Modelpublic

scales-okn/ontology-consent-judgment-resolution

sourceHugging Facegpl-3.0updated 2mo agoView on Hugging Face

0likes16downloads

Model Card

ontology-consent-judgment-resolution

This model is a fine-tuned version of scales-okn/docket-language-model on the None dataset. It achieves the following results on the evaluation set:

—Loss: 0.2063
—Accuracy: 0.9618
—F1: 0.9318
—Precision: 0.8723
—Recall: 1.0

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: 3e-05
—trainbatchsize: 4
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 8
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—lrschedulerwarmup_ratio: 0.06
—num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.0203	0.89	100	0.2183	0.9363	0.8889	0.8163	0.9756
0.1801	1.79	200	0.2063	0.9618	0.9318	0.8723	1.0
0.001	2.68	300	0.2282	0.9618	0.9318	0.8723	1.0
0.0004	3.57	400	0.2530	0.9618	0.9318	0.8723	1.0
0.0003	4.46	500	0.2977	0.9618	0.9318	0.8723	1.0

Framework versions

—Transformers 4.23.1
—Pytorch 1.13.0+cu117
—Datasets 2.6.1
—Tokenizers 0.13.1

Public release information

This model is released by the SCALES Open Knowledge Network under the GNU General Public License v3.0. It is derived from scales-okn/docket-language-model and is intended for research and development involving legal-document classification or information extraction. It is not legal advice.

The organization has reviewed the release decision and confirmed that the model's training data and resulting weights are legally and ethically releasable. Users are responsible for evaluating accuracy, bias, privacy, and fitness for their own use.

The repository includes PyTorch .bin artifacts. Hugging Face's server-side security scan reported no file issues before publication. As with any serialized model artifact, load it only with maintained libraries and in an appropriately isolated environment.