led-base-16384-finetuned-summscreen-bestval-100-genlen-10-epochs

This model is a fine-tuned version of allenai/led-base-16384 on the SummScreen dataset. It achieves the following results on the evaluation set:

• Loss: 3.1833
• Rouge1: 31.6225
• Rouge2: 6.7688
• Rougel: 18.5526
• Rougelsum: 27.3033
• Gen Len: 81.8209

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-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
3.1147	0.99	3500	3.0409	30.2997	6.0055	17.6836	26.3062	92.0363
2.7805	1.99	7000	2.9981	31.0651	6.1553	18.2326	26.6136	84.1678
2.6362	2.98	10500	3.0013	31.0598	6.4268	18.3805	26.7816	81.1338
2.4646	3.98	14000	3.0267	31.46	6.5898	18.5842	27.165	83.4966
2.2546	4.97	17500	3.0448	31.4952	6.3709	18.2124	27.111	86.1701
2.1432	5.97	21000	3.0840	31.6475	6.663	18.504	27.2444	81.3583
2.0585	6.96	24500	3.1105	31.3432	6.7862	18.4461	27.0081	76.4671
1.9243	7.96	28000	3.1558	31.5712	6.6456	18.3928	27.2066	81.1769
1.8828	8.95	31500	3.1686	31.6361	6.4783	18.4167	27.2049	83.0658
1.8403	9.95	35000	3.1833	31.6225	6.7688	18.5526	27.3033	81.8209

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1
• Datasets 2.9.0
• Tokenizers 0.13.2