Question about the 0.04 training result #318
Description
I’m seeing a learning result of 0.04.
I am currently training the model using the MMFI dataset with the warm-up epochs set to 5.
Even after reaching 50 epochs, the validation metrics remain completely stagnant at 0.04, showing no signs of improvement whatsoever. Given these configurations, is this the intended performance? It’s hard to believe this is functioning correctly when the learning progress is essentially non-existent.
2026-03-25T15:02:49.091901Z INFO Epoch 5: loss=0.0006 val_pck=0.0450 val_oks=0.0099 lr=1.50e-4
2026-03-25T15:35:12.112450Z INFO Epoch 10: loss=0.0008 val_pck=0.0451 val_oks=0.0101 lr=3.00e-4