my_awesome_billsum_model

This model is a fine-tuned version of t5-small on the billsum dataset. It achieves the following results on the evaluation set:

• Loss: 2.5057
• Rouge1: 0.1437
• Rouge2: 0.0544
• Rougel: 0.12
• Rougelsum: 0.12
• Gen Len: 19.0

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: 2e-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
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
No log	1.0	62	2.7978	0.1276	0.039	0.1077	0.1076	19.0
No log	2.0	124	2.5889	0.1371	0.0489	0.1153	0.1151	19.0
No log	3.0	186	2.5234	0.1429	0.054	0.1196	0.1194	19.0
No log	4.0	248	2.5057	0.1437	0.0544	0.12	0.12	19.0

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu116
• Datasets 2.8.0
• Tokenizers 0.13.2