Kwan0/layoutlmv3-base-finetune-DocLayNet-100k

layoutlmv3-finetuned-funsd

This model is a fine-tuned version of microsoft/layoutlmv3-base on the pierreguillou/DocLayNet-large using bounding boxes and categories for lines (not for for paragraphs). It achieves the following results on the evaluation set:

• —Loss: 0.33888205885887146,
• —Precision: 0.8478835766832817,
• —Recall: 0.8934488524091807,
• —F1: 0.8700700634847538,
• —Accuracy: 0.9574140990541197

The script for training can be found here: https://github.com/huggingface/transformers/tree/main/examples/research_projects/layoutlmv3

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 1e-05
• —trainbatchsize: 2
• —evalbatchsize: 2
• —training_steps: 100000

Framework versions

• —Transformers 4.33.3
• —Pytorch 1.11.0+cu115
• —Datasets 2.14.5
• —Tokenizers 0.13.3