accelerate
https://github.com/huggingface/accelerate
Python
🚀 A simple way to train and use PyTorch models with multi-GPU, TPU, mixed-precision
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported1 Subscribers
Add a CodeTriage badge to accelerate
Help out
- Issues
- allow numpy arrays through weights_only load
- drop shared tensors when saving a plain dict
- Preserve no_grad when composing model hooks
- Stateful split-batch dispatcher replays consumed samples after restoring
- Fix split-batch stateful dispatcher resume
- save_state(total_limit=...) deletes the newest checkpoint and load_state() resumes from an old one on Windows
- Optimizer exceptions leave a stale AMP step marker and advance the scheduler after overflow
- Restore optimizer step state after AMP errors
- Release owned FSDP2 source state before assigning offloaded shards
- Speed up forward with offloaded weights
- Docs
- Python not yet supported