trocr-bigram4-BY-450k

This model is a fine-tuned version of cyttic/exp2-frozen-benyehuda-cont on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.3969
Cer: 0.0154
Wer: 0.0479

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 16
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: 8400
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
4.0171	0.0714	2000	1.7101	0.2032	0.4148
3.1964	0.1429	4000	1.3900	0.1315	0.2951
2.8672	0.2143	6000	1.2906	0.1099	0.2564
2.7257	0.2857	8000	1.1934	0.0993	0.2372
2.3802	0.3571	10000	1.0854	0.0801	0.2089
2.2460	0.4286	12000	0.9850	0.0716	0.1887
2.1261	0.5	14000	0.9230	0.0641	0.1727
2.0339	0.5714	16000	0.8716	0.0590	0.1564
1.9418	0.6429	18000	0.8187	0.0498	0.1379
1.8719	0.7143	20000	0.7716	0.0456	0.1267
1.7357	0.7857	22000	0.7384	0.0442	0.1271
1.7185	0.8571	24000	0.7005	0.0414	0.1184
1.6163	0.9286	26000	0.6719	0.0365	0.1075
1.5521	1.0	28000	0.6427	0.0322	0.0990
1.2175	1.0714	30000	0.6224	0.0351	0.0988
1.2979	1.1429	32000	0.5963	0.0322	0.0929
1.1303	1.2143	34000	0.5869	0.0301	0.0890
1.0653	1.2857	36000	0.5731	0.0285	0.0860
1.1797	1.3571	38000	0.5536	0.0255	0.0782
1.1670	1.4286	40000	0.5495	0.0272	0.0812
0.9935	1.5	42000	0.5208	0.0243	0.0746
1.0177	1.5714	44000	0.5122	0.0241	0.0730
1.0507	1.6429	46000	0.4999	0.0242	0.0740
0.9875	1.7143	48000	0.4906	0.0222	0.0679
0.9788	1.7857	50000	0.4825	0.0192	0.0623
0.9224	1.8571	52000	0.4694	0.0193	0.0620
0.9621	1.9286	54000	0.4567	0.0196	0.0600
0.9634	2.0	56000	0.4493	0.0178	0.0563
0.8980	2.0714	58000	0.4430	0.0188	0.0579
0.7686	2.1429	60000	0.4360	0.0195	0.0571
0.7952	2.2143	62000	0.4306	0.0187	0.0565
0.7972	2.2857	64000	0.4282	0.0179	0.0550
0.7620	2.3571	66000	0.4236	0.0184	0.0555
0.7173	2.4286	68000	0.4195	0.0174	0.0540
0.7364	2.5	70000	0.4131	0.0173	0.0526
0.7875	2.5714	72000	0.4133	0.0170	0.0529
0.7530	2.6429	74000	0.4079	0.0172	0.0521
0.7464	2.7143	76000	0.4041	0.0168	0.0517
0.7617	2.7857	78000	0.4019	0.0165	0.0510
0.7221	2.8571	80000	0.4005	0.0164	0.0503
0.7381	2.9286	82000	0.3979	0.0163	0.0502
0.7615	3.0	84000	0.3969	0.0154	0.0479

Framework versions

Transformers 5.15.0
Pytorch 2.11.0+cu128
Datasets 5.0.1
Tokenizers 0.22.2

cyttic/trocr-bigram4-BY-450k

trocr-bigram4-BY-450k

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Seed this model