alibaba / alibaba/ChatLearn

[Feature]support to parameter sync when trainer_tp divides inference_tp for megatron core model

Open
#105 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
450
Forks
40
PR merge metrics
No merged PRs in 30d

Description

we already support this feature for legacy models and qwen2 since [PR](https://github.com/alibaba/ChatLearn/pull/92). but models of mcore format might break.

Contributor guide

No contributing guide indexed for this repository

Research direction

Start by comparing the legacy-model and Qwen2 support described in PR #92, then locate the Megatron Core model parameter-sync entry point. Reproduce the case where trainer_tp divides inference_tp and verify that parameter synchronization works for the Megatron Core format without breaking existing model paths.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
distributed-systems, machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.