Modelpublic

personalizedrefrigerator/trocr-base

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes21downloads

Model Card

TrOCR Small (Finetuned on French)

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

—Loss: 0.0871
—Model Preparation Time: 0.0066
—Cer: 0.0134
—Wer: 0.0350
—Ratio: 97.4287

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
—trainbatchsize: 8
—evalbatchsize: 8
—seed: 42
—optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: linear
—training_steps: 12000
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Cer	Wer	Ratio
1.359	0.0333	400	1.0638	0.0066	0.1715	0.3841	89.6898
1.1123	0.0667	800	1.2707	0.0066	0.1715	0.4437	89.7413
0.872	0.1	1200	0.9230	0.0066	0.1305	0.3576	91.0810
0.6831	0.1333	1600	0.7046	0.0066	0.1195	0.2848	92.7617
0.6659	0.1667	2000	0.5952	0.0066	0.0841	0.2450	94.7742
0.612	0.2	2400	0.5830	0.0066	0.1029	0.2517	93.2609
0.4865	0.2333	2800	0.5312	0.0066	0.0973	0.2318	94.4422
0.4922	0.2667	3200	0.5842	0.0066	0.0918	0.1987	95.5454
0.4152	0.3	3600	0.4215	0.0066	0.0664	0.1921	95.9747
0.3648	0.3333	4000	0.4264	0.0066	0.0608	0.1722	96.4262
0.3272	0.3667	4400	0.5209	0.0066	0.0653	0.1921	95.3742
0.3172	0.4	4800	0.4229	0.0066	0.0531	0.1788	95.8131
0.2672	0.4333	5200	0.4071	0.0066	0.0586	0.1921	96.0787
0.2747	0.4667	5600	0.3494	0.0066	0.0586	0.1656	96.2269
0.2576	0.5	6000	0.3687	0.0066	0.0642	0.1523	96.6562
0.2138	0.5333	6400	0.3945	0.0066	0.0564	0.1391	96.8775
0.2197	0.5667	6800	0.3698	0.0066	0.0420	0.1391	97.3293
0.1908	1.0141	7200	0.3288	0.0066	0.0420	0.1126	97.6500
0.145	1.0474	7600	0.2655	0.0066	0.0332	0.0993	97.6602
0.1347	1.0808	8000	0.2659	0.0066	0.0365	0.1258	97.2893
0.1092	1.1141	8400	0.2496	0.0066	0.0343	0.1192	97.6671
0.111	1.1474	8800	0.2205	0.0066	0.0221	0.0861	98.4693
0.1033	1.1807	9200	0.2226	0.0066	0.0254	0.0927	98.1761
0.0919	1.2141	9600	0.1787	0.0066	0.0210	0.0728	98.6440
0.074	1.2474	10000	0.1756	0.0066	0.0188	0.0464	99.3030
0.0833	1.2808	10400	0.1830	0.0066	0.0232	0.0728	98.8762
0.057	1.3141	10800	0.1675	0.0066	0.0133	0.0530	99.3276
0.038	1.3474	11200	0.1709	0.0066	0.0188	0.0596	98.9563
0.0409	1.3807	11600	0.1400	0.0066	0.0133	0.0530	99.2905
0.0404	1.4141	12000	0.1423	0.0066	0.0122	0.0464	99.2598

Framework versions

—Transformers 4.46.3
—Pytorch 2.5.1+cu121
—Tokenizers 0.20.3