EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Revert "[Bugfix] Fix Qwen2.5-Omni-7B accuarcy test (#4556)" (#4619)

Browse Source 
This reverts commit 71e9b379c8. It breaks vllm-ascend/Qwen3-30B-A3B-W8A8 test
This commit is contained in:
wangxiyuan
2025-12-02 13:15:47 +08:00
committed by GitHub
parent e18e3067a7
commit 6360eb1dea
2 changed files with 5 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm_ascend/ops/register_custom_ops.py
View File

@@ -173,9 +173,7 @@ def _maybe_prefetch_mlp_down_proj_impl(x_dependency: torch.Tensor) -> None:
except AssertionError:
return
prefetch_mlp_enabled = getattr(forward_context, 'prefetch_mlp_enabled',
False)
if not prefetch_mlp_enabled:
if not forward_context.prefetch_mlp_enabled:
return
forward_context.prefetch_mlp_down_proj = True
model_instance = forward_context.model_instance
@@ -204,9 +202,7 @@ def _maybe_wait_prefetch_done_impl(x: torch.Tensor) -> None:
except AssertionError:
return
prefetch_mlp_enabled = getattr(forward_context, 'prefetch_mlp_enabled',
False)
if not prefetch_mlp_enabled:
if not forward_context.prefetch_mlp_enabled:
return
if forward_context.prefetch_mlp_gate_up_proj or \
forward_context.prefetch_mlp_down_proj: