Skip to content

[Model] Fix Skywork R1V mlp #26673

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Oct 13, 2025
Merged

[Model] Fix Skywork R1V mlp #26673

merged 2 commits into from
Oct 13, 2025

Conversation

@jeejeelee
Copy link
Collaborator

@jeejeelee jeejeelee commented Oct 13, 2025 •
edited by github-actions bot
Loading

Purpose

FIX #20818

Test Plan

Test Result

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.
@jeejeelee
Done
70d8b89 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
@jeejeelee jeejeelee requested a review from Isotr0py October 13, 2025 03:27
gemini-code-assist[bot]
gemini-code-assist bot reviewed Oct 13, 2025
View reviewed changes
Copy link
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 aims to fix the MLP component of the Skywork R1V model, likely to enable correct quantization. The changes correctly propagate the quant_config and prefix to the ReplicatedLinear layers within the mlp1 module. However, there's a critical issue in how the prefix is constructed at the call site for _init_mlp1, which will lead to incorrect layer prefixes and likely cause quantization to fail. I've provided a suggestion to correct this.
chatgpt-codex-connector[bot]
chatgpt-codex-connector bot reviewed Oct 13, 2025
View reviewed changes
Copy link

@chatgpt-codex-connector chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

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

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.
vllm/model_executor/models/skyworkr1v.py
Comment on lines 748 to +756
vit_hidden_size * int(1 / self.downsample_ratio) ** 2,
llm_hidden_size,
return_bias=False,
quant_config=quant_config,
Copy link

@chatgpt-codex-connector chatgpt-codex-connector bot Oct 13, 2025

Choose a reason for hiding this comment

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

P1 Badge Prefix omits mlp1 when registering quantized projector

The new _init_mlp1 forwards the outer prefix directly into the ReplicatedLinear layers (prefix=f"{prefix}.1"/.3). Because this module is attached as self.mlp1, the actual state‑dict path for the quantized weights is …mlp1.1 and …mlp1.3. AWQ/GPTQ configs rely on the prefix string to find the precomputed qweight/scales or to decide which modules to skip; registering the layers under prefix.1 means the loader will not match the checkpoint entries (mlp1.1.*) and the quantized projector still fails to load. Wrap the prefix with maybe_prefix(prefix, "mlp1") before appending the numeric suffix so that weight lookup works for quantized Skywork R1V checkpoints.

Useful? React with 👍 / 👎.
@jeejeelee
Fix
0f0e931 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
DarkLight1337
DarkLight1337 approved these changes Oct 13, 2025
View reviewed changes
Copy link
Member

@DarkLight1337 DarkLight1337 left a comment

Choose a reason for hiding this comment

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

Thanks for fixing!
@DarkLight1337 DarkLight1337 enabled auto-merge (squash) October 13, 2025 03:58
@github-actions github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Oct 13, 2025
@jeejeelee jeejeelee changed the title [Moel] Fix Skywork R1V mlp [Model] Fix Skywork R1V mlp Oct 13, 2025
@vllm-bot vllm-bot merged commit 98f30b8 into vllm-project:main Oct 13, 2025
52 of 55 checks passed
@jeejeelee jeejeelee deleted the fix-skyr1v-qconfig branch October 13, 2025 07:40
bbeckca pushed a commit to bbeckca/vllm that referenced this pull request Oct 13, 2025
@jeejeelee @bbeckca
[Model] Fix Skywork R1V mlp (vllm-project#26673)
b9803b5 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
1994 pushed a commit to 1994/vllm that referenced this pull request Oct 14, 2025
@jeejeelee @1994
[Model] Fix Skywork R1V mlp (vllm-project#26673)
02905ae 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: 1994 <1994@users.noreply.github.com>
Dhruvilbhatt pushed a commit to Dhruvilbhatt/vllm that referenced this pull request Oct 14, 2025
@jeejeelee @Dhruvilbhatt
[Model] Fix Skywork R1V mlp (vllm-project#26673)
49cf9c0 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: Dhruvil Bhatt <bhattdbh@amazon.com>
bbartels pushed a commit to bbartels/vllm that referenced this pull request Oct 16, 2025
@jeejeelee @bbartels
[Model] Fix Skywork R1V mlp (vllm-project#26673)
7d68465 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: bbartels <benjamin@bartels.dev>
lywa1998 pushed a commit to lywa1998/vllm that referenced this pull request Oct 20, 2025
@jeejeelee @lywa1998
[Model] Fix Skywork R1V mlp (vllm-project#26673)
7935fed 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
alhridoy pushed a commit to alhridoy/vllm that referenced this pull request Oct 24, 2025
@jeejeelee @alhridoy
[Model] Fix Skywork R1V mlp (vllm-project#26673)
e7c558c 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Oct 24, 2025
@jeejeelee @xuebwang-amd
[Model] Fix Skywork R1V mlp (vllm-project#26673)
2170d6b 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: xuebwang-amd <xuebwang@amd.com>
xuebwang-amd pushed a commit to xuebwang-amd/vllm that referenced this pull request Oct 24, 2025
@jeejeelee @xuebwang-amd
[Model] Fix Skywork R1V mlp (vllm-project#26673)
a9791ed 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: xuebwang-amd <xuebwang@amd.com>
0xrushi pushed a commit to 0xrushi/vllm that referenced this pull request Oct 26, 2025
@jeejeelee @0xrushi
[Model] Fix Skywork R1V mlp (vllm-project#26673)
0150e6b 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: 0xrushi <6279035+0xrushi@users.noreply.github.com>
0xrushi pushed a commit to 0xrushi/vllm that referenced this pull request Oct 26, 2025
@jeejeelee @0xrushi
[Model] Fix Skywork R1V mlp (vllm-project#26673)
f3245b1 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: 0xrushi <6279035+0xrushi@users.noreply.github.com>
rtourgeman pushed a commit to rtourgeman/vllm that referenced this pull request Nov 10, 2025
@jeejeelee @rtourgeman
[Model] Fix Skywork R1V mlp (vllm-project#26673)
bfe0e2c 
Signed-off-by: Jee Jee Li <pandaleefree@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

3 participants

@jeejeelee @DarkLight1337 @vllm-bot