BARTxiv

See the model implementation here.

This model is a fine-tuned version of facebook/bart-large-cnn on the arxiv-summarization dataset. It achieves the following results on the validation set:

• Loss: 0.86
• Rouge1: 41.70
• Rouge2: 15.13
• Rougel: 22.85
• Rougelsum: 37.77

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: 1e-6
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Adafactor
• num_epochs: 9

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
1.24	1.0	1073	1.24	38.32	12.80	20.55	34.50
1.04	2.0	2146	1.04	39.65	13.74	21.28	35.83
0.979	3.0	3219	0.98	40.19	14.30	21.87	36.38
0.970	4.0	4292	0.97	40.87	14.44	22.14	36.89
0.918	5.0	5365	0.92	41.17	14.94	22.54	37.40
0.901	6.0	6438	0.90	41.02	14.65	22.46	37.05
0.889	7.0	7511	0.89	41.32	15.09	22.64	37.42
0.900	8.0	8584	0 .90	41.23	15.02	22.67	37.28
0.869	9.0	9657	0.87	41.70	15.13	22.85	37.77

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu117
• Datasets 2.6.1
• Tokenizers 0.13.1