learning_rate: 3e-05
train_batch_size: 4
eval_batch_size: 8
seed: 42
optimizer: Use adamw_torch 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: 2000
num_epochs: 30
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer
3.2047	1.0	672	2.8873	1.0
2.6034	2.0	1344	2.5893	0.9774
2.2041	3.0	2016	2.1982	0.9428
2.0373	4.0	2688	1.8445	0.9076
1.5318	5.0	3360	0.9850	0.4850
1.1824	6.0	4032	0.5831	0.1897
1.0263	7.0	4704	0.4764	0.1663
0.9864	8.0	5376	0.4037	0.1554
0.8761	9.0	6048	0.3606	0.1496
0.877	10.0	6720	0.3315	0.1446
0.841	11.0	7392	0.3187	0.1425
0.8028	12.0	8064	0.3008	0.1397
0.7991	13.0	8736	0.2966	0.1385
0.7624	14.0	9408	0.2879	0.1348
0.7529	15.0	10080	0.2796	0.1337
0.7272	16.0	10752	0.2720	0.1329
0.702	17.0	11424	0.2665	0.1315
0.7025	18.0	12096	0.2664	0.1290
0.7048	19.0	12768	0.2571	0.1288
0.6472	20.0	13440	0.2581	0.1293
0.6599	21.0	14112	0.2570	0.1286
0.671	22.0	14784	0.2528	0.1298
0.6698	23.0	15456	0.2487	0.1284
0.6461	24.0	16128	0.2517	0.1283
0.6589	25.0	16800	0.2473	0.1270
0.6527	26.0	17472	0.2498	0.1262
0.6205	27.0	18144	0.2484	0.1272
0.6406	28.0	18816	0.2487	0.1265
0.6154	29.0	19488	0.2487	0.1273
0.6234	30.0	20160	0.2485	0.1272

duck-hug-567/xls-r-300m-engphon-collapsed-v2

xls-r-300m-engphon-collapsed-v2