Pretrain, finetune ANY AI model of ANY size on 1 or 10,000+ GPUs with zero code changes.
COMMITS
February 1, 2024
T
map operator: Add support for non absolute input_dir and output_dir (#19378)
thomas chaton committed
A
Enable saving and loading stateful DataLoaders in Trainer (#19361)
awaelchli committed
W
Support TQDM_MINITERS env variable (#19381)
Wouter Zwerink committed
January 31, 2024
A
Compile guide for Fabric (#19330)
awaelchli committed
J
precommit: drop Black in favor of Ruff (#19380)
Jirka Borovec committed
January 30, 2024
A
Refactor BoringFabric in tests (#19364)
awaelchli committed
T
StreamingDataloader: Resolve typo (#19370)
thomas chaton committed
T
JPEGSerializer: Fix serializer io.bytes image (#19369)
thomas chaton committed
T
Bump Lightning Cloud 0.5.64 (#19372)
thomas chaton committed
M
Shorten docstring (for CLI compat) (#19356)
Michael Pilosov, PhD committed
A
Error message to inform bitsandbytes is only supported on CUDA (#19360)
awaelchli committed
January 29, 2024
A
Update Trainer's ckpt_path type for pathlib Path (#19362)
awaelchli committed
T
map operator: Add support for nested folders (#19366)
thomas chaton committed
T
map operator: Add weights to evenly distributed works among workers (#19365)
thomas chaton committed
January 28, 2024
J
ci: adding missing requirements for generating legacy ckpt (#19353)
Jirka Borovec committed
January 26, 2024
A
Drop support for PyTorch 1.12 (#19300)
awaelchli committed
J
CI: enable testing with coming PT 2.2 (#19289)
Jirka Borovec committed
January 25, 2024
T
Downloader: Resolve race condition (#19348)
thomas chaton committed
T
BC: Switch map operator arguments order (#19345)
thomas chaton committed
January 24, 2024
T
StreamingDataloader: Add profiling support (#19338)
thomas chaton committed
T
Streaming Dataset: tiny optimisations (#19342)
thomas chaton committed
T
tiny improvement (#19341)
thomas chaton committed
A
Allow any AWS authentication method in studios (#19336)
Andy☼ McSherry☼ committed
A
Remove `__len__` from CombinedStreamingDataset (#19321)
awaelchli committed
C
Fallback to `ACCELERATOR_TYPE` for TPU flops (#19314)
Carlos Mocholí committed
A
Reapply `torch.compile` in Fabric.setup() (#19280)
awaelchli committed
January 23, 2024
A
Update Lightning AI multi-node guide (#19324)
awaelchli committed
L
`_restricted_classmethod`: add wrapper, to allow inspection (#19332)
Laurits Fredsgaard Larsen committed
A
Utility to consolidate sharded checkpoints (#19213)
awaelchli committed