Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update bitsandbytes, hivemind, transformers #290

Merged
merged 7 commits into from
Mar 28, 2023
Merged

Update bitsandbytes, hivemind, transformers #290

merged 7 commits into from
Mar 28, 2023

Conversation

justheuristic
Copy link
Collaborator

@justheuristic justheuristic commented Mar 13, 2023

  • new bitsandbytes supports newer and older GPUs
  • new hivemind supports a better bfloat16 codec

Future work:

  • remove CustomLinear8bitlt
  • check bnb backward without memory_efficient_backward

@borzunov borzunov changed the title bump dependency versions Bump dependency versions Mar 13, 2023
@borzunov borzunov mentioned this pull request Mar 28, 2023
4 tasks
Co-authored-by: Alexander Borzunov <[email protected]>
@borzunov borzunov changed the title Bump dependency versions Update bitsandbytes, hivemind, transformers Mar 28, 2023
@borzunov borzunov merged commit 987f4d2 into main Mar 28, 2023
@borzunov borzunov deleted the bump branch March 28, 2023 21:20
borzunov added a commit that referenced this pull request Mar 29, 2023
This PR fixes issues of #290:

- hivemind bfloat16 codec crashed on dummy tensors (with 0 elements), see learning-at-home/hivemind#560 (this PR makes Petals depend on the latest hivemind version from the repo, it's temporary)
- transformers version check mismatched with the version allowed in `setup.cfg`

Also:

- This PR enables 8-bit by default for TP. Even though TP in 8-bit may be slower, we currently prefer to host more blocks to increase the network's stability.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants