NVIDIA / NVIDIA/cutlass

[QST]How can I implement W4A16 gemm kernel with cute api?

Open
#2,168 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

? - Needs Triage inactive-30d inactive-90d question
Dominant language
C++
Stars
10.5k
Forks
2.1k
Avg merge
3d 11h
Merged PRs (30d)
7

Description

What is your question?
in gemm Activation*Weight = C, Activation is f16 type, Weight is int4 type, after Weight gmem->smem, i want to use mma_op = SM80_16x8x16_F16F16F16F16_TN, how can i move Weight smem int4 type to register with cute api using m16n8k16 B(type f16) matrix layout? then I can transform int4 to fp16 in register and complete gemm.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the W4A16 GEMM question and the referenced SM80_16x8x16_F16F16F16F16_TN operation. Investigate CUTLASS CuTe handling of an int4 Weight in shared memory with the m16n8k16 B layout, including conversion to fp16 in registers. Done means a documented or working path for completing the requested GEMM.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
hpc
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.