120 kB

Ctrl+K

1 contributor

trainer: the post-stop assertions are rank-0 facts (E-044: hub_cb.steps is empty on rank 1 by construction, so the guard killed every correctly-checkpointed segment)

e26f3f1 verified 11 days ago

hubckpt.py

23.7 kB
hubckpt: Content-Range + size guards on the bounded read-back, and prune_verified() (review E-042) 11 days ago
preflight.py

34.7 kB
Upload train/preflight.py with huggingface_hub 12 days ago
shard_dataset.py

9.61 kB
perm_sha: identify the visit order by hashing the permutation, not just its seed (E-035 residual) -- cursor gains shuffle_perm_sha, asserted on resume; session repins the two files 12 days ago
train_ounce100m.py

52.5 kB
trainer: the post-stop assertions are rank-0 facts (E-044: hub_cb.steps is empty on rank 1 by construction, so the guard killed every correctly-checkpointed segment) 11 days ago