deepspeedai / deepspeedai/DeepSpeed

AutoTP for training skips lm_head ?

Open
#7,407 3 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
43.1k
Forks
5k
Avg merge
4d 15h
Merged PRs (30d)
112

Description

I noticed that when using AutoTP for training purpose, the lm_head is skipped. It seems to be only working for inference but not training. Any ideas why ? cc:@inkcherry

Contributor guide

Open the contributing guide

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

The issue names AutoTP and lm_head but no file or test. Start by locating the AutoTP training and inference entry points and comparing how lm_head is handled; reproduce the training path and define done as having lm_head included during training without regressing inference behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
distributed-systems, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.