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

Revert "[cherry-pick][refactor]support gatingtopk operator generalization (#4050)" (#4352)

Browse Source 
This reverts commit c87a77e8b4.

it breaks ops e2e test

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-11-21 23:03:20 +08:00
committed by GitHub
parent 5ad0ccdc31
commit a2e4c3fe78
3 changed files with 69 additions and 74 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
vllm_ascend/ascend_forward_context.py
View File

@@ -96,7 +96,6 @@ def set_ascend_forward_context(
ep_size = (get_ep_group().world_size if
vllm_config.parallel_config.enable_expert_parallel else 1)
# fused_moe_state is used in torchair, it will be deleted along with torchair
is_deepseek_v3_r1 = hasattr(
vllm_config.model_config.hf_config, 'n_routed_experts'
) and vllm_config.model_config.hf_config.n_routed_experts == 256