pytorch-lightning
https://github.com/pytorchlightning/pytorch-lightning
Python
The lightweight PyTorch wrapper for high-performance AI research. Scale your models, not the boilerplate.
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 supported12 Subscribers
Add a CodeTriage badge to pytorch-lightning
Help out
- Issues
- Allow scheduled gradient accumulation across repeated fits
- Delegate async checkpoint teardown to the wrapped backend
- Recheck model FLOPs capability when reusing ThroughputMonitor
- perf(fabric): avoid per-element host syncs and use tuple slicing in _gather_all_tensors
- Add `val_check_interval_unit` to validate every N optimizer steps
- Avoid per-parameter device sync and redundant transfers in `StochasticWeightAveraging.update_parameters`
- Stage AsyncCheckpointIO tensor snapshots on CPU
- Fix `_load` treating local relative paths starting with `http` as URLs
- Fix `save_hyperparameters` crash when `__init__` has array/tensor args alongside a container
- AsyncCheckpointIO retains live NumPy arrays while snapshotting model tensors
- Docs
- Python not yet supported