[tml] Inkling model support - #1683

Merged

Zhichenzzz merged 52 commits into

mainfrom

Aug 3, 2026

Commits on Jul 15, 2026

[tml] inkling model support
Zhichenzzz
committed
56156a7

Commits on Jul 17, 2026

[tml] Add Inkling 975B fully-async RL script (8 train + 4 rollout nodes)
Zhichenzzz
committed
a61b986

Commits on Jul 28, 2026

Merge origin/main into zhichen/feat (Inkling)
Zhichenzzz
committed
9c54802
[tml] fix black formatting in the Inkling launchers
Zhichenzzz
committed
0a7b31f
[tml] fix three merge regressions against main's fast tests
Zhichenzzz
committed
e206905
[tml] format the atomic-group assert the way black 24.3.0 wants
Zhichenzzz
committed
05d0a68
[tml] add an Inkling 4-layer megatron model-script CI smoke test
Zhichenzzz
committed
038b9b4
[tml] read args.load defensively in the missing-checkpoint gate
Zhichenzzz
committed
3a3a1ef
[tml] only skip the checkpoint load when --load is staged but empty
Zhichenzzz
committed
15838c5
[tml] pass --experts-shared-outer-loras on the Inkling LoRA path
Zhichenzzz
committed
1a290d1
[tml] restore unconditional --no-offload-* on the LoRA path, with the reason
Zhichenzzz
committed
efb2a1f
[tml] repack the LoRA adapter onto fresh storage before CUDA IPC
Zhichenzzz
committed
e1e7ae7
[tml] let the LoRA path take the offload barrier on smaller engines
Zhichenzzz
committed
ae15475
[tml] gate the LoRA IPC repack on offload_train
Zhichenzzz
committed
5f91016
[tml] drop the verbose LoRA-flag comments from the launcher
Zhichenzzz
committed
13033a0
[tml] strip explanatory comments from the Inkling code
Zhichenzzz
committed
d05468f
[tml] add an Inkling 4-layer LoRA model-script CI smoke test
Zhichenzzz
committed
f309bab

Commits on Jul 29, 2026

Commits on Jul 30, 2026

Commits on Jul 31, 2026

Commits on Aug 1, 2026

Commits on Aug 3, 2026