personalizedrefrigerator/trocr-small-fr

TrOCR Small (Finetuned on French)

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

• Loss: 0.1007
• Model Preparation Time: 0.0057
• Cer: 0.0138
• Wer: 0.0455
• Ratio: 98.3979

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

See https://github.com/personalizedrefrigerator/trocr_finetuning/tree/main/trocr

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Cer	Wer	Ratio
0.1386	0.0333	400	0.1543	0.0057	0.0190	0.0819	98.6922
0.1298	0.0667	800	0.1300	0.0057	0.0130	0.0526	98.9649
0.1171	0.1	1200	0.1622	0.0057	0.0200	0.0760	98.3437
0.1035	0.1333	1600	0.1538	0.0057	0.0190	0.0760	98.6841
0.1186	0.1667	2000	0.1605	0.0057	0.0170	0.0760	98.9547
0.1285	0.2	2400	0.1675	0.0057	0.0190	0.0643	98.5663
0.1043	0.2333	2800	0.1511	0.0057	0.0220	0.0702	98.4283
0.1294	0.2667	3200	0.1647	0.0057	0.0150	0.0526	98.9361
0.0954	0.3	3600	0.1532	0.0057	0.0160	0.0526	98.7555
0.111	0.3333	4000	0.1577	0.0057	0.0210	0.0643	98.1890
0.114	0.3667	4400	0.1378	0.0057	0.0160	0.0585	98.6565
0.1183	0.4	4800	0.1163	0.0057	0.0070	0.0351	99.3075
0.1277	0.4333	5200	0.1571	0.0057	0.0160	0.0760	98.8328
0.1219	0.4667	5600	0.1571	0.0057	0.0150	0.0526	98.7910
0.1101	0.5	6000	0.1245	0.0057	0.0130	0.0526	99.0524
0.1069	0.5333	6400	0.1470	0.0057	0.0130	0.0585	99.0389
0.1126	0.5667	6800	0.1302	0.0057	0.0140	0.0526	98.9437
0.0837	1.0137	7200	0.1323	0.0057	0.0200	0.0702	98.4624
0.0809	1.047	7600	0.1180	0.0057	0.0100	0.0409	99.4630
0.0889	1.0803	8000	0.1241	0.0057	0.0180	0.0702	98.7486
0.0711	1.1137	8400	0.1174	0.0057	0.0150	0.0585	98.8769
0.0736	1.147	8800	0.1166	0.0057	0.0120	0.0468	99.0708
0.0786	1.1803	9200	0.1080	0.0057	0.0080	0.0351	99.5225
0.0686	1.2137	9600	0.1037	0.0057	0.0070	0.0292	99.5887
0.0738	1.2470	10000	0.1127	0.0057	0.0140	0.0468	99.0132
0.07	1.2803	10400	0.1051	0.0057	0.0120	0.0409	99.0954
0.0697	1.3137	10800	0.1003	0.0057	0.0090	0.0292	99.2171
0.0686	1.347	11200	0.1038	0.0057	0.0120	0.0351	98.9317
0.0763	1.3803	11600	0.1028	0.0057	0.0120	0.0351	98.9317
0.0717	1.4137	12000	0.1018	0.0057	0.0120	0.0351	98.9317

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3