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
8.537	1.0	479	6.2801	1.0
4.1924	2.0	958	3.7917	1.0
3.518	3.0	1437	3.3861	0.9980
2.9231	4.0	1916	2.4444	0.7940
2.3797	5.0	2395	1.4271	0.3890
2.0441	6.0	2874	1.0737	0.2580
1.8602	7.0	3353	0.9219	0.1967
1.7634	8.0	3832	0.8330	0.1937
1.7072	9.0	4311	0.7777	0.1852
1.6978	10.0	4790	0.7503	0.1760
1.6251	11.0	5269	0.7250	0.1744
1.6029	12.0	5748	0.7061	0.1723
1.5616	13.0	6227	0.6999	0.1657
1.5867	14.0	6706	0.7165	0.1636
1.5382	15.0	7185	0.6676	0.1618
1.5525	16.0	7664	0.6844	0.1710
1.5822	17.0	8143	0.7085	0.1638
1.6695	18.0	8622	0.7505	0.1866
1.7377	19.0	9101	0.7970	0.2001
1.807	20.0	9580	0.8281	0.2602
2.0481	21.0	10059	1.1861	0.2890

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

xls-r-300m-gaphon-collapsed-v2