different generated result when using llama2_flash_attn_monkey_patch
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 39.5k
- Forks
- 4.8k
- PR merge metrics
- No merged PRs in 30d
Description
When use_cache=False, whether fastchat.train.llama2_flash_attn_monkey_patch set True or False, we get the same generated results.
However, when use_cache=True, whether fastchat.train.llama2_flash_attn_monkey_patch set True or False, we get different generated results. About ten generated tokens are the same, afterwards tokens are different.
Does the flash_attn op generate tiny differently and the results are reasonable? Thanks.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing the generated-token difference with use_cache=True and the fastchat.train.llama2_flash_attn_monkey_patch setting enabled and disabled. Inspect the monkey patch and flash_attn operation while comparing the cached and uncached results. Done means establishing whether the divergence is expected or identifying a concrete correction and documenting the result.
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