[GLM 4.7] Add RTX 6000 Pro aka sm120

[koush]

Motivation

GLM 4.7 fails to run on RTX Pro 6000 because it uses the default Hopper kernels which allocates more SMEM than is available on sm120.

Modifications

Add moe json configs.

Accuracy Tests

Benchmarking and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @koush, 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 enhances the compatibility of GLM 4.7 with the NVIDIA RTX 6000 Pro GPU by introducing specialized Triton kernel configurations. These new configurations resolve shared memory allocation conflicts that previously prevented the model from running on this hardware, thereby expanding the range of supported devices and improving performance on the RTX 6000 Pro.

Highlights

• RTX 6000 Pro (sm120) Compatibility: This pull request addresses a critical issue where GLM 4.7 failed to run on the NVIDIA RTX 6000 Pro GPU (sm120 architecture) due to default Hopper kernels allocating more shared memory (SMEM) than available on the sm120.
• New Triton MoE Kernel Configurations: New Mixture-of-Experts (MoE) Triton kernel configuration files have been added, specifically tailored for the NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition. These configurations provide optimized kernel parameters for both standard and FP8 quantized models with per-channel quantization.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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 by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

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 pull request 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 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. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

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.

Footnotes

1. 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. ↩

[gemini-code-assist]

Code Review

This pull request adds Triton MoE kernel configurations for the NVIDIA RTX PRO 6000 Blackwell GPU (sm120), which is necessary to avoid shared memory errors with the default Hopper kernels. The changes consist of two new JSON configuration files, one for fp8_w8a8 with per-channel quantization and another for default dtypes like fp16/bf16.

My main feedback is regarding the very specific device name used in the filenames. I've left comments on both files suggesting to double-check the device name string to ensure these configurations can be correctly loaded at runtime. Otherwise, the changes look good and address the issue described.

[gemini-code-assist]

The device name NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition in the filename is very specific and seems to contain potentially conflicting terms like Max-Q (typically for laptops) and Workstation_Edition (for desktops). Could you please double-check that this is the exact output of torch.cuda.get_device_name() on your system? If this name is not precise, these configurations might not be loaded. If other variants of this GPU exist, it might be better to use a more generic name, though that would require changes to the config loading logic to support partial matching.