[Refactor] Switch gradient checkpointing to the non-reentrant implementation (1/3) - #1979

Open

HAOCHENYE wants to merge 8 commits into

mainfrom

feat/sac-nonlegacy-base

Commits on Aug 14, 2026

[Refactor] Switch gradient checkpointing to the non-reentrant implementation

HAOCHENYE
authored and
jayhenry
committed
2f793bf
[Fix] Keep MTP gradients and the DSA top-k lifecycle correct under the new checkpointing

HAOCHENYE
authored and
jayhenry
committed
1e79a3b
[Refactor] Drop the legacy reentrant checkpoint path

HAOCHENYE
authored and
jayhenry
committed
b9daa9e
[Refactor] Remove the now-unused DSA lifecycle predicate

HAOCHENYE
authored and
jayhenry
committed
7b8bace
[Fix] Let a checkpointed layer's input tensors reach the caller's saved-tensor hooks

HAOCHENYE
authored and
jayhenry
committed
1104050
[Fix] Preserve binding for checkpointed compiled forwards
jayhenry
committed
a236826
[Fix] Consume keyed decoder layer outputs
jayhenry
committed
da6f136
[Fix] Keep FP8 EP checkpoint replay deterministic
jayhenry
committed
386bd7b