Abhinavsing/elastic - #4434

Open

abhinavclemson wants to merge 31 commits into

lukebaumann/elastic-snapshotfrom

abhinavsing/elastic

Commits on Jun 30, 2026

Commits on Jul 1, 2026

Add pathways elastic training resiliency (asynchronous host snapshotting, dynamic mesh recovery, and GKE scale-up)
lukebaumann
committed
f5dbe2c
Add scale test debugging and logging
lukebaumann
committed
463e6dc
Fix elasticity bugs: wait for slices at startup and recalculate batch sizes on recovery
lukebaumann
committed
4f0d0b3
Fix save_pytree mismatch: save state_dict instead of TrainState after scale-up
lukebaumann
committed
78304e2
Fix elasticity recovery: wait for active slices and add empty snapshot check
lukebaumann
committed
1986e9e
Remove comment and avoid unpacking gbs_load_start in recover
lukebaumann
committed
33a9817
Fix recover to use metric_logger_instance
lukebaumann
committed
5749742
Use Snapshotter for metrics recovery and handle non-Array leaves
lukebaumann
committed
e834104

Commits on Jul 2, 2026

Commits on Jul 11, 2026

Fix Data Loss issue in DataLoader.
abhinavclemson
committed
0c6ee16
Fix maxtext import issue.
abhinavclemson
committed
551a688