mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-19 23:50:51 +00:00
A checkpoint is taken at a round boundary, so with rounds longer than the gap between checkpoints it landed at the first boundary past its batch, and a run of one long round checkpointed only at its end. A round now hands out no batch past the next multiple of `every` or the next epoch's end, and the checkpoint lands on its batch. hanzo-train 0.11.20.