deepseek-ai / deepseek-ai/DeepGEMM

[DSv4] IS W4A16(mxfp4, act bf16) on the planning list?

Open
#374 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Cuda
Stars
7.8k
Forks
1.3k
Avg merge
3d 7h
Merged PRs (30d)
3

Description

For Hopper, SGLang supports both w8a8 and w4a16 path. The latter path supports only marlin/flashinfer_mxfp4. So is there any plan to support this fea in DeepGeMM?

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by reviewing the existing w8a8 and w4a16 paths mentioned in the issue, including the marlin and flashinfer_mxfp4 implementations, then inspect DeepGEMM's planning context for Hopper support. Done means the project has a decided plan or implementation scope for W4A16 with mxfp4 weights and bf16 activations.

Written by the indexing model from the issue text.

Assessment

Domain
performance
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.