summarise_v8 / README.md
debbiesoon's picture
Update README.md
432e4a0
metadata
license: apache-2.0
tags:
  - generated_from_trainer
model-index:
  - name: summarise_v8
    results: []

SGH logo.png

This model is a fine-tuned version of allenai/led-base-16384 on the SGH news articles and summaries dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8163
  • Rouge2 Precision: 0.3628
  • Rouge2 Recall: 0.3589
  • Rouge2 Fmeasure: 0.3316

Model description

This model was created to generate summaries of news articles.

Intended uses & limitations

The model takes up to maximum article length of 768 tokens and generates a summary of maximum length of 512 tokens, and minimum length of 100 tokens.

Training and evaluation data

This model was trained on 100+ articles and summaries from SGH.

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 5e-05
  • train_batch_size: 2
  • eval_batch_size: 2
  • seed: 42
  • 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 Rouge2 Precision Rouge2 Recall Rouge2 Fmeasure
1.5952 0.23 10 1.0414 0.2823 0.3908 0.3013
1.8116 0.47 20 0.9171 0.3728 0.273 0.3056
1.6289 0.7 30 0.8553 0.3284 0.2892 0.291
1.5074 0.93 40 0.8163 0.3628 0.3589 0.3316

Framework versions

  • Transformers 4.21.3
  • Pytorch 1.12.1+cu113
  • Datasets 1.2.1
  • Tokenizers 0.12.1