learning_rate: 0.0003
train_batch_size: 8
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
training_steps: 8000
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer
19.1252	0.9390	400	3.4716	100.0
6.2270	1.8779	800	3.4558	100.0
4.0283	2.8169	1200	3.3956	100.0
2.8016	3.7559	1600	3.2941	99.96
2.2193	4.6948	2000	3.2600	100.0
1.7547	5.6338	2400	3.2170	99.76
1.3964	6.5728	2800	3.3102	101.75
1.1963	7.5117	3200	3.1533	100.75
0.9914	8.4507	3600	3.2595	99.84
0.9010	9.3897	4000	3.1565	101.07
0.8122	10.3286	4400	3.1109	99.64
0.7482	11.2676	4800	3.0935	99.72
0.6616	12.2066	5200	3.1038	99.88
0.6589	13.1455	5600	3.1592	100.6
0.6076	14.0845	6000	3.0983	99.68
0.5958	15.0235	6400	3.1282	99.84
0.6226	15.9624	6800	3.2371	98.77
0.6571	16.9014	7200	3.2466	99.92
0.6573	17.8404	7600	3.3030	99.92
0.6681	18.7793	8000	3.2899	99.92

GwadaDLT/wav2vec2-gcf-r11

wav2vec2-gcf-r11