Skip to content

fix(wan): un-shadow XLMRoberta's SelfAttention/AttentionBlock in wan_video_image_encoder #1688

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Anai-Guo wants to merge 1 commit into
base: main
Choose a base branch
from
+4 −4
Open

fix(wan): un-shadow XLMRoberta's SelfAttention/AttentionBlock in wan_video_image_encoder #1688

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions diffsynth/models/wan_video_image_encoder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
from .wan_video_dit import flash_attention


class SelfAttention(nn.Module):
class XLMRobertaSelfAttention(nn.Module):

def __init__(self, dim, num_heads, dropout=0.1, eps=1e-5):
assert dim % num_heads == 0
Expand Down Expand Up @@ -50,7 +50,7 @@ def forward(self, x, mask):
return x


class AttentionBlock(nn.Module):
class XLMRobertaAttentionBlock(nn.Module):

def __init__(self, dim, num_heads, post_norm, dropout=0.1, eps=1e-5):
super().__init__()
Expand All @@ -60,7 +60,7 @@ def __init__(self, dim, num_heads, post_norm, dropout=0.1, eps=1e-5):
self.eps = eps

# layers
self.attn = SelfAttention(dim, num_heads, dropout, eps)
self.attn = XLMRobertaSelfAttention(dim, num_heads, dropout, eps)
self.norm1 = nn.LayerNorm(dim, eps=eps)
self.ffn = nn.Sequential(
nn.Linear(dim, dim * 4), nn.GELU(), nn.Linear(dim * 4, dim),
Expand Down Expand Up @@ -112,7 +112,7 @@ def __init__(self,

# blocks
self.blocks = nn.ModuleList([
AttentionBlock(dim, num_heads, post_norm, dropout, eps)
XLMRobertaAttentionBlock(dim, num_heads, post_norm, dropout, eps)
for _ in range(num_layers)
])

Expand Down