[Fixbug] Fix soc_version for 310p

[zhangxinyuehfad]

What this PR does / why we need it?

1. fix soc_version for 310p
2. refactor _build_info and add ascend_soc_version(A2, A3, 310P) into _build_info
3. set default SOC_VERSION(ASCEND910B1, Ascend910_9392, ASCEND310P3) for ascend_soc_version

(EngineCore_0 pid=7454)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700] EngineCore failed to start.
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700] Traceback (most recent call last):
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/v1/engine/core.py", line 691, in run_engine_core
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     engine_core = EngineCoreProc(*args, **kwargs)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/v1/engine/core.py", line 492, in __init__
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     super().__init__(vllm_config, executor_class, log_stats,
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/v1/engine/core.py", line 89, in __init__
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     self._initialize_kv_caches(vllm_config)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/v1/engine/core.py", line 179, in _initialize_kv_caches
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     self.model_executor.determine_available_memory())
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/v1/executor/abstract.py", line 76, in determine_available_memory
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     output = self.collective_rpc("determine_available_memory")
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/executor/uniproc_executor.py", line 58, in collective_rpc
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     answer = run_method(self.driver_worker, method, args, kwargs)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-empty/vllm/utils/__init__.py", line 3007, in run_method
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     return func(*args, **kwargs)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]            ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-ascend/vllm_ascend/worker/worker_v1.py", line 161, in determine_available_memory
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     self.model_runner.profile_run()
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-ascend/vllm_ascend/worker/model_runner_v1.py", line 2163, in profile_run
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     hidden_states = self._dummy_run(self.max_num_tokens,
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/usr/local/python3.11.13/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     return func(*args, **kwargs)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]            ^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-ascend/vllm_ascend/worker/model_runner_v1.py", line 2015, in _dummy_run
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     moe_comm_method = self._select_moe_comm_method(num_tokens)
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]   File "/__w/vllm-benchmarks/vllm-benchmarks/vllm-ascend/vllm_ascend/worker/model_runner_v1.py", line 1633, in _select_moe_comm_method
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700]     raise ValueError(f"Unsupported soc_version: {soc_version}")
(EngineCore_0 pid=7454) ERROR 09-01 08:01:59 [core.py:700] ValueError: Unsupported soc_version: AscendSocVersion.UNDEFINED

Does this PR introduce any user-facing change?

Users can use 310p nomarlly

How was this patch tested?

• vLLM version: v0.10.2
• vLLM main: vllm-project/vllm@1a0a04d

[gemini-code-assist]

Code Review

This pull request correctly fixes a bug for Ascend 310p devices by adding support for its soc_version. The changes properly identify the new soc_version and configure the appropriate MoE communication method. I have one high-severity suggestion to improve maintainability by replacing a magic number with a named constant.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[wangxiyuan]

I don't like this way, how about refactor to build_info way totally?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.84%. Comparing base (2693196) to head (5fc0f77).
⚠️ Report is 24 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2676      +/-   ##
==========================================
+ Coverage   72.61%   73.84%   +1.22%     
==========================================
  Files         154      155       +1     
  Lines       21319    21338      +19     
==========================================
+ Hits        15480    15756     +276     
+ Misses       5839     5582     -257     

Flag	Coverage Δ
unittests	73.84% <100.00%> (+1.22%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[AlphaINF]

same issue! Want this branch merge!

[zhangxinyuehfad]

I don't like this way, how about refactor to build_info way totally?

It's impossible to import torch_npu and using get_soc_version() in the isolated environment, We need to get ascend_soc_version in the run phase like before。

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.

[github-actions]

This pull request has conflicts, please resolve those before we can evaluate the pull request.