COMMITS
August 11, 2026
A
(chore) prepare for release v2.6.0
Akshita Bhagia committed
A
fix dion pypi (#816)
Akshita Bhagia committed
A
(chore) prepare for release v2.6.0
Akshita Bhagia committed
P
Update anonymous paths in public scripts to point to data in hugging face bucket (#802)
Paul Laskowski committed
July 29, 2026
Y
bump fla to 0.5.2 (#798)
Yashas Samaga B L committed
A
Misc fixes (#800)
Akshita Bhagia committed
June 22, 2026
T
Add max_checkpoints to limit permanent checkpoint retention (#694)
Tim Dettmers committed
June 18, 2026
M
CI: cache HuggingFace models for the CPU Test job (#732)
Michael committed
June 12, 2026
A
Add OutputDiscardCheckpoint (#682)
Akshita Bhagia committed
M
Add Qwen3.5 model support (0.8B, 4B, 9B, 27B) (#684)
Michael committed
F
Add conversion overrides for Llama, Qwen3, and Gemma 4 models so they roundtrip properly (#677)
Finbarr Timbers committed
June 11, 2026
F
Add configurable determinism_check to activation checkpointing (#713)
Finbarr Timbers committed
June 9, 2026
J
[1/n] Add vision transformer, connector, and MultimodalLM (#692)
Jason Ren committed
May 28, 2026
A
Misc training-utility improvements (#691)
Akshita Bhagia committed
A
Improve checkpoint/S3 IO robustness (#690)
Akshita Bhagia committed
May 27, 2026
A
CI: run GPU tests at low priority (#696)
Akshita Bhagia committed
May 15, 2026
A
Fix PP FLOPs: capture full model before pipeline split (#680)
Akshita Bhagia committed
A
Increase tolerance for flaky test (#681)
Akshita Bhagia committed
May 14, 2026
A
Two small transformer core fixes: TE Dynamo + PP init seed (#679)
Akshita Bhagia committed
A
Add PowerLR scheduler (#674)
Akshita Bhagia committed
May 13, 2026
A
Add ComposableScheduler (#671)
Akshita Bhagia committed
May 12, 2026
A
Update task timeout for GPU tests (#675)
Akshita Bhagia committed
F
Revert "Add position_ids-based varlen RoPE support for packed inputs" (#672)
Finbarr Timbers committed
May 10, 2026
T
Add position_ids-based varlen RoPE support for packed inputs (#654)
Tai Vu committed
May 8, 2026
F
Fixes the Qwen3 implementation to match HF (#663)
Finbarr Timbers committed
May 6, 2026
F
Record checkpoint save/load durations as trainer metrics (#665)
Finbarr Timbers committed
B
Fix broken path in 32B LC (#670)
Bailey Kuehl committed
A
Fix breaking paths (#669)
Akshita Bhagia committed
May 1, 2026
F
Use case-insensitive HTTP endpoint to check beaker secret existence (#666)
Finbarr Timbers committed
A
Pin flash-linear-attention version to 0.4.1 (#664)
Akshita Bhagia committed