README.md

metadata

tags:
  - generated_from_trainer
  - trocr
language: ar
model-index:
  - name: TrOCR-Ar-Small
    results: []

TrOCR-Ar-Small

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

• Loss: 2.2771
• Cer: 0.8211

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: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
3.6363	0.14	1000	2.7594	0.9370
2.7508	0.29	2000	2.6589	0.8901
2.6519	0.43	3000	2.6059	0.8647
2.5936	0.57	4000	2.5360	0.7941
2.5069	0.72	5000	2.4701	0.8262
2.4606	0.86	6000	2.4427	0.7552
2.4046	1.0	7000	2.4262	0.7822
2.3628	1.15	8000	2.3880	0.8186
2.3458	1.29	9000	2.3589	0.8262
2.3062	1.43	10000	2.3704	0.8693
2.2884	1.58	11000	2.3065	0.8034
2.263	1.72	12000	2.3413	0.8545
2.2473	1.86	13000	2.3314	0.7996
2.2318	2.01	14000	2.3034	0.8254
2.2004	2.15	15000	2.3068	0.8461
2.1774	2.29	16000	2.2799	0.8207
2.1684	2.44	17000	2.2746	0.8249
2.1637	2.58	18000	2.2540	0.7797
2.1418	2.72	19000	2.2595	0.7937
2.1309	2.87	20000	2.2771	0.8211

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.0+cu111
• Datasets 2.0.0
• Tokenizers 0.11.6