NVIDIA / NVIDIA/Megatron-LM

FusedDispatch and FusedCombine use default config

Open
#4,316 10 comments 0 reactions 1 assignee Claimed by @YangFei1990 View on GitHub
community-request enhancement
Dominant language
Python
Stars
17.9k
Forks
4.5k
Avg merge
4d 6h
Merged PRs (30d)
271

Description

Currently, FusedDispatch and FusedCombine use default config, which limit performance. Is it possible to allow user-defined configurations? For example, by reading a config file to set the chunk size accordingly.
[fused_a2a.py](https://github.com/NVIDIA/Megatron-LM/blob/b3426025fcf36f33d916590f2dee81e0e425ddd5/megatron/core/transformer/moe/fused_a2a.py#L113)

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.