Skip to content

Refine test cases - #238

Merged
jiangkuaixue123 merged 1 commit into
vllm-project:mainfrom
yujuancao07:ci
Aug 13, 2026
Merged

Refine test cases#238
jiangkuaixue123 merged 1 commit into
vllm-project:mainfrom
yujuancao07:ci

Conversation

@yujuancao07

@yujuancao07 yujuancao07 commented Aug 13, 2026

Copy link
Copy Markdown
Contributor

Purpose

This PR extends the DeepSeek-V2-Lite E2E coverage with two opt-in four-device cases while keeping the existing three-device PR gate unchanged.

  • Add afd-graph-dbo-2a2f for GPU and NPU:
    • Attention: DP=2, TP=1
    • FFN: DP=2, TP=1, EP=2
    • Runs graph + DBO with GSM8K-7
  • Add afd-eager-async-cam for NPU:
    • Uses CAMAsyncAFDConnector
    • Attention: DP=1, TP=2
    • FFN: DP=2, TP=1, EP=2
    • Runs in eager mode without graph or DBO
    • Sends one accounting prompt and validates a non-empty 32-token completion
  • Extend the shared E2E runner to support the new topologies and completion-based smoke test.
  • Update the E2E README, design document, and run-e2e skill.

Test Plan

afd-graph-dbo-2a2f

export AFD_E2E_DEVICES=0,1,2,3
python -m pytest -q -s \
  "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-graph-dbo-2a2f]"

afd-eager-async-cam

python -m pytest -q -s \
  "tests/e2e/models/deepseek_v2_lite/test_deepseek_v2_lite.py::test_deepseek_v2_lite[afd-graph-dbo]"

Signed-off-by: yujuancao07 <yujuancao07@gmail.com>
@jiangkuaixue123
jiangkuaixue123 merged commit fe79d23 into vllm-project:main Aug 13, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants