[Test] Add ut test for torchair #4287

ForBetterCodeNine · 2025-11-20T01:27:11Z

What this PR does / why we need it?

The current community lacks unit tests (UT) for files such as torchair_worker, mtp_proposer, and model_runner. Therefore, UT coverage for these files needs to be added.

Does this PR introduce any user-facing change?

NO

How was this patch tested?

vLLM version: v0.11.0
vLLM main: vllm-project/vllm@2918c1b

Signed-off-by: CodeNine-CJ <[email protected]>

github-actions · 2025-11-20T01:27:20Z

👋 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.

gemini-code-assist

Code Review

This pull request adds unit tests for several torchair components. While adding tests is a great initiative, the new test files for NPUTorchairModelRunner and TorchairMtpProposer contain several critical syntax and logical errors that will prevent them from running. I've left detailed comments on these issues. Please address them to ensure the new tests are effective.

gemini-code-assist · 2025-11-20T01:29:11Z

tests/ut/torchair/test_torchair_model_runner.py

+from vllm.config import VllmConfig
+
+
+class TestNPUTorchairModelRunner(PytestBase):


PytestBase is not defined, which will cause a NameError at runtime. You need to import it, likely from tests.ut.base.

gemini-code-assist · 2025-11-20T01:29:12Z

tests/ut/torchair/test_torchair_model_runner.py

+        device = torch.device("npu:0")
+
+        ascend_config = MagicMock()
+        ascend_config = enable_shared_expert_dp = False


This line incorrectly reassigns ascend_config to the boolean value False, overwriting the MagicMock object created on the previous line. This will cause an AttributeError on the next line. It should be an attribute assignment.

Suggested change

ascend_config = enable_shared_expert_dp = False

ascend_config.enable_shared_expert_dp = False

gemini-code-assist · 2025-11-20T01:29:12Z

tests/ut/torchair/test_torchair_mtp_proposer.py

+sys.modules[__name__].vllm_version_is = vllm_version_is
+
+
+class TestTorchairMtpProposer(PytestBase):


PytestBase is not defined, which will cause a NameError at runtime. You need to import it, likely from tests.ut.base.

gemini-code-assist · 2025-11-20T01:29:12Z

tests/ut/torchair/test_torchair_mtp_proposer.py

+    def test_init(self, setup_torchair_mtp_proposer):
+        proposer, _, _, = setup_torchair_mtp_proposer
+
+        assert isinstance(proposer, setup_torchair_mtp_proposer)
+        assert proposer.torchair_compiled_model is None
+        assert proposer.torchair_compiled_models = {}
+        Mock.assert_called_once_with(
+            proposer.__class__.__bases__[0],
+            proposer.vllm_config,
+            proposer.device,
+            proposer.runner
+        )


The test_init method has several syntax and logical errors that will prevent it from running correctly:

Line 65: proposer, _, _, = ... is a syntax error due to the trailing comma.

Line 67: isinstance(proposer, setup_torchair_mtp_proposer) is incorrect. The second argument should be the TorchairMtpProposer class, not the fixture function. You'll also need to import TorchairMtpProposer.

Line 69: assert proposer.torchair_compiled_models = {} is an assignment, not a comparison. This is a syntax error.

Lines 70-75: Mock.assert_called_once_with is called on the Mock class, not a mock instance. You should patch the superclass __init__ and assert on the mock object returned by the patch.

gemini-code-assist · 2025-11-20T01:29:12Z

tests/ut/torchair/test_torchair_mtp_proposer.py

+        )
+
+        mock_get_layers = mocker.patch("vllm.config.get_layers_from_vllm_config")
+        assert mock_get_layers.call_count = 2


This line contains an assignment (=) within an assert statement, which is a syntax error. It should be a comparison (==).

assert mock_get_layers.call_count == 2

Signed-off-by: CodeNine-CJ <[email protected]>

…cend into eplb_ci_bugfix * 'eplb_ci_bugfix' of https://github.com/845473182/vllm-ascend: (31 commits) [Test] Add ut test for torchair (vllm-project#4287) [Feat][Doc] Add a load_balance_dp_proxy in examples and external dp doc. (vllm-project#4265) [CI] Defaultly compile vllm with multimodal audio feature in dockerfile (vllm-project#4324) [MM][Bugfix] Add error log for VL models when enabling FLASHCOMM (vllm-project#4272) [Readme] EPLB Support Scenarios (vllm-project#4314) eplb redundant expert bugfix (vllm-project#4291) [Feat][BugFix]Support the Qwen3-Next-80B-A3B-Instruct quantization model&Fix the NZ issue (vllm-project#4245) [Test] Add ACL graph capture/replay DP test (vllm-project#4259) [Test] quick fix mla ut (vllm-project#4318) [Feat] Support MTP to running in full graph mode (vllm-project#3892) [CI] Add mla ut (vllm-project#4280) [Test] Add tests for the multi-node DeepSeek-V2-Lite network in GE Graph (vllm-project#4039) avoid mrope fusion op when running qwen2.5-vl on a+x machine (vllm-project#4270) [Bugfix] fix nightly multi-node EPLB tests' "DYNAMIC_EPLB=true" environment not working (vllm-project#4223) [long seq feat]GQA support long-prefill-token-threshold and fixbug (vllm-project#4209) [misc] clean up get_metadata_cls (vllm-project#4276) [Docs] Improve the AISBench multi-modal testing docs (vllm-project#4255) [doc]fix readme for kv pool user guide (vllm-project#4271) remove get_metadata_cls (vllm-project#4087) [Bugfix] fix hang in async scheduling (vllm-project#4233) ...

### What this PR does / why we need it? The current community lacks unit tests (UT) for files such as torchair_worker, mtp_proposer, and model_runner. Therefore, UT coverage for these files needs to be added. ### Does this PR introduce _any_ user-facing change? NO ### How was this patch tested? - vLLM version: v0.11.0 - vLLM main: vllm-project/vllm@2918c1b --------- Signed-off-by: CodeNine-CJ <[email protected]> Signed-off-by: 白永斌 <[email protected]>

worker ut test

160495a

Signed-off-by: CodeNine-CJ <[email protected]>

github-actions bot added the module:tests label Nov 20, 2025

gemini-code-assist bot reviewed Nov 20, 2025

View reviewed changes

ForBetterCodeNine force-pushed the worker_ut branch 6 times, most recently from ab75b53 to f6bee66 Compare November 20, 2025 03:29

worker ut

fedd8cc

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch 17 times, most recently from 1c24929 to 0b8b27c Compare November 20, 2025 12:37

fixup mocker

87f3f61

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch from 0b8b27c to 87f3f61 Compare November 20, 2025 13:19

fixup

42a6004

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch from bb239b5 to 42a6004 Compare November 20, 2025 13:34

test ut fixup

519ca7e

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch from 75a7d82 to 519ca7e Compare November 20, 2025 13:57

ut test

cb81376

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch 11 times, most recently from 7f84226 to ed14a1a Compare November 21, 2025 06:26

fixup

80a87c1

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch from ed14a1a to 80a87c1 Compare November 21, 2025 06:50

test ut

df7ed30

Signed-off-by: CodeNine-CJ <[email protected]>

ForBetterCodeNine force-pushed the worker_ut branch from 6acb5ba to df7ed30 Compare November 21, 2025 07:11

wangxiyuan approved these changes Nov 21, 2025

View reviewed changes

wangxiyuan merged commit e332e27 into vllm-project:main Nov 21, 2025
18 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Test] Add ut test for torchair #4287

[Test] Add ut test for torchair #4287

ForBetterCodeNine commented Nov 20, 2025 •

edited by github-actions bot

Loading

Uh oh!

github-actions bot commented Nov 20, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Nov 20, 2025

Uh oh!

gemini-code-assist bot Nov 20, 2025

Uh oh!

gemini-code-assist bot Nov 20, 2025

Uh oh!

gemini-code-assist bot Nov 20, 2025

Uh oh!

gemini-code-assist bot Nov 20, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		from vllm.config import VllmConfig


		class TestNPUTorchairModelRunner(PytestBase):

	ascend_config = enable_shared_expert_dp = False
	ascend_config.enable_shared_expert_dp = False

		sys.modules[__name__].vllm_version_is = vllm_version_is


		class TestTorchairMtpProposer(PytestBase):

[Test] Add ut test for torchair #4287

[Test] Add ut test for torchair #4287

Conversation

ForBetterCodeNine commented Nov 20, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this PR does / why we need it?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

github-actions bot commented Nov 20, 2025

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Nov 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ForBetterCodeNine commented Nov 20, 2025 •

edited by github-actions bot

Loading