[Test] Add nightly test for DeepSeek-V3.2-Exp #3908
Conversation
|
👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:
If CI fails, you can run linting and testing checks locally according Contributing and Testing. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds nightly tests for the DeepSeek-V3.2-Exp model. The changes include a new single-node e2e test, a multi-node test configuration, and updates to the multi-node test runner script. My review identifies a critical issue in the multi-node YAML configuration where an incorrect model version is served without the necessary quantization flag, which would invalidate the test results. Additionally, I've pointed out several high-severity issues related to maintainability and potential for confusion, such as hardcoded values and typos in scripts and test files.
zhangxinyuehfad
force-pushed
the
zxy_dpsk32
branch
2 times, most recently
from
be74728 to
da61f4f
Compare
|
This pull request has conflicts, please resolve those before we can evaluate the pull request. |
Signed-off-by: hfadzxy <[email protected]>
zhangxinyuehfad
force-pushed
the
zxy_dpsk32
branch
from
da61f4f to
7156cbf
Compare
Signed-off-by: hfadzxy <[email protected]>
zhangxinyuehfad
force-pushed
the
zxy_dpsk32
branch
from
7156cbf to
50d965b
Compare
| OMP_NUM_THREADS: 100 | ||
| HCCL_BUFFSIZE: 1024 | ||
| SERVER_PORT: 8080 | ||
| disaggregated_prefill: |
There was a problem hiding this comment.
I think you want to add multi-dp test cross 2 nodes for deepseekv3.2 right? so we should not pass disaggregated_prefill related config
| --data-parallel-rpc-port 13389 | ||
| --tensor-parallel-size 16 | ||
| --seed 1024 | ||
| --served-model-name deepseek_v3.2 |
There was a problem hiding this comment.
| --served-model-name deepseek_v3.2 |
There was a problem hiding this comment.
Since we use the model name instead of this short tag when checking if the service is running, passing this parameter will result in a timeout.
| --data-parallel-rpc-port 13389 | ||
| --tensor-parallel-size 16 | ||
| --seed 1024 | ||
| --served-model-name deepseek_v3.2 |
2 participants
What this PR does / why we need it?
Add nightly test for DeepSeek-V3.2-Exp
Does this PR introduce any user-facing change?
How was this patch tested?