Edit model card

bart-cnn-pubmed-arxiv-pubmed-v3-e10

This model is a fine-tuned version of theojolliffe/bart-cnn-pubmed-arxiv-pubmed on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8410
  • Rouge1: 56.5123
  • Rouge2: 41.1641
  • Rougel: 43.4495
  • Rougelsum: 54.544
  • Gen Len: 141.6667

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: 2e-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: 10
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum Gen Len
1.254 1.0 795 0.9244 52.4478 32.5958 34.8756 49.8059 142.0
0.6985 2.0 1590 0.8156 52.4786 33.2296 35.5063 49.737 141.7963
0.5252 3.0 2385 0.7821 52.0494 32.953 36.5502 49.7292 142.0
0.3389 4.0 3180 0.7422 53.5408 36.2206 39.8389 51.6693 142.0
0.26 5.0 3975 0.7670 54.4279 36.5972 40.255 52.0877 142.0
0.1678 6.0 4770 0.8106 54.6811 37.8329 40.8512 52.3482 141.963
0.1243 7.0 5565 0.7926 54.5081 37.9596 41.912 52.5097 142.0
0.0967 8.0 6360 0.8079 56.0795 40.0954 43.7055 54.2041 142.0
0.0709 9.0 7155 0.8390 55.5257 38.5546 42.1562 53.5524 141.963
0.0691 10.0 7950 0.8410 56.5123 41.1641 43.4495 54.544 141.6667

Framework versions

  • Transformers 4.18.0
  • Pytorch 1.11.0+cu113
  • Datasets 2.1.0
  • Tokenizers 0.12.1
Downloads last month
1
Hosted inference API
This model can be loaded on the Inference API on-demand.

Evaluation results