit ebef4d9990
1 changed files with 19 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
comfy/ldm/pixart/blocks.py
View File

@ -1,7 +1,6 @@
# Based on:
# https://github.com/PixArt-alpha/PixArt-alpha [Apache 2.0 license]
# https://github.com/PixArt-alpha/PixArt-sigma [Apache 2.0 license]
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
@ -9,6 +8,7 @@ from einops import rearrange
from comfy import model_management
from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder, Mlp
from comfy.ldm.modules.attention import optimized_attention
if model_management.xformers_enabled():
import xformers.ops
@ -46,14 +46,14 @@ class MultiHeadCrossAttention(nn.Module):
kv = self.kv_linear(cond).view(1, -1, 2, self.num_heads, self.head_dim)
k, v = kv.unbind(2)
# TODO: figure out mask logic and use optimized_attention instead
# TODO: xformers needs separate mask logic here
if model_management.xformers_enabled():
attn_bias = None
if mask is not None:
attn_bias = block_diagonal_mask_from_seqlens([N] * B, mask)
x = xformers.ops.memory_efficient_attention(q, k, v, p=0, attn_bias=attn_bias)
else:
q, k, v = map(lambda t: t.permute(0, 2, 1, 3),(q, k, v),)
q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v),)
attn_mask = None
if mask is not None and len(mask) > 1:
# Create equivalent of xformer diagonal block mask, still only correct for square masks
@ -66,9 +66,11 @@ class MultiHeadCrossAttention(nn.Module):
attn_mask = torch.block_diag(attn_mask_template)
# create a mask on the diagonal for each mask in the batch
for n in range(B - 1):
for _ in range(B - 1):
attn_mask = torch.block_diag(attn_mask, attn_mask_template)
x = torch.nn.functional.scaled_dot_product_attention(q, k, v, dropout_p=0, attn_mask=attn_mask).permute(0, 2, 1, 3).contiguous()
x = optimized_attention(q, k, v, self.num_heads, mask=attn_mask, skip_reshape=True)
x = x.view(B, -1, C)
x = self.proj(x)
x = self.proj_drop(x)
@ -157,24 +159,19 @@ class AttentionKVCompress(nn.Module):
k = k.reshape(B, new_N, self.num_heads, C // self.num_heads).to(dtype)
v = v.reshape(B, new_N, self.num_heads, C // self.num_heads).to(dtype)
attn_bias = None
if mask is not None:
attn_bias = torch.zeros([B * self.num_heads, q.shape[1], k.shape[1]], dtype=q.dtype, device=q.device)
attn_bias.masked_fill_(mask.squeeze(1).repeat(self.num_heads, 1, 1) == 0, float('-inf'))
# Switch between torch / xformers attention
if model_management.xformers_enabled():
x = xformers.ops.memory_efficient_attention(
q, k, v,
p=0,
attn_bias=attn_bias
)
else:
q, k, v = map(lambda t: t.transpose(1, 2),(q, k, v),)
x = torch.nn.functional.scaled_dot_product_attention(
q, k, v,
dropout_p=0,
attn_mask=attn_bias
).transpose(1, 2).contiguous()
raise NotImplementedError("Attn mask logic not added for self attention")
# This is never called at the moment
# attn_bias = None
# if mask is not None:
# attn_bias = torch.zeros([B * self.num_heads, q.shape[1], k.shape[1]], dtype=q.dtype, device=q.device)
# attn_bias.masked_fill_(mask.squeeze(1).repeat(self.num_heads, 1, 1) == 0, float('-inf'))
# attention 2
q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v),)
x = optimized_attention(q, k, v, self.num_heads, mask=None, skip_reshape=True)
x = x.view(B, N, C)
x = self.proj(x)
return x