led-arxiv-10240

This model is a fine-tuned version of allenai/led-base-16384 on the arxiv-summarization dataset. It achieves the following results on the evaluation set:

  • Loss: 2.0623
  • Rouge1: 0.4423
  • Rouge2: 0.1739
  • Rougel: 0.2521
  • Rougelsum: 0.4004

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: 1e-05
  • train_batch_size: 8
  • 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

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum
2.2112 0.39 10000 2.1196 0.4367 0.1702 0.2468 0.3963
2.1591 0.79 20000 2.0623 0.4423 0.1739 0.2521 0.4004

Framework versions

  • Transformers 4.25.1
  • Pytorch 1.13.1+cu116
  • Datasets 2.8.0
  • Tokenizers 0.13.2
Downloads last month
107
Safetensors
Model size
162M params
Tensor type
F32
ยท
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.

Spaces using NielsV/led-arxiv-10240 2

Evaluation results