Why training speed slows down after 314 batch size and accuracy=nan in Step 3 of Experiment3 ?

[yanglinxiabuaaa]

trafficstars

Skipped 6000 image pairs 2022-01-08 18:00:28,652 - root - INFO -

2022-01-08 18:00:28,652 - root - INFO - Before pruning: 2022-01-08 18:00:28,652 - root - INFO - Sparsity range: 0.0 -> 0.1 2022-01-08 18:00:28,652 - root - INFO -

Train Ep. #1: 4%|##3 | 465/12274 [05:52<1:17:12, 2.55it/s, loss=9.35, accuracy=nan, lr=0.0005, sparsity=1, network_width_mpl=1]

Is it normal to have accuracy=nan in the training beginning ?

[yanglinxiabuaaa]

Super slow , 28.95s/it !!!

Train Ep. #1: 5%|##8 | 571/12274 [10:11<94:06:56, 28.95s/it, loss=9.34, accuracy=nan, lr=0.0005, sparsity=1, network_width_mpl=1]