learning_rate: 0.0001
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
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: 5000
num_epochs: 8
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.6196	0.3909	500	inf	100.0	65.7730
1.0801	0.7819	1000	inf	38.9103	10.8535
0.69	1.1728	1500	inf	28.1298	7.8743
0.5238	1.5637	2000	inf	24.6450	6.9218
0.4787	1.9547	2500	inf	23.5140	6.6324
0.41	2.3456	3000	inf	23.2639	6.4844
0.3868	2.7365	3500	inf	23.0202	6.4139
0.4987	3.1274	4000	inf	22.1255	6.1609
0.5173	3.5184	4500	inf	22.0348	6.1361
0.5159	3.9093	5000	inf	22.2732	6.2526
0.2712	4.3002	5500	inf	22.0759	6.1940
0.2565	4.6912	6000	inf	21.4397	5.9482
0.4129	5.0821	6500	inf	21.1116	5.8439
0.3925	5.4730	7000	inf	21.1506	5.9633
0.3943	5.8640	7500	inf	19.8160	5.4331
0.1971	6.2549	8000	inf	18.7778	5.1171
0.1898	6.6458	8500	inf	18.2018	4.8967
0.2553	7.0367	9000	inf	18.1142	4.9026
0.2291	7.4277	9500	inf	17.5730	4.7604
0.2655	7.8186	10000	inf	17.8979	4.8286

davidilag/wav2vec2-xls-r-1b-scandinavian-E4-100h-30-epochs-20250208_v3

wav2vec2-xls-r-1b-scandinavian-E4-100h-30-epochs-20250208_v3