A8W4 fuse moe kernel in testx branch issue

Open
#614 2 comments 0 reactions 1 assignee View on GitHub

@coderfeli is already working on this.

Since Jul 4, 2025.

Assessment

This issue has not been assessed yet.

Description

I modified the inter_dim in https://github.com/ROCm/aiter/blob/d9c1a7390ec27b7bd948f9233f1e49a8fef82d54/op_tests/test_moe_2stage_int4.py#L397 from 4096 to 256, and got

RuntimeError: wrong! device_gemm with the specified compilation parameters does not support this GEMM problem

This inter_dim is needed when deploying deepseek on MI300x with TP8.

To solve this problem, I tried to change the PrefetchStages in https://github.com/ROCm/composable_kernel/blob/b20173a49471a9e22154654fdddec8fe088a4e27/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_b_preshuffle_dequant_v1.hpp#L145 from 2 to 1, and rebuild aiter. After that the test_moe_2stage_int4.py unit test was able to run, but gave the wrong results.

Dominant language
Python
Stars
565
Forks
585
Avg merge
3d 4h
Merged PRs (30d)
366

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.

More from ROCm/aiter

All issues in ROCm/aiter

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.