BART_pretrained_on_billsum_finetuned_on_small_SCOTUS_extracted_dataset_2

This model is a fine-tuned version of bheshaj/bart-large-billsum-epochs20 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 9.5858
• Rouge1: 0.06
• Rouge2: 0.0
• Rougel: 0.06
• Rougelsum: 0.0598
• Gen Len: 20.0

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: 0.02
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
23.3274	0.98	10	35.8778	0.0289	0.0	0.0288	0.0289	20.0
43.999	1.98	20	39.9056	0.0475	0.0045	0.0453	0.0453	20.0
32.0042	2.98	30	25.4613	0.0516	0.0005	0.0499	0.0499	20.0
21.3151	3.98	40	17.6485	0.0021	0.0	0.0021	0.0021	20.0
15.4017	4.98	50	12.6187	0.06	0.0	0.06	0.0598	20.0
10.9491	5.98	60	9.5858	0.06	0.0	0.06	0.0598	20.0

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1+cu116
• Datasets 2.10.1
• Tokenizers 0.13.2