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

[BugFix]Fix grouplist type of mc2. (#4049)

Browse Source 
### What this PR does / why we need it?
Fix accrucy problem of eplb because of PTA upgrade. This is a backport
of #4047

### How was this patch tested?
Mian:
    baseline:
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 87.50 |

   EPLB:

| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 87.50 |
- vLLM version: v0.11.0
- vLLM main:
83f478bb19

Signed-off-by: offline0806 <3337230449@qq.com>
Co-authored-by: offline0806 <3337230449@qq.com>
This commit is contained in:
offline893
2025-11-07 17:43:23 +08:00
committed by GitHub
parent 016337eaec
commit 8e72758645
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm_ascend/ops/moe/token_dispatcher.py
View File

@@ -212,7 +212,7 @@ class TokenDispatcherWithMC2(MoETokenDispatcher):
if shared_experts is not None:
shared_gate_up, _ = shared_experts.gate_up_proj(hidden_states)
self.shared_act = shared_experts.act_fn(shared_gate_up)
group_list_type = 1 if dynamic_eplb else 0
group_list_type = 0
return {
"group_list_type": group_list_type,
"hidden_states": expand_x,