TGL-3

This model is a fine-tuned version of t5-small on an abstract-summary dataset, 23000 pieces of data for training. The data was acquired by openreview.net. It achieves the following results on the evaluation set:

• Loss: 2.4435
• Rouge1: 36.4998
• Rouge2: 17.8322
• Rougel: 31.8632
• Rougelsum: 31.8341

Model description

Here is the paper https://arxiv.org/abs/1910.10683

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: 5.6e-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: 8

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
2.9096	1.0	1240	2.5721	36.234	17.8214	31.5514	31.5549
2.7259	2.0	2480	2.5258	36.2572	17.9912	31.6249	31.6441
2.6434	3.0	3720	2.4957	36.4623	17.9657	31.7693	31.7542
2.5896	4.0	4960	2.4663	36.3692	17.8372	31.5909	31.6089
2.5491	5.0	6200	2.4511	36.4775	17.8094	31.8102	31.8003
2.5183	6.0	7440	2.4440	36.5892	17.906	31.9058	31.8985
2.4997	7.0	8680	2.4438	36.3747	17.8309	31.7314	31.7178
2.4863	8.0	9920	2.4435	36.4998	17.8322	31.8632	31.8341

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1