learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
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
num_epochs: 10
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer
4.9721	0.7722	300	0.9371	0.2075
1.4698	1.5431	600	0.8236	0.1890
1.1612	2.3140	900	0.7892	0.1803
0.9061	3.0849	1200	0.7871	0.1841
0.6954	3.8571	1500	0.8228	0.1732
0.5219	4.6281	1800	0.9034	0.1777
0.3814	5.3990	2100	1.0512	0.1701
0.2725	6.1699	2400	1.1763	0.1751
0.1867	6.9421	2700	1.2330	0.1768
0.1216	7.7130	3000	1.3450	0.1704
0.0827	8.4839	3300	1.5420	0.1713
0.0515	9.2548	3600	1.6782	0.1687

dhasmana/Kumaoni-Bhojpuri-Chhattisgarhi-Angika-w2v-bert-2.0

Kumaoni-Bhojpuri-Chhattisgarhi-Angika-w2v-bert-2.0