COMMITS
August 26, 2026
V
Ignore a stale best checkpoint recorded in a resumed trainer state (#48319)
Vaggelis Giannopoulos committed
J
Keep MXFP4 weights quantized on XPU when use_kernels is set (#47923)
jiqing-feng committed
W
Fix missing FP8 TP layer overrides (#48343)
Wang, Chang committed
S
[docs] Fix links and remove TokenizerFast (#47748)
Steven Liu committed
M
Fix incorrect token classification prefix for ESMC (#48348)
Matt committed
Q
Create the continuous batching CPU group with local synchronization (#48302)
Quentin Gallouédec committed
Q
Resolve continuous batching config against the text config for composite models (#48299)
Quentin Gallouédec committed
S
[Glm 5.3 Flash] GLM 5.3 Flash Support (#48342)
Shijin Zhang committed
M
Fix kernel commit and repo paths for ESMFold2 (#48186)
Matt committed
A
[`CI`] Unblock fast CI for now (failing tests) (#48344)
Anton Vlasjuk committed
Y
Y
H
Normalize HunYuanVL's legacy field aliases via attribute_map (#48261)
Harry Mellor committed
C
Add Qwen4Exp model (#48337)
Cyril Vallez committed
A
Restore BC for the tensor-parallel API (#48300)
Arthur committed
R
Fix video-llama modular conversion (#48336)
Raushan Turganbay committed
T
CI: gate the hunyuan-moe slow test (#48330)
Tarek Ziade committed
L
D
Q
Fix `scores` type in stopping criteria docstrings (#47676)
Quentin Gallouédec committed
R
Docstring check didn't match some file - fix it (#48121)
Raushan Turganbay committed
G
Add shared ImageProcessingTester (#47745)
guarin committed
K
fix bug for blt model parallel bug (#48327)
kaixuanliu committed
J
Cpmant fix use cache (#48013)
jiqing-feng committed
August 25, 2026
V
Stop pairing two-word pretokenized documents in the python tokenizer (#48279)
Vaggelis Giannopoulos committed
S
[docs] Fix failing doctests (#47687)
Steven Liu committed
G
Fix build_2d_sinusoidal_position_embedding on MPS (#47897)
guarin committed
G
Refactor audio buffer creation to use NumPy (#48295)
George Saon committed
S
[docs] Kernel supported models (#48258)
Steven Liu committed