t5-base-finetuned-summscreen-bestval-100-genlen-10-epochs

This model is a fine-tuned version of t5-base on the SummScreen dataset. It achieves the following results on the evaluation set:

• Loss: 2.9499
• Rouge1: 29.2769
• Rouge2: 5.5288
• Rougel: 17.5141
• Rougelsum: 25.345
• Gen Len: 86.7596

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
3.2858	0.99	3500	3.1088	27.292	4.7003	16.4176	23.7005	84.6259
3.2085	1.99	7000	3.0425	28.3997	5.0233	17.0582	24.5815	86.8322
3.107	2.98	10500	3.0110	28.7042	5.3326	17.429	24.7691	84.8549
3.074	3.98	14000	2.9886	28.8975	5.371	17.302	25.0658	86.6327
2.9899	4.97	17500	2.9769	29.0185	5.6415	17.6407	24.7669	82.8435
2.9857	5.97	21000	2.9647	29.5476	5.5332	17.4855	25.2605	87.3152
2.9542	6.96	24500	2.9586	29.4713	5.5729	17.5815	25.2393	88.0295
2.9301	7.96	28000	2.9536	29.8483	5.7355	17.8895	25.774	87.195
2.9118	8.95	31500	2.9503	29.3014	5.5802	17.5983	25.3476	86.0476
2.9033	9.95	35000	2.9499	29.2769	5.5288	17.5141	25.345	86.7596

Framework versions

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