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

[main] mlp weight prefetch in Qwen Dense Models (#2816)

Browse Source 
### What this PR does / why we need it?
This PR prefetchs the weight of mlp layers in Qwen Dense Models to
optimize the performance in Decode phase mainly.

### Does this PR introduce _any_ user-facing change?
 No.

### How was this patch tested?
CI passed with new added/existing test.

- vLLM version: main
- vLLM main:
a1213fae5f

Signed-off-by: rjg-lyh <1318825571@qq.com>
Co-authored-by: Shuming19 <313093131@qq.com>
This commit is contained in:
rjg-lyh
2025-09-11 21:20:09 +08:00
committed by GitHub
parent c3c2221503
commit 0005479b9c
17 changed files with 313 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/vllm_ascend_test.yaml vendored
View File

@@ -258,4 +258,4 @@ jobs:
VLLM_WORKER_MULTIPROC_METHOD: spawn
VLLM_USE_MODELSCOPE: True
run: |
pytest -sv tests/e2e/multicard/test_qwen3_moe.py::test_models_distributed_Qwen3_MOE_TP2_WITH_EP
pytest -sv tests/e2e/multicard/test_qwen3_moe.py::test_models_distributed_Qwen3_MOE_TP2_WITH_EP