bart_bertsum_1024_250_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: 1.0191
• Rouge1: 0.6894
• Rouge2: 0.4262
• Rougel: 0.6274
• Rougelsum: 0.6272
• Wer: 0.4606
• Bleurt: -0.5228

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	0.13	250	1.2589	0.6432	0.3644	0.5764	0.5763	0.5168	-0.3132
2.1861	0.27	500	1.1641	0.6562	0.3824	0.591	0.591	0.4985	-0.867
2.1861	0.4	750	1.1326	0.6626	0.3917	0.5988	0.5987	0.4904	-0.5078
1.2496	0.53	1000	1.1111	0.6657	0.3958	0.6015	0.6014	0.4859	-0.484
1.2496	0.66	1250	1.0959	0.6708	0.4014	0.6052	0.6051	0.4814	-0.4774
1.193	0.8	1500	1.0774	0.6724	0.4041	0.609	0.609	0.4787	-0.494
1.193	0.93	1750	1.0662	0.681	0.4127	0.6177	0.6176	0.4742	-0.4464
1.14	1.06	2000	1.0593	0.6795	0.4157	0.6178	0.6177	0.4709	-0.5849
1.14	1.2	2250	1.0504	0.6784	0.4158	0.6161	0.616	0.4685	-0.3624
1.0439	1.33	2500	1.0427	0.6815	0.418	0.6196	0.6195	0.4667	-0.5998
1.0439	1.46	2750	1.0357	0.6833	0.4198	0.6209	0.6207	0.465	-0.6198
1.045	1.6	3000	1.0286	0.6872	0.4238	0.6251	0.6251	0.4635	-0.4564
1.045	1.73	3250	1.0248	0.6829	0.4214	0.6222	0.6221	0.4622	-0.5228
1.0242	1.86	3500	1.0198	0.69	0.4273	0.6284	0.6283	0.4601	-0.4592
1.0242	1.99	3750	1.0191	0.6894	0.4262	0.6274	0.6272	0.4606	-0.5228

Framework versions

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