sinhala-ocr-model-v3

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

• Loss: 4.7242
• Cer: 0.2764

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
3.6711	6.54	500	4.9311	0.4178
2.3499	13.07	1000	4.5366	0.3482
1.5601	19.61	1500	4.4634	0.3204
0.987	26.14	2000	4.4804	0.3011
0.6487	32.68	2500	4.6310	0.2863
0.3816	39.22	3000	4.6093	0.2788
0.3494	45.75	3500	4.6291	0.2827
0.2357	52.29	4000	4.6399	0.2780
0.2188	58.82	4500	4.6313	0.2798
0.1413	65.36	5000	4.6828	0.2768
0.0985	71.9	5500	4.7135	0.2772
0.1086	78.43	6000	4.7242	0.2764

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.0
• Datasets 2.16.0
• Tokenizers 0.15.0