base_model: google/pegasus-cnn_dailymail
tags:
- generated_from_trainer
datasets:
- xlsum
model-index:
- name: pegsasus_xlsum
results: []
language:
- en
metrics:
- rouge
my_awesome_pegsasus_model
This model is a fine-tuned version of google/pegasus-cnn_dailymail on the xlsum dataset.
Model description
pegasus_xlsum is a state-of-the-art model fine-tuned on the English subset of the csebuetnlp/xlsum dataset. This data source is one of the most comprehensive and diverse sets available, originally composed of 1.35 million professional article-summary pairs sourced from BBC across 45 languages. Despite its multilingual nature, we intentionally selected the English language subset, consisting of approximately 330,000 records, as the focus for our fine-tuning process.
The goal was to adapt the model for the text summarization task, and we're thrilled to report that the fine-tuned pegasus_xlsum model exceeded our expectations. It outperformed the established csebuetnlp/mT5_multilingual_XLSum model in terms of ROUGE scores, demonstrating superior summary generation capabilities. The pegasus_xlsum model leverages the powerful PEGASUS architecture, proving its efficiency and effectiveness in handling English text summarization tasks.
Intended uses & limitations
pegasus_xlsum is to provide a reliable, high-performance solution for English text summarization, making the most of the rich, professional, and diverse source dataset it was trained on. We hope you find this model as useful in your applications as we did in our experiments.
Training and evaluation data
| Metric | Score |
|---|---|
| ROUGE-1 F1 | 0.391215 |
| ROUGE-2 F1 | 0.174674 |
| ROUGE-L F1 | 0.308941 |
| ROUGE-LSUM F1 | 0.308923 |
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-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: 4
Framework versions
- Transformers 4.31.0
- Pytorch 2.0.1+cu118
- Datasets 2.14.0
- Tokenizers 0.13.3