learning_rate: 3e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
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
10.6762	1.0	479	8.1054	1.0
4.8891	2.0	958	4.1721	1.0
3.6215	3.0	1437	3.4540	1.0
3.1004	4.0	1916	2.9628	0.9844
2.79	5.0	2395	2.4702	0.8111
2.4069	6.0	2874	1.9406	0.6473
2.1003	7.0	3353	1.4747	0.4970
1.8618	8.0	3832	1.2153	0.3861
1.7019	9.0	4311	1.0300	0.3043
1.6456	10.0	4790	0.9045	0.2520
1.5454	11.0	5269	0.8245	0.2346
1.453	12.0	5748	0.7660	0.2212
1.3653	13.0	6227	0.7162	0.2044
1.3144	14.0	6706	0.6711	0.1974
1.2981	15.0	7185	0.6509	0.1801
1.2577	16.0	7664	0.6265	0.1812
1.2226	17.0	8143	0.6019	0.1787
1.206	18.0	8622	0.5879	0.1707
1.2062	19.0	9101	0.5715	0.1638
1.1892	20.0	9580	0.5666	0.1643
1.1798	21.0	10059	0.5556	0.1577
1.1321	22.0	10538	0.5481	0.1560
1.1376	23.0	11017	0.5414	0.1553
1.1062	24.0	11496	0.5306	0.1510
1.1071	25.0	11975	0.5339	0.1523
1.1203	26.0	12454	0.5238	0.1493
1.0882	27.0	12933	0.5190	0.1481
1.0983	28.0	13412	0.5172	0.1488
1.0942	29.0	13891	0.5148	0.1490
1.1035	29.9383	14340	0.5146	0.1486

duck-hug-567/xls-r-300m-paired

xls-r-300m-paired