lars
5d8567ef1e
CI / Sync project version with tag (push) Has been skipped
CI / Format (ruff format) (push) Successful in 32s
CI / Lint (ruff check) (push) Successful in 33s
CI / Type check (ty) (push) Successful in 3m54s
CI / Tests (push) Successful in 4m47s
CI / Bump version, tag, and update changelog on merge to master (push) Has been skipped
The per-epoch training fan-out only makes sense if epoch k is the same epoch either way, and the shuffle fix alone wasn't enough: run_train_job calls seed_everything(train.seed) at process start, so a fresh job restarted the torch/numpy stream at epoch 1's state and drew different flow/WGAN noise than the corresponding epoch of a single long run. giant.config.epoch_seed derives a per-epoch seed, and the training loop reseeds from it at the top of every epoch. Verified on a 3-epoch toy run: the chained workflow's concatenated metrics.csv is now byte-identical to a single `giant train --epochs 3` with the same seed (it matched only on epoch 1 before). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Symbolic link
1 line
50 B
Plaintext
Symbolic link
1 line
50 B
Plaintext
run-20260826_120646-single/logs/debug-internal.log |