microsoft / microsoft/aurora

bf16_mode finetuning

Open
#119 9 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1k
Forks
174
PR merge metrics
No merged PRs in 30d

Description

Hi !
I tried to finetune the Air Pollution model using bf16_mode, but it does not seem to work well with the flash attention (at least for me):
Error detected in ScaledDotProductFlashAttentionBackward0. No forward pass information available
RuntimeError: Function 'ScaledDotProductFlashAttentionBackward0' returned nan values in its 1th output

It works fine with the previous implementation of autocast though.
Thought you might be interested in the feedback !

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

Reproduce finetuning the Air Pollution model with bf16_mode and flash attention, then compare it with the previous autocast implementation. Investigate the reported ScaledDotProductFlashAttentionBackward0 error and NaN output; done means bf16_mode finetuning completes without these failures.

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
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.