learning_rate: 5e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 500
num_epochs: 15
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Cer	Decode Mode	Validation Loss	Norm Lev Sim Mean	Wer
20.1827	1.0	95	0.9775	greedy	26.9940	0.0230	1.0
18.5724	2.0	190	1.0594	greedy	25.8965	0.0261	1.0
18.9031	3.0	285	0.9740	greedy	23.9979	0.0265	1.0
14.6475	4.0	380	0.9683	greedy	21.1389	0.0323	1.0
14.3493	4.9973	470	17.5431	0.0321	1.0	0.9682
11.6066	6.0	565	12.5254	0.0280	1.0	0.9725
7.5114	7.0	660	8.1257	0.0254	1.0	0.9749
5.8601	8.0	755	5.3355	0.0219	1.0	0.9780
4.3301	9.0	850	4.2174	0.0187	1.0	0.9812
4.1288	10.0	945	3.9223	0.0191	1.0	0.9808
3.7416	11.0	1040	3.9099	0.0188	1.0	0.9812
4.0712	12.0	1135	3.9099	0.0181	1.0	0.9818
3.7301	13.0	1230	3.9099	0.0185	1.0	0.9814
4.0605	14.0	1325	3.9098	0.0187	1.0	0.9812
4.1702	14.9019	1410	3.9099	0.0187	1.0	0.9812

pawmeow/wav2vec2-bn-checkpoints

wav2vec2-bn-checkpoints