Skip to content

[BugFix] Fix minimax m2 model rotary_dim#30384

Merged
vllm-bot merged 1 commit intovllm-project:mainfrom
rogeryoungh:m2-fix-rotary
Dec 10, 2025
Merged

[BugFix] Fix minimax m2 model rotary_dim#30384
vllm-bot merged 1 commit intovllm-project:mainfrom
rogeryoungh:m2-fix-rotary

Conversation

@rogeryoungh
Copy link
Copy Markdown
Contributor

@rogeryoungh rogeryoungh commented Dec 10, 2025

Purpose

After #29966, get_rope always reads partial_rotary_factor from the configuration and performs the multiplication again, even if rotary_dim has already been scaled. This leads to the factor being applied repeatedly, incorrectly reducing the effective rotational dimension.

Another way to fix #30349.

Test Result

lm_eval --model local-completions \
    --model_args base_url=http://localhost:10086/v1/completions,tokenizer=/model,model=/model \
    --tasks gsm8k_cot  \
    --batch_size 128 \
    --num_fewshot 5 

local-completions (base_url=http://localhost:10086/v1/completions,tokenizer=/model,model=/model), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: 128
|  Tasks  |Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|---------|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k_cot|      3|flexible-extract|     5|exact_match||0.9219|±  |0.0074|
|         |       |strict-match    |     5|exact_match||0.9098|±  |0.0079|

Essential Elements of an Effective PR Description Checklist
  • The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
  • The test plan, such as providing test command.
  • The test results, such as pasting the results comparison before and after, or e2e results
  • (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
  • (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

Signed-off-by: xuebi <xuebi@minimaxi.com>
@chatgpt-codex-connector
Copy link
Copy Markdown

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

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

Code Review

This pull request correctly addresses a bug in the minimax_m2 model where the rotary dimension was being scaled twice. The issue stemmed from passing an already-scaled rotary_dim to the get_rope function, which would then apply the partial_rotary_factor again. The proposed fix of passing self.head_dim as the rotary_dim to get_rope is the correct approach, as it provides the unscaled base dimension, allowing get_rope to perform the scaling correctly. The change is logical, minimal, and effectively resolves the described problem.

@jeejeelee jeejeelee added the ready ONLY add when PR is ready to merge/full CI is needed label Dec 10, 2025
Copy link
Copy Markdown
Member

@hmellor hmellor left a comment

Choose a reason for hiding this comment

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

LGTM! Thanks for the fix

@vllm-bot vllm-bot merged commit d017bce into vllm-project:main Dec 10, 2025
50 of 56 checks passed
Majid-Taheri pushed a commit to Majid-Taheri/vllm that referenced this pull request Dec 23, 2025
Signed-off-by: xuebi <xuebi@minimaxi.com>
Co-authored-by: xuebi <xuebi@minimaxi.com>
Signed-off-by: Ubuntu <mjtaheri68@gmail.com>
dsuhinin pushed a commit to dsuhinin/vllm that referenced this pull request Jan 21, 2026
Signed-off-by: xuebi <xuebi@minimaxi.com>
Co-authored-by: xuebi <xuebi@minimaxi.com>
Signed-off-by: dsuhinin <suhinin.dmitriy@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ready ONLY add when PR is ready to merge/full CI is needed

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants