legal-flan-t5-base

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

Legal Flan-T5-Base

This model is a fine-tuned version of google/flan-t5-base on an LegalBench dataset. It achieves the following results on the evaluation set:

• Loss: 1.1885
• Rouge1: 65.4762
• Rouge2: 0.0
• Rougel: 65.4762
• Rougelsum: 65.4762
• Gen Len: 2.1905

Model description

We finetune Flan-T5-Base LLM on the LegalBench.

Prompt

The prompt should be formatted as follows: {{Question}} {{Clause}}

Question: Does the clause grant one party an “enterprise,” “all you can eat” or unlimited usage license?

Clause: Except as the parties may otherwise agree in writing, Converge, to the extent it has the legal right to do so, hereby grants to Vert an irrevocable, perpetual, world-wide, non-exclusive right and license to use, load, store, transmit, execute, copy, market, distribute, in any medium or distribution technology whatsoever, known or unknown, display, perform and sublicense the Converge-Independent Materials and the Third-Party Materials, in both Source Code and Object Code formats, and to make unlimited instantiations thereof, for any and all purposes.

Prompt: Does the clause grant one party an “enterprise,” “all you can eat” or unlimited usage license? Except as the parties may otherwise agree in writing, Converge, to the extent it has the legal right to do so, hereby grants to Vert an irrevocable, perpetual, world-wide, non-exclusive right and license to use, load, store, transmit, execute, copy, market, distribute, in any medium or distribution technology whatsoever, known or unknown, display, perform and sublicense the Converge-Independent Materials and the Third-Party Materials, in both Source Code and Object Code formats, and to make unlimited instantiations thereof, for any and all purposes.

Intended uses & limitations

More information needed

Training and evaluation data

We used LegalBench for training and evaluation.

Training procedure

Tutorial: Finetune Flan-T5 with Baseten.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-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: 20
• num_epochs: 50

Training results

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1+cu117
• Datasets 2.10.1
• Tokenizers 0.13.2