prompt-compressor

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

• Loss: 0.8566
• Rouge1: 0.4106
• Rouge2: 0.3099
• Rougel: 0.3802
• Rougelsum: 0.3806
• Length Ratio To Reference: 3.2022

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: 3e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Length Ratio To Reference
0.7994	0.992	62	0.9094	0.4043	0.2832	0.3667	0.3665	3.239
0.4782	2.0	125	0.9544	0.3673	0.2642	0.3358	0.336	3.5952
0.2149	2.992	187	1.0604	0.4005	0.2792	0.3608	0.3605	3.0623
0.1182	4.0	250	1.1147	0.3914	0.2734	0.3539	0.3541	3.2979

Framework versions

• Transformers 4.44.0
• Pytorch 2.8.0+cu128
• Datasets 2.21.0
• Tokenizers 0.19.1