longt5_xl_gov_5

This model is a fine-tuned version of google/long-t5-tglobal-xl on the tau/scrolls gov_report dataset. It achieves the following results on the evaluation set:

• Loss: 1.4141
• Rouge1: 54.2522
• Rouge2: 24.7528
• Rougel: 27.2444
• Rougelsum: 51.5916
• Gen Len: 889.25

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: 0.001
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 128
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• num_epochs: 5.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
1.6209	1.0	136	1.5434	45.0062	18.1618	23.3808	41.7616	904.1996
1.418	1.99	272	1.4141	54.2522	24.7528	27.2444	51.5916	889.25
1.2626	3.0	409	1.4249	52.4332	23.708	27.2902	49.8071	878.4095
1.0992	4.0	545	1.4211	55.2041	26.5229	29.9951	52.6487	670.7047
0.9974	4.99	680	1.4569	55.9961	26.2205	29.0409	53.3109	883.0463

Framework versions

• Transformers 4.34.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.14.5
• Tokenizers 0.13.3