deepspeedai / deepspeedai/DeepSpeedExamples

[BUG] DeepSpeed-Chat Step3 - actor model repeats generating the same token when hybrid engine enabled

Open
#821 9 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
6.8k
Forks
1.1k
Avg merge
2d 16h
Merged PRs (30d)
1

Description

Keep other settings the same, when enabling the hybrid engine, the actor model in Step 3 generates the same token one by one until reaching the max length of the answer (id 29962 is the end of my prompt, the repeated token id is 517):

Screenshot 2023-11-30 at 17 53 08

when I disabled the hybrid engine, the actor model generates normally:

Screenshot 2023-11-30 at 17 46 02

Is there anything wrong with the hybrid engine? Thanks!

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.

Research direction

Start with the DeepSpeed-Chat Step3 actor-model entry point and reproduce the reported behavior with the hybrid engine enabled and disabled while keeping other settings unchanged. Compare the two configurations and runtime output; done means identifying and correcting the repeated-token behavior without changing the normal generation path.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.