t5-base-pt-asqa-ob

This model is a fine-tuned version of din0s/t5-base-msmarco-nlgen-ob on the ASQA dataset. It achieves the following results on the evaluation set:

• Loss: 1.7481
• Rougelsum: 12.3722

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rougelsum
No log	1.0	355	1.8760	11.5138
2.1344	2.0	710	1.8322	11.6843
1.979	3.0	1065	1.8109	11.8592
1.979	4.0	1420	1.7967	11.9466
1.9493	5.0	1775	1.7871	12.0333
1.9099	6.0	2130	1.7778	12.0805
1.9099	7.0	2485	1.7720	12.1659
1.8748	8.0	2840	1.7668	12.2039
1.8584	9.0	3195	1.7628	12.2506
1.8362	10.0	3550	1.7601	12.2557
1.8362	11.0	3905	1.7575	12.2718
1.8134	12.0	4260	1.7562	12.2789
1.7996	13.0	4615	1.7538	12.3179
1.7996	14.0	4970	1.7529	12.3035
1.8049	15.0	5325	1.7519	12.3317
1.7898	16.0	5680	1.7510	12.3717
1.7872	17.0	6035	1.7497	12.3750
1.7872	18.0	6390	1.7486	12.3580
1.7759	19.0	6745	1.7483	12.3698
1.785	20.0	7100	1.7481	12.3722

Framework versions

• Transformers 4.23.0.dev0
• Pytorch 1.12.1+cu102
• Datasets 2.4.0
• Tokenizers 0.12.1