[2/N] Quantization Refactor: Compressed tensors MoE schemes

[TamirBaydasov]

Motivation

Add MoE schemes to compressed-tensors instead of storing all classes in a single file. This should make it easier to implement additional functionality as well as clearly show that compressed-tensors also supports NPU hardware in some cases.

Images and motivation for this PR can be viewed in our roadmap: #15194

Modifications

Moved all classes from compressed_tensors_moe.py to new schemes in quantization/compressed_tensors/schemes/
Removed compressed_tensors_moe.py file
Added get_moe_scheme function to compressed_tensors.py file
Fixed quant method check in FusedMoE class to reflect new changes

Accuracy Tests

GPU tests:

Qwen3-30B-A3B-Instruct-2507-AWQ

Qwen3-30B-A3B-Instruct-2507-FP8

NPU tests:

Qwen3-30B-A3B-Instruct-2507-W8A8

Benchmarking and Profiling

Not affected

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 @TamirBaydasov, 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 significantly refactors the Mixture-of-Experts (MoE) quantization implementation within the compressed-tensors library. The primary goal is to transition from a monolithic class structure to a more flexible scheme-based design. This change not only simplifies the integration of future quantization functionalities but also explicitly highlights and enhances support for NPU hardware, ensuring better compatibility and performance across different platforms.

Highlights

• MoE Quantization Refactoring: The Mixture-of-Experts (MoE) quantization logic has been refactored from a class-based approach to a more modular scheme-based architecture, improving extensibility and clarity.
• New MoE Quantization Schemes: Several new MoE quantization schemes have been introduced, including CompressedTensorsW4A4Nvfp4MoE, NPUCompressedTensorsW4A8Int8DynamicMoE, CompressedTensorsW8A8Fp8MoE, NPUCompressedTensorsW8A8Int8DynamicMoE, CompressedTensorsWNA16MoE, and NPUCompressedTensorsW4A16Int4DynamicMoE.
• NPU Hardware Support: Explicit support for NPU hardware has been integrated into the compressed-tensors framework through dedicated NPU-specific MoE schemes, making it clearer that NPU is supported in certain configurations.
• Unified MoE Method Handling: The FusedMoE class now dynamically retrieves the appropriate MoE scheme via a new get_moe_scheme function, streamlining the process of applying different quantization methods.

🧠 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 refactors the Mixture of Experts (MoE) schemes for compressed tensors by moving them into a new schemes directory. This is a positive structural change that should improve modularity and extensibility. However, the review has identified several critical issues that appear to be oversights from the refactoring process. These include typos that will lead to runtime errors, incorrect method signatures and implementations in the new CompressedTensorsFusedMoEMethod class, and missing methods in one of the new scheme files. Addressing these issues is crucial for the correctness of the implementation.

[HandH1998]

/rerun-failed-ci

[HandH1998]

/rerun-failed-ci

[HandH1998]

/rerun-failed-ci

[ping1jing2]

/rerun-failed-ci

[ping1jing2]

/rerun-failed-ci

[ping1jing2]

i merge it since both HandH1998 and
AniZpZ approved it. The single failing AMD UT(Exception: Capture cuda graph failed: Could not find CUDA installation. Please set CUDA_HOME environment variable.) is unrelated to the code change, please let me know if there are any issues.

[JustinTong0323]

This should be apply_weights @ping1jing2 @TamirBaydasov , causing this error:

  File "/root/xinyuan/sglang/python/sglang/srt/models/deepseek_v2.py", line 1778, in <lambda>
    lambda idx, prefix: DeepseekV2DecoderLayer(
                        ^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/xinyuan/sglang/python/sglang/srt/models/deepseek_v2.py", line 1522, in __init__
    self.mlp = DeepseekV2MoE(
               ^^^^^^^^^^^^^^
  File "/root/xinyuan/sglang/python/sglang/srt/models/deepseek_v2.py", line 386, in __init__
    self.experts = get_moe_impl_class(quant_config)(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/xinyuan/sglang/python/sglang/srt/layers/moe/fused_moe_triton/layer.py", line 277, in __init__
    self.quant_method = quant_config.get_quant_method(self, prefix)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/xinyuan/sglang/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py", line 179, in get_quant_method
    layer.scheme = self.get_moe_scheme(layer=layer, layer_name=prefix)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/xinyuan/sglang/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py", line 680, in get_moe_scheme
    return CompressedTensorsMxInt4MoE(self)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Can't instantiate abstract class CompressedTensorsMxInt4MoE without an implementation for abstract method 'apply_weights'

[ping1jing2]

got it, we will hotfix it tomorrow