minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[diffusion] fix: enable torch.compile for UlyssesAttention (#18840)

Browse Source
This commit is contained in:
Ratish P
2026-02-15 15:54:27 +08:00
committed by GitHub
parent ad1bdb93df
commit 274bf6607a
1 changed files with 0 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/multimodal_gen/runtime/layers/attention/layer.py
-2
View File
@@ -78,7 +78,6 @@ class UlyssesAttention(nn.Module):
self.backend = attn_backend.get_enum() self.backend = attn_backend.get_enum()
self.dtype = dtype self.dtype = dtype
@torch.compiler.disable
def forward( def forward(
self, self,
q: torch.Tensor, q: torch.Tensor,
@@ -157,7 +156,6 @@ class UlyssesAttention(nn.Module):
class UlyssesAttention_VSA(UlyssesAttention): class UlyssesAttention_VSA(UlyssesAttention):
"""Distributed attention layer with VSA support.""" """Distributed attention layer with VSA support."""
@torch.compiler.disable
def forward( def forward(
self, self,
q: torch.Tensor, q: torch.Tensor,