Skip to content

Fix acc mismatch for GCU (Enflame) #330

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
liuyancong-enflame-tech wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Fix acc mismatch for GCU (Enflame) #330

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c5ab62b
change attention backend, improve speed and acc
liuyancong-enflame-tech Jul 28, 2026
a1ab318
Merge branch 'flagos-ai:main' into gcu_fix
liuyancong-enflame-tech Jul 28, 2026
74031c7
update attention backend
liuyancong-enflame-tech Jul 29, 2026
0af6ea3
Merge branch 'flagos-ai:main' into gcu_fix
liuyancong-enflame-tech Jul 30, 2026
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
12 changes: 4 additions & 8 deletions vllm_fl/dispatch/backends/vendor/gcu/gcu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,15 +73,11 @@ def rotary_embedding(
)

def attention_backend(self, use_mla: bool = False, use_sparse: bool = False) -> str:
from vllm.v1.attention.backends.registry import AttentionBackendEnum

if use_mla:
if use_sparse:
raise NotImplementedError("GCU does not support sparse attention yet")
raise NotImplementedError("GCU does not support MLA yet")

import flash_attn.vllm_flash_attn

sys.modules["vllm.vllm_flash_attn"] = flash_attn.vllm_flash_attn

return AttentionBackendEnum.FLASH_ATTN.get_path()
# GCU uses a standalone flash_attn backend (AttentionGCUBackend) that calls
# Enflame's native flash_attn_varlen_func directly, without depending on
# vllm upstream FlashAttentionBackend / FlashAttentionImpl.
return "vllm_fl.dispatch.backends.vendor.gcu.impl.attention.AttentionGCUBackend"
Loading
Loading