Skip to content

[chore]: Add Qwen DBO scenarios; fix #270 typos and mypy findings; bugfix #290 - #286

Open
yujuancao07 wants to merge 4 commits into
vllm-project:mainfrom
yujuancao07:weekly
Open

[chore]: Add Qwen DBO scenarios; fix #270 typos and mypy findings; bugfix #290#286
yujuancao07 wants to merge 4 commits into
vllm-project:mainfrom
yujuancao07:weekly

Conversation

@yujuancao07

@yujuancao07 yujuancao07 commented Aug 29, 2026

Copy link
Copy Markdown
Contributor

Purpose

  1. Add the Qwen DBO weekly scenarios.

  2. Fix the typos and mypy-3.10 findings that surfaced when the pre-commit
    suite scanned the fix: gate GPU FFN graph replay on Attention run mode #270 changes

  3. **bugfix [Bug]: FFN CUDA graph replay illegal memory access in 2A1F DBO decode (persists after #270) #290 **

Issue

Scope

In scope

  • test-weekly.yml step contents and the matching README /
    design-doc wording. No runner or model-code changes.
  • determindetermine comment fix; SPDX headers on the
    three test files added with fix: gate GPU FFN graph replay on Attention run mode #270; mixin attribute declarations on
    AFDMetadataProviderMixin; Optional narrowing (asserts / replay guards)
    in the FFN runners; duck-typed helpers annotated Any where the
    contract is dynamic; test-fake annotations. Every hunk maps to a
    reported mypy/typos finding.

Out of scope

@jiangkuaixue123 jiangkuaixue123 added the weekly-test Used to trigger weekly CI in PRs. label Aug 29, 2026
Add Qwen3 MoE and Qwen3.6 MoE DBO scenarios

Signed-off-by: yujuancao07 <yujuancao07@gmail.com>
…w-up checks

- Fix the 'determin' typo in attention_model_runner.py and add the missing
  SPDX headers to the three test files introduced with vllm-project#270.
- Declare the connector/vllm_config/metadata attributes on
  AFDMetadataProviderMixin so the shared helpers type-check.
- Narrow Optional control-plane/model/graph accesses in the FFN runners,
  replace the bool aggregated-key predicate with a sizes-tuple helper in
  cuda_graph.py, and type the duck-typed helpers as Any where the contract
  is dynamic (vLLM fakes, HF config, forward context).
- Convert the dynamic fake-module attribute assignments in
  test_p2p_connector.py to Any-typed locals and fix the test fakes'
  annotations; all changed files pass mypy 3.10, ruff, and the affected
  unit suites.

Signed-off-by: yujuancao07 <yujuancao07@gmail.com>
Signed-off-by: yujuancao07 <yujuancao07@gmail.com>
@yujuancao07 yujuancao07 changed the title [chore]: Add Qwen DBO scenarios; fix #270 typos and mypy findings [chore]: Add Qwen DBO scenarios; fix #270 typos and mypy findings; bugfix #290 Sep 3, 2026
Signed-off-by: yujuancao07 <yujuancao07@gmail.com>
@jiangkuaixue123 jiangkuaixue123 added weekly-test Used to trigger weekly CI in PRs. and removed weekly-test Used to trigger weekly CI in PRs. labels Sep 3, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

weekly-test Used to trigger weekly CI in PRs.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug]: FFN CUDA graph replay illegal memory access in 2A1F DBO decode (persists after #270)

2 participants