bart-summarizer

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

• Loss: 2.1898
• Rouge1: 51.7683
• Rouge2: 36.3956
• Rougel: 45.7626
• Rougelsum: 45.7512
• Bert F1: 89.7697

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Bert F1
2.48	1.0	766	2.3197	46.084	31.1672	40.7261	40.733	88.58
2.203	2.0	1532	2.2230	49.9815	34.8577	44.2515	44.2457	89.3509
2.1447	3.0	2298	2.1980	50.7333	35.3908	44.6146	44.6091	89.4589
2.0614	4.0	3064	2.1907	51.6468	36.4567	45.7548	45.7343	89.7909
2.0515	4.9941	3825	2.1898	51.7683	36.3956	45.7626	45.7512	89.7697

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0