norhand-hcr-beta1

This model is a fine-tuned version of microsoft/trocr-base-stage1 on the NorHand dataset. It achieves the following results on the evaluation set:

• Loss: 1.4687
• Cer: 0.0439

It achieves the following results on the test set:

• CER: 8.3
• WER: 22.07

This is work in progress.

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: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.7606	0.71	5000	1.0489	0.1266
0.5971	1.43	10000	1.2096	0.1028
0.4448	2.14	15000	1.3738	0.0771
0.3919	2.85	20000	1.2733	0.0670
0.2806	3.56	25000	1.3048	0.0597
0.2055	4.28	30000	1.3568	0.0557
0.2486	4.99	35000	0.9717	0.0527
0.1631	5.7	40000	1.3159	0.0506
0.1159	6.41	45000	1.3730	0.0480
0.0605	7.13	50000	1.3399	0.0469
0.0626	7.84	55000	1.3642	0.0460
0.0532	8.55	60000	1.3870	0.0450
0.0177	9.26	65000	1.4912	0.0439
0.0308	9.98	70000	1.4687	0.0439

Framework versions

• Transformers 4.32.1
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.13.3