distilbert-legal-definitions

This model is a fine-tuned version of distilbert-base-cased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0034
• Precision: 0.9668
• Recall: 0.9707
• Macro F1: 0.9688
• Micro F1: 0.9688
• Accuracy: 0.9994
• Term F1: 0.9688
• Term Precision: 0.9668
• Term Recall: 0.9707

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	Macro F1	Micro F1	Accuracy	Term F1	Term Precision	Term Recall
0.0049	1.0	2325	0.0034	0.9790	0.9580	0.9684	0.9684	0.9993	0.9684	0.9790	0.9580
0.0023	2.0	4650	0.0032	0.9669	0.9786	0.9727	0.9727	0.9994	0.9727	0.9669	0.9786
0.0013	3.0	6975	0.0018	0.9836	0.9794	0.9815	0.9815	0.9997	0.9815	0.9836	0.9794
0.0006	4.0	9300	0.0016	0.9879	0.9828	0.9854	0.9854	0.9997	0.9854	0.9879	0.9828

Framework versions

• Transformers 4.21.3
• Pytorch 1.12.1+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1