lilt-en-funsd

This model is a fine-tuned version of SCUT-DLVCLab/lilt-roberta-en-base on the funsd-layoutlmv3 dataset. It achieves the following results on the evaluation set:

• Loss: 1.7928
• Answer: {'precision': 0.8716763005780347, 'recall': 0.9228886168910648, 'f1': 0.8965517241379309, 'number': 817}
• Header: {'precision': 0.5648148148148148, 'recall': 0.5126050420168067, 'f1': 0.5374449339207047, 'number': 119}
• Question: {'precision': 0.8945454545454545, 'recall': 0.9136490250696379, 'f1': 0.9039963252181902, 'number': 1077}
• Overall Precision: 0.8678
• Overall Recall: 0.8937
• Overall F1: 0.8806
• Overall Accuracy: 0.7985

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 2500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Answer	Header	Question	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
0.4236	10.53	200	0.9583	{'precision': 0.8623962040332147, 'recall': 0.8898408812729498, 'f1': 0.8759036144578314, 'number': 817}	{'precision': 0.5131578947368421, 'recall': 0.3277310924369748, 'f1': 0.39999999999999997, 'number': 119}	{'precision': 0.8450704225352113, 'recall': 0.947075208913649, 'f1': 0.893169877408056, 'number': 1077}	0.8401	0.8872	0.8630	0.8016
0.0421	21.05	400	1.4064	{'precision': 0.8573113207547169, 'recall': 0.8898408812729498, 'f1': 0.8732732732732732, 'number': 817}	{'precision': 0.4301675977653631, 'recall': 0.6470588235294118, 'f1': 0.5167785234899329, 'number': 119}	{'precision': 0.8667883211678832, 'recall': 0.8820798514391829, 'f1': 0.87436723423838, 'number': 1077}	0.8262	0.8713	0.8482	0.7733
0.0121	31.58	600	1.5114	{'precision': 0.8534090909090909, 'recall': 0.9192166462668299, 'f1': 0.8850913376546846, 'number': 817}	{'precision': 0.5930232558139535, 'recall': 0.42857142857142855, 'f1': 0.4975609756097561, 'number': 119}	{'precision': 0.8824577025823687, 'recall': 0.9201485608170845, 'f1': 0.9009090909090909, 'number': 1077}	0.8583	0.8907	0.8742	0.8044
0.0058	42.11	800	1.4988	{'precision': 0.8361391694725028, 'recall': 0.9118727050183598, 'f1': 0.8723653395784543, 'number': 817}	{'precision': 0.5203252032520326, 'recall': 0.5378151260504201, 'f1': 0.5289256198347108, 'number': 119}	{'precision': 0.8798206278026905, 'recall': 0.9108635097493036, 'f1': 0.8950729927007299, 'number': 1077}	0.8408	0.8892	0.8643	0.7982
0.004	52.63	1000	1.5823	{'precision': 0.8455467869222097, 'recall': 0.9179926560587516, 'f1': 0.880281690140845, 'number': 817}	{'precision': 0.5263157894736842, 'recall': 0.5042016806722689, 'f1': 0.5150214592274679, 'number': 119}	{'precision': 0.867595818815331, 'recall': 0.924791086350975, 'f1': 0.8952808988764045, 'number': 1077}	0.8404	0.8972	0.8679	0.7996
0.0028	63.16	1200	1.6518	{'precision': 0.8492822966507177, 'recall': 0.8690330477356181, 'f1': 0.8590441621294616, 'number': 817}	{'precision': 0.5855855855855856, 'recall': 0.5462184873949579, 'f1': 0.5652173913043478, 'number': 119}	{'precision': 0.88, 'recall': 0.9192200557103064, 'f1': 0.899182561307902, 'number': 1077}	0.8518	0.8768	0.8641	0.7939
0.0013	73.68	1400	1.8819	{'precision': 0.8378672470076169, 'recall': 0.9424724602203183, 'f1': 0.8870967741935485, 'number': 817}	{'precision': 0.6794871794871795, 'recall': 0.44537815126050423, 'f1': 0.5380710659898478, 'number': 119}	{'precision': 0.9006622516556292, 'recall': 0.8839368616527391, 'f1': 0.8922211808809747, 'number': 1077}	0.8642	0.8818	0.8729	0.7931
0.0013	84.21	1600	1.8234	{'precision': 0.8519362186788155, 'recall': 0.9155446756425949, 'f1': 0.8825958702064898, 'number': 817}	{'precision': 0.5585585585585585, 'recall': 0.5210084033613446, 'f1': 0.5391304347826087, 'number': 119}	{'precision': 0.9120982986767486, 'recall': 0.8960074280408542, 'f1': 0.9039812646370023, 'number': 1077}	0.8671	0.8818	0.8744	0.7996
0.0008	94.74	1800	1.7898	{'precision': 0.844170403587444, 'recall': 0.9216646266829865, 'f1': 0.8812170860152135, 'number': 817}	{'precision': 0.5294117647058824, 'recall': 0.5294117647058824, 'f1': 0.5294117647058824, 'number': 119}	{'precision': 0.8756613756613757, 'recall': 0.9220055710306406, 'f1': 0.898236092265943, 'number': 1077}	0.8434	0.8987	0.8701	0.7901
0.0004	105.26	2000	1.8115	{'precision': 0.8396436525612472, 'recall': 0.9228886168910648, 'f1': 0.8793002915451895, 'number': 817}	{'precision': 0.6063829787234043, 'recall': 0.4789915966386555, 'f1': 0.5352112676056338, 'number': 119}	{'precision': 0.8909090909090909, 'recall': 0.9099350046425255, 'f1': 0.90032154340836, 'number': 1077}	0.8561	0.8897	0.8726	0.7939
0.0004	115.79	2200	1.7928	{'precision': 0.8716763005780347, 'recall': 0.9228886168910648, 'f1': 0.8965517241379309, 'number': 817}	{'precision': 0.5648148148148148, 'recall': 0.5126050420168067, 'f1': 0.5374449339207047, 'number': 119}	{'precision': 0.8945454545454545, 'recall': 0.9136490250696379, 'f1': 0.9039963252181902, 'number': 1077}	0.8678	0.8937	0.8806	0.7985
0.0003	126.32	2400	1.8271	{'precision': 0.863013698630137, 'recall': 0.9253365973072215, 'f1': 0.8930891907855877, 'number': 817}	{'precision': 0.6105263157894737, 'recall': 0.48739495798319327, 'f1': 0.5420560747663552, 'number': 119}	{'precision': 0.8935395814376706, 'recall': 0.9117920148560817, 'f1': 0.9025735294117648, 'number': 1077}	0.8676	0.8922	0.8797	0.7983

Framework versions

• Transformers 4.25.1
• Pytorch 1.12.1+cu102
• Datasets 2.8.0
• Tokenizers 0.13.1