metadata

tags:
  - summarization
language:
  - it
metrics:
  - rouge
model-index:
  - name: summarization_mbart_fanpage4epoch
    results: []
datasets:
  - ARTeLab/fanpage

mbart-summarization-fanpage

This model is a fine-tuned version of facebook/mbart-large-cc25 on Fanpage dataset for Abstractive Summarization.

It achieves the following results:

Loss: 2.1833
Rouge1: 36.5027
Rouge2: 17.4428
Rougel: 26.1734
Rougelsum: 30.2636
Gen Len: 75.2413

Usage

from transformers import MBartTokenizer, MBartForConditionalGeneration
tokenizer = MBartTokenizer.from_pretrained("ARTeLab/mbart-summarization-fanpage")
model = MBartForConditionalGeneration.from_pretrained("ARTeLab/mbart-summarization-fanpage")

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-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: 4.0

Framework versions

Transformers 4.15.0.dev0
Pytorch 1.10.0+cu102
Datasets 1.15.1
Tokenizers 0.10.3