NVIDIA / NVIDIA/Megatron-LM

[QUESTION] HyperConnectionModule implementation of sinkhorn algorithm problem

Open
#5,106 3 comments 0 reactions 1 assignee Claimed by @jingqiny-99 View on GitHub
community-request question waiting-on-customer
Dominant language
Python
Stars
17.9k
Forks
4.5k
Avg merge
4d 6h
Merged PRs (30d)
271

Description

In the dev branch, the `HyperConnectionModule` implements the **sinkhorn** algorithm, but the `native` and `Fusion versions` are not aligned. Furthermore, the native and Deepseek implementations are also not aligned. What is the reason for this implementation discrepancy?

### MCore native version

Image

### MCore fusion version
Image

### DeepSeek tilekernel version

Image

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.