bart_billsum_abstractive_1024_1000

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

• Loss: 1.0789
• Rouge1: 0.6795
• Rouge2: 0.4076
• Rougel: 0.6139
• Rougelsum: 0.6139
• Wer: 0.4803
• Bleurt: -0.0583

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: 6
• eval_batch_size: 6
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Wer	Bleurt
No log	0.14	250	1.3122	0.6345	0.3515	0.5637	0.5638	0.5303	-0.3533
2.3005	0.27	500	1.2468	0.6452	0.3662	0.5767	0.5767	0.5174	-0.4992
2.3005	0.41	750	1.1909	0.6513	0.3745	0.5823	0.5823	0.5094	-0.4679
1.3108	0.55	1000	1.1685	0.6605	0.3827	0.5928	0.5928	0.5037	-0.1431
1.3108	0.68	1250	1.1505	0.6671	0.3894	0.5984	0.5984	0.4996	-0.0701
1.2615	0.82	1500	1.1334	0.6616	0.3883	0.5949	0.5949	0.4953	-0.3277
1.2615	0.96	1750	1.1226	0.6692	0.3948	0.6035	0.6035	0.492	-0.0701
1.1939	1.09	2000	1.1148	0.6669	0.3942	0.6007	0.6007	0.4892	-0.2128
1.1939	1.23	2250	1.1110	0.6741	0.4003	0.6072	0.6072	0.4884	-0.3492
1.1268	1.36	2500	1.1111	0.6746	0.4018	0.6093	0.6094	0.4865	-0.0701
1.1268	1.5	2750	1.0927	0.6717	0.4001	0.6054	0.6054	0.4837	-0.467
1.0977	1.64	3000	1.0840	0.6756	0.4048	0.6099	0.61	0.4814	-0.2661
1.0977	1.77	3250	1.0834	0.673	0.4034	0.6077	0.6077	0.4808	-0.2082
1.079	1.91	3500	1.0789	0.6795	0.4076	0.6139	0.6139	0.4803	-0.0583

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2