bart-samsung-test

This model is a fine-tuned version of facebook/bart-base on the samsum dataset. It achieves the following results on the evaluation set:

• Loss: 1.5511
• Rouge1: 46.7195
• Rouge2: 23.3711
• Rougel: 39.5121
• Rougelsum: 43.2091
• Gen Len: 17.7738

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
1.6838	1.0	1841	1.5511	46.7195	23.3711	39.5121	43.2091	17.7738

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu116
• Datasets 2.7.1
• Tokenizers 0.13.2