bert2gpt2_med_v3

This model is a fine-tuned version of Chemsseddine/bert2gpt2_med_v2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.5474
• Rouge1: 31.8871
• Rouge2: 14.4411
• Rougel: 31.6716
• Rougelsum: 31.579
• Gen Len: 22.8412

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: 2e-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
2.5621	1.0	900	1.9724	30.3731	13.8412	29.9606	29.9716	22.6353
1.3692	2.0	1800	1.9634	29.6409	13.7674	29.5202	29.5207	22.5059
0.8308	3.0	2700	2.1431	30.9317	14.5594	30.8021	30.7287	22.6118
0.4689	4.0	3600	2.2970	30.1132	14.6407	29.9657	30.0182	23.3235
0.2875	5.0	4500	2.3787	30.9378	14.7108	30.861	30.9097	22.7529
0.1564	6.0	5400	2.4137	30.5338	13.9702	30.1252	30.1975	23.1588
0.1007	7.0	6300	2.4822	30.872	14.9353	30.835	30.7694	23.0529
0.0783	8.0	7200	2.4974	29.9825	14.1702	29.7507	29.7271	23.1882
0.0504	9.0	8100	2.5175	31.96	15.0705	31.9669	31.9839	23.0588
0.0339	10.0	9000	2.5474	31.8871	14.4411	31.6716	31.579	22.8412

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1