COMMITS
August 5, 2026
Y
ci(minimax): extend M2.7 LoRA timeout (#3428)
Yuhe Zhang committed
A
perf(moe): vectorize deterministic expert bias gradients (#3387)
Alexandros Koumparoulis committed
A
fix(transformers): default missing THD capability (#3406)
Alexandros Koumparoulis committed
L
docs(fern): add observed 404 redirects (#3425)
Lawrence Lane committed
A
fix(ci): run Kimi K3 HellaSwag on GB200 (#3420)
Alexandros Koumparoulis committed
S
fix(docs): AUT-1324 qualify LTX model coverage slug (#3418)
svcnemo-autobot committed
T
feat: Add {% generation %} chat template for DiffusionGemma SFT/LoRA examples (#3353)
Tianqianjin Lin committed
August 4, 2026
Y
test(checkpoint): harden MoE reload parity checks (#3335)
Yuhe Zhang committed
D
fix(deps): resolve OSS CVE findings (#3398)
Dong Hyuk Chang committed
A
fix(gemma4): enable expandable_segments on the CP tulu3 recipes (#3408)
Abhishree Thittenamane committed
A
fix(moe): support EP-free DTensor state-dict conversion (#3397)
Alexandros Koumparoulis committed
A
docs(models): automate latest model coverage (#3351)
Alexandros Koumparoulis committed
Y
fix(checkpoint): harden PEFT PP checkpointing and Step-3.7 coverage (#3316)
Yuhe Zhang committed
Y
fix(retrieval): guard optional W&B imports (#3380)
Yuhe Zhang committed
A
perf: remove Python overhead from model hot paths (#3374)
Alexandros Koumparoulis committed
Y
docs(training): mark Mamba prewarm sections for review (#3340)
Yuhe Zhang committed
A
ci: add gradient-determinism rule to PR review guidelines (#3392)
Alexandros Koumparoulis committed
S
chore(ci): AUT-1298 bump community workflow to v1.8.7 (#3378)
svcnemo-autobot committed
Y
ci: tune Nemotron single-GPU model load threads (#3370)
Yuhe Zhang committed
Y
fix(checkpoint): survive an interrupted save instead of hanging all ranks (#3261)
Yuhe Zhang committed
Y
fix(model): correct Mistral4 attention and distributed MoE routing (#3348)
Yuhe Zhang committed
A
fix(ci): enable LTX-2.3 diffusion finetuning (#3372)
Alexandros Koumparoulis committed
A
perf(checkpoint): reduce distributed save overhead (#3369)
Alexandros Koumparoulis committed
N
fix(fsdp): resolve fp32 master-weight compute dtype per parameter (#3328)
NancyFyong committed
August 3, 2026
H
feat(models): make Inkling implementation standalone (#3358)
Huiying committed
D
fix(ci): use a single container cache donor (#3377)
Dong Hyuk Chang committed
C
fix(tools): save processor artifacts in LoRA merge for #3256 (#3270)
Chenhe committed
A
fix(distributed): stabilize SAC replay with TE and FSDP (#3330)
Alexandros Koumparoulis committed
A
fix(tokenizer): make special token insertion opt-in (#3337)
Alexandros Koumparoulis committed