SriDoc/TrOCR-Sin-Printed

TrOCR-SIN(DeiT)

This model is a fine-tuned version of on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4335
• Cer: 0.1445

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 75000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
1.3019	1.78	5000	0.6416	1.7769
0.6387	3.55	10000	0.4048	0.8457
0.3402	5.33	15000	0.2808	0.6898
0.1332	7.11	20000	0.2377	0.5765
0.1141	8.89	25000	0.2223	0.4460
0.0481	10.66	30000	0.1868	0.4128
0.0391	12.44	35000	0.1563	0.4172
0.0357	14.22	40000	0.1981	0.4756
0.0215	16.0	45000	0.1983	0.5838
0.0129	17.77	50000	0.1757	0.5511
0.0087	19.55	55000	0.1699	0.5568
0.003	21.33	60000	0.1648	0.4532
0.0042	23.11	65000	0.1582	0.4650
0.0066	24.88	70000	0.1654	0.4740
0.0014	26.66	75000	0.1448	0.4337

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.1