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: 1.6853
  • Answer: {'precision': 0.8719723183391004, 'recall': 0.9253365973072215, 'f1': 0.8978622327790974, 'number': 817}
  • Header: {'precision': 0.6224489795918368, 'recall': 0.5126050420168067, 'f1': 0.5622119815668203, 'number': 119}
  • Question: {'precision': 0.908411214953271, 'recall': 0.9025069637883009, 'f1': 0.9054494643688868, 'number': 1077}
  • Overall Precision: 0.8791
  • Overall Recall: 0.8887
  • Overall F1: 0.8839
  • Overall Accuracy: 0.8067

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
0.4358 10.53 200 1.0235 {'precision': 0.8292964244521338, 'recall': 0.8800489596083231, 'f1': 0.8539192399049881, 'number': 817} {'precision': 0.4657534246575342, 'recall': 0.5714285714285714, 'f1': 0.5132075471698114, 'number': 119} {'precision': 0.8694852941176471, 'recall': 0.8783658310120706, 'f1': 0.8739030023094689, 'number': 1077} 0.8248 0.8609 0.8425 0.7868
0.0513 21.05 400 1.2438 {'precision': 0.8439306358381503, 'recall': 0.8935128518971848, 'f1': 0.8680142687277052, 'number': 817} {'precision': 0.7045454545454546, 'recall': 0.5210084033613446, 'f1': 0.5990338164251209, 'number': 119} {'precision': 0.8937893789378938, 'recall': 0.9220055710306406, 'f1': 0.9076782449725778, 'number': 1077} 0.8648 0.8867 0.8756 0.8066
0.0139 31.58 600 1.3473 {'precision': 0.8443935926773455, 'recall': 0.9033047735618115, 'f1': 0.872856298048492, 'number': 817} {'precision': 0.6111111111111112, 'recall': 0.5546218487394958, 'f1': 0.5814977973568282, 'number': 119} {'precision': 0.8945945945945946, 'recall': 0.9220055710306406, 'f1': 0.9080932784636488, 'number': 1077} 0.8590 0.8927 0.8755 0.8101
0.0087 42.11 800 1.3432 {'precision': 0.8778718258766627, 'recall': 0.8886168910648715, 'f1': 0.8832116788321168, 'number': 817} {'precision': 0.5813953488372093, 'recall': 0.6302521008403361, 'f1': 0.6048387096774193, 'number': 119} {'precision': 0.9113573407202216, 'recall': 0.9164345403899722, 'f1': 0.9138888888888889, 'number': 1077} 0.8769 0.8882 0.8825 0.8161
0.0039 52.63 1000 1.5068 {'precision': 0.8678362573099415, 'recall': 0.9082007343941249, 'f1': 0.8875598086124402, 'number': 817} {'precision': 0.5564516129032258, 'recall': 0.5798319327731093, 'f1': 0.5679012345679013, 'number': 119} {'precision': 0.8998144712430427, 'recall': 0.9006499535747446, 'f1': 0.9002320185614848, 'number': 1077} 0.8658 0.8847 0.8752 0.8028
0.0028 63.16 1200 1.5721 {'precision': 0.8624277456647399, 'recall': 0.9130966952264382, 'f1': 0.8870392390011891, 'number': 817} {'precision': 0.6288659793814433, 'recall': 0.5126050420168067, 'f1': 0.5648148148148148, 'number': 119} {'precision': 0.9085714285714286, 'recall': 0.8857938718662952, 'f1': 0.8970380818053596, 'number': 1077} 0.8752 0.8748 0.8750 0.8145
0.0027 73.68 1400 1.5657 {'precision': 0.8695150115473441, 'recall': 0.9216646266829865, 'f1': 0.8948306595365418, 'number': 817} {'precision': 0.6310679611650486, 'recall': 0.5462184873949579, 'f1': 0.5855855855855856, 'number': 119} {'precision': 0.8944494995450409, 'recall': 0.9127205199628597, 'f1': 0.9034926470588234, 'number': 1077} 0.8709 0.8947 0.8826 0.8130
0.0012 84.21 1600 1.6853 {'precision': 0.8719723183391004, 'recall': 0.9253365973072215, 'f1': 0.8978622327790974, 'number': 817} {'precision': 0.6224489795918368, 'recall': 0.5126050420168067, 'f1': 0.5622119815668203, 'number': 119} {'precision': 0.908411214953271, 'recall': 0.9025069637883009, 'f1': 0.9054494643688868, 'number': 1077} 0.8791 0.8887 0.8839 0.8067
0.0007 94.74 1800 1.6321 {'precision': 0.8642117376294591, 'recall': 0.9192166462668299, 'f1': 0.8908659549228943, 'number': 817} {'precision': 0.5964912280701754, 'recall': 0.5714285714285714, 'f1': 0.5836909871244635, 'number': 119} {'precision': 0.9101964452759589, 'recall': 0.903435468895079, 'f1': 0.9068033550792172, 'number': 1077} 0.8733 0.8902 0.8817 0.8045
0.0004 105.26 2000 1.7732 {'precision': 0.8535469107551488, 'recall': 0.9130966952264382, 'f1': 0.8823181549379067, 'number': 817} {'precision': 0.5752212389380531, 'recall': 0.5462184873949579, 'f1': 0.5603448275862069, 'number': 119} {'precision': 0.8991825613079019, 'recall': 0.9192200557103064, 'f1': 0.909090909090909, 'number': 1077} 0.8625 0.8947 0.8783 0.7991
0.0003 115.79 2200 1.7988 {'precision': 0.8785714285714286, 'recall': 0.9033047735618115, 'f1': 0.8907664453832227, 'number': 817} {'precision': 0.6190476190476191, 'recall': 0.5462184873949579, 'f1': 0.5803571428571429, 'number': 119} {'precision': 0.8940639269406393, 'recall': 0.9090064995357474, 'f1': 0.9014732965009209, 'number': 1077} 0.8735 0.8852 0.8793 0.7950
0.0003 126.32 2400 1.8038 {'precision': 0.8584686774941995, 'recall': 0.9057527539779682, 'f1': 0.8814770696843359, 'number': 817} {'precision': 0.63, 'recall': 0.5294117647058824, 'f1': 0.5753424657534247, 'number': 119} {'precision': 0.8943533697632058, 'recall': 0.9117920148560817, 'f1': 0.9029885057471265, 'number': 1077} 0.8665 0.8867 0.8765 0.7953

Framework versions

  • Transformers 4.35.0
  • Pytorch 2.1.0+cu118
  • Datasets 2.14.6
  • Tokenizers 0.14.1
Downloads last month
38
Safetensors
Model size
130M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for T-Brockhouse/lilt-en-funsd

Finetuned
(46)
this model