Skip to content

refactor: align the Vime adapter with the linear_logp contract - #352

Merged
inaniloquentee merged 2 commits into
testfrom
codex/linear-logp-vime-provider
Aug 28, 2026
Merged

inaniloquentee merged 2 commits into
testfrom
codex/linear-logp-vime-provider

Conversation

@inaniloquentee

Copy link
Copy Markdown
Collaborator

Summary

  • move the Vime adapter to the explicit linear_logp_provider path
  • consume the model-neutral structural context without importing Vime dataclasses
  • update runtime hook identity, the Qwen3 TP2/CP2 example, docs, and tests

Validation

  • python -m pytest -q tests/test_vime_linear_logp_provider.py tests/test_vime_qwen3_example.py (13 passed)
  • python -m compileall -q rl_engine tests/test_vime_linear_logp_provider.py tests/test_vime_qwen3_example.py
  • git diff --check

@coderabbitai

coderabbitai Bot commented Aug 28, 2026 •

Copy link
Copy Markdown

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 45487860-cbbd-419f-ae56-c669161c3d4b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review

Comment @coderabbitai help to get the list of available commands.

@Flink-ddd Flink-ddd added the platform: cuda Specific optimizations or bugs in NVIDIA graphics cards (such as FlashInfer, TMA optimizations) label Aug 28, 2026

@Flink-ddd Flink-ddd left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@maxiaosong1124
maxiaosong1124 self-requested a review August 28, 2026 13:49
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

platform: cuda Specific optimizations or bugs in NVIDIA graphics cards (such as FlashInfer, TMA optimizations)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants