legal-bert-base-uncased-regnlp-obligation-classifier

This model is a fine-tuned version of nlpaueb/legal-bert-base-uncased on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0205
• Accuracy: 0.9957
• F1: 0.9964
• Precision: 1.0
• Recall: 0.9928

This repo is created for easy sharing model weights among our team which is participating in the RegNLP challenge. Please don't request access for it

You can find the finetuning scripts at this Github Repo.

All credits for the work and data belong to the RegNLP Paper authors.

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.4237	1.0	115	0.1827	0.9783	0.9819	0.9855	0.9784
0.0429	2.0	230	0.0438	0.9870	0.9892	0.9928	0.9856
0.0781	3.0	345	0.0579	0.9891	0.9910	0.9928	0.9892
0.009	4.0	460	0.0753	0.9891	0.9909	1.0	0.9820
0.0004	5.0	575	0.0205	0.9957	0.9964	1.0	0.9928
0.0001	6.0	690	0.0640	0.9913	0.9928	0.9964	0.9892
0.0001	7.0	805	0.0671	0.9913	0.9928	0.9964	0.9892
0.0001	8.0	920	0.0696	0.9913	0.9928	0.9964	0.9892

Framework versions

• Transformers 4.43.1
• Pytorch 2.3.0+cu121
• Datasets 2.14.7
• Tokenizers 0.19.1