diff options
| author | skal <pascal.massimino@gmail.com> | 2026-03-25 22:08:19 +0100 |
|---|---|---|
| committer | skal <pascal.massimino@gmail.com> | 2026-03-25 22:08:19 +0100 |
| commit | 4ca498277b033ae10134045dae9c8c249a8d2b2b (patch) | |
| tree | 0b834c3f5cacc6cbf7cdab37c4d41ac217fb0d9a /workspaces/test/test_assets/test_asset_1.txt | |
| parent | 4ad0e121108261884cdf49374481e04095a6d9c7 (diff) | |
fix(cnn_v3/training): rebuild optimizer before loading state on resume past FiLM warmup
When resuming a checkpoint saved after the FiLM warmup phase, the optimizer
was created with frozen (fewer) param groups, causing a size mismatch when
loading the saved optimizer state. Fix: detect ckpt['epoch'] >= film_warmup,
unfreeze FiLM MLP, and rebuild the optimizer before loading its state dict.
handoff(Gemini): train_cnn_v3.py --resume now works past epoch 1500.
Diffstat (limited to 'workspaces/test/test_assets/test_asset_1.txt')
0 files changed, 0 insertions, 0 deletions