GLM-4 Update #20736

zRzRzRzRzRzRzR · 2025-07-10T05:44:00Z

GLM-4 Update of Moe Support

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

github-actions · 2025-07-10T05:44:10Z

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

gemini-code-assist

Summary of Changes

Hello @zRzRzRzRzRzRzR, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces comprehensive support for the GLM-4 Mixture-of-Experts (MoE) model within vLLM. It provides the full architectural implementation, including specialized MoE components, and integrates the new model into the existing benchmarking and model loading infrastructure.

Highlights

New Model Support: I've added the complete implementation for the Glm4MoeForCausalLM model architecture, enabling vLLM to support GLM-4 Mixture-of-Experts models.
Mixture-of-Experts (MoE) Implementation: The new glm4_moe.py file includes custom components like Glm4MoeTopkRouter for expert selection and Glm4MoeSparseMoeBlock to handle the specific routing and shared expert mechanisms of GLM-4's MoE architecture.
System Integration: Updates were made to the MoE benchmarking script (benchmark_moe.py) and the model registry (registry.py) to properly recognize and integrate the new Glm4MoeForCausalLM model within vLLM's ecosystem.
Weight Loading Logic: The new model implementation includes robust weight loading logic to correctly map and load parameters from HuggingFace checkpoints, covering stacked linear layers (QKV, gate/up projections) and individual expert weights.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in issue comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist is currently in preview and may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments to provide feedback.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

gemini-code-assist

Code Review

The pull request adds support for GLM-4 MoE models to vLLM. The changes include modifying the benchmark script to include GLM-4 MoE, adding the GLM-4 MoE model to the model registry, and creating a new file for the GLM-4 MoE model architecture. There's a potential typo in the benchmark script that needs to be addressed, and the model path in the registry should be a real model name.

benchmarks/kernels/benchmark_moe.py

tests/models/registry.py

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

vllm/model_executor/models/glm4_moe.py

jeejeelee

Thank you, we also need to update the document.

vllm/transformers_utils/configs/ovis.py

zRzRzRzRzRzRzR · 2025-07-11T15:42:37Z

Thank you, we also need to update the document.

you mean supported_models.md, isnt it generate auto?

jeejeelee · 2025-07-12T08:44:02Z

Thank you, we also need to update the document.

Thank you, we also need to update the document.

you mean supported_models.md, isnt it generate auto?

Yes, we need to update manually.

Signed-off-by: Isotr0py <[email protected]>

zRzRzRzRzRzRzR · 2025-07-13T15:57:18Z

We also have the MTP model, but if this PR meets the requirements, it can be merged first. The implementation of MTP can be proposed in a separate PR.

zRzRzRzRzRzRzR · 2025-07-19T04:05:32Z

This seems to be due to the example model not being released, making it impossible to find, or is it an error with the VoxtralForConditionalGeneration model? @DarkLight1337

DarkLight1337 · 2025-07-19T04:14:13Z

You can ignore the problem related to Voxtral. However, the GLM-4-MoE-MTP model is also failing the test which needs to be fixed here

zRzRzRzRzRzRzR · 2025-07-19T05:14:08Z

AssertionError: function <function can_initialize at 0x7ff0e802a520> failed when called with args ('Glm4MoeMTPModel', <_pytest.monkeypatch.MonkeyPatch object at 0x7ff2635270e0>, <tests.models.registry.HfExampleModels object at 0x7ff0ea18aae0>) and kwargs {}

I only seem to have found this one error related to GLM-4-MoE. But is this caused by the model being too large? It looks like the error is because the model can’t be found. Our model hasn’t been open-sourced yet, and the model is indeed named GLM-4.5.

ERROR 07-19 05:12:42 [config.py:129] Error retrieving file list: (MaxRetryError("HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /api/models/THUDM/GLM-4.5/tree/main?recursive=True&expand=False (Caused by NewConnectionError('<urllib3.connection.HTTPSConnection object at 0x7fa5e4e76170>: Failed to establish a new connection: [Errno 101] Network is unreachable'))"), '(Request ID: 1c6445ab-986e-4e61-ac9f-964c2099442e)'), retrying 1 of 2

DarkLight1337 · 2025-07-19T05:16:16Z

If the checkpoint is not released yet, set is_available_offline=False for that model in the registry

mergify · 2025-07-19T05:51:03Z

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @zRzRzRzRzRzRzR.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

DarkLight1337 · 2025-07-19T06:46:45Z

Sorry there are more conflicts now that need to be solved

zRzRzRzRzRzRzR · 2025-07-19T07:05:27Z

done and test again

zRzRzRzRzRzRzR · 2025-07-19T09:09:19Z

@DarkLight1337 Now it is not GLM-MoE Error

DarkLight1337 · 2025-07-19T09:12:35Z

https://buildkite.com/vllm/ci/builds/24404/steps/canvas?jid=01982180-cbbf-43c2-91ea-a4efe642c99e this one may be caused by your changes

zRzRzRzRzRzRzR · 2025-07-19T09:29:58Z

Let me try again, I modified some local files. It seems there are only errors from llama3.2 left. I'm really not very familiar with this part of the code.

vllm/config.py

Isotr0py

LGTM if all tests can pass. Assuming that model outputs are aligned on your side.

DarkLight1337

Same, thanks for working on this!

Signed-off-by: zRzRzRzRzRzRzR <[email protected]> Signed-off-by: Isotr0py <[email protected]> Signed-off-by: Lu Fang <[email protected]> Co-authored-by: Isotr0py <[email protected]> Co-authored-by: Lu Fang <[email protected]>

init

e11b28a

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

zRzRzRzRzRzRzR requested review from DarkLight1337 and ywang96 as code owners July 10, 2025 05:44

gemini-code-assist bot reviewed Jul 10, 2025

View reviewed changes

mergify bot added the performance Performance-related issues label Jul 10, 2025

ovis aimv2 model type need changed, mark

818db59

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

gemini-code-assist bot reviewed Jul 10, 2025

View reviewed changes

benchmarks/kernels/benchmark_moe.py Outdated Show resolved Hide resolved

tests/models/registry.py Outdated Show resolved Hide resolved

zRzRzRzRzRzRzR added 3 commits July 10, 2025 14:01

format

c6b8eb6

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

format

f406a09

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

use ds loading(not work)

efcff2b

mergify bot added the new-model Requests to new models label Jul 10, 2025

zRzRzRzRzRzRzR added 8 commits July 10, 2025 18:42

update

e6ad576

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

update 1

d49c5bc

Signed-off-by: zRzRzRzRzRzRzR <[email protected]>

test

c8ad31e

Update glm4_moe.py

2440756

update

5e38b14

Update glm4_moe.py

c227471

1

9f3ab70

use ds imp

8feace6

jeejeelee reviewed Jul 11, 2025

View reviewed changes

vllm/model_executor/models/glm4_moe.py Show resolved Hide resolved

jeejeelee reviewed Jul 11, 2025

View reviewed changes

vllm/transformers_utils/configs/ovis.py Outdated Show resolved Hide resolved

zRzRzRzRzRzRzR added 2 commits July 11, 2025 22:26

Merge branch 'vllm-project:main' into glm-moe

a875d98

update and merge

12666d3

Isotr0py and others added 2 commits July 12, 2025 22:28

fix partial_rotary_factor

ffe9d62

Signed-off-by: Isotr0py <[email protected]>

Merge branch 'vllm-project:main' into glm-moe

19de827

Update for doc

dbf1719

2

7b36970

update

554e467

mergify bot added the needs-rebase label Jul 19, 2025

Update registry.py

c829435

Merge remote-tracking branch 'upstream/main' into glm-moe

a305bc2

mergify bot removed the needs-rebase label Jul 19, 2025

add glm_4_moe

1065a13

Isotr0py reviewed Jul 19, 2025

View reviewed changes

zRzRzRzRzRzRzR added 3 commits July 19, 2025 21:28

Merge branch 'vllm-project:main' into glm-moe

09620f9

1

3973a74

Update config.py

6618ce6

Isotr0py approved these changes Jul 19, 2025

View reviewed changes

Isotr0py enabled auto-merge (squash) July 19, 2025 16:08

DarkLight1337 approved these changes Jul 19, 2025

View reviewed changes

Isotr0py merged commit 10eb24c into vllm-project:main Jul 19, 2025
74 checks passed

github-project-automation bot moved this to Done in Tool Calling Jul 19, 2025

zRzRzRzRzRzRzR deleted the glm-moe branch July 20, 2025 03:07

zRzRzRzRzRzRzR restored the glm-moe branch July 21, 2025 13:26

zRzRzRzRzRzRzR deleted the glm-moe branch July 21, 2025 13:26

Uh oh!

GLM-4 Update #20736

GLM-4 Update #20736

Conversation

zRzRzRzRzRzRzR commented Jul 10, 2025

Uh oh!

github-actions bot commented Jul 10, 2025

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jeejeelee left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

zRzRzRzRzRzRzR commented Jul 11, 2025

Uh oh!

jeejeelee commented Jul 12, 2025

Uh oh!

zRzRzRzRzRzRzR commented Jul 13, 2025

Uh oh!

zRzRzRzRzRzRzR commented Jul 19, 2025

Uh oh!

DarkLight1337 commented Jul 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

zRzRzRzRzRzRzR commented Jul 19, 2025

Uh oh!

DarkLight1337 commented Jul 19, 2025

Uh oh!

mergify bot commented Jul 19, 2025

Uh oh!

DarkLight1337 commented Jul 19, 2025

Uh oh!

zRzRzRzRzRzRzR commented Jul 19, 2025

Uh oh!

zRzRzRzRzRzRzR commented Jul 19, 2025

Uh oh!

DarkLight1337 commented Jul 19, 2025

Uh oh!

zRzRzRzRzRzRzR commented Jul 19, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Isotr0py left a comment

Choose a reason for hiding this comment

Uh oh!

DarkLight1337 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

DarkLight1337 commented Jul 19, 2025 •

edited

Loading