arabert2arabert-finetuned-ar-wikilingua

This model is a fine-tuned version of on the wiki_lingua dataset. It achieves the following results on the evaluation set:

  • Loss: 4.6877
  • Rouge-1: 13.2
  • Rouge-2: 3.43
  • Rouge-l: 12.45
  • Gen Len: 20.0
  • Bertscore: 64.88

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: 5e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 16
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 250
  • num_epochs: 8
  • label_smoothing_factor: 0.1

Training results

Training Loss Epoch Step Validation Loss Rouge-1 Rouge-2 Rouge-l Gen Len Bertscore
6.7667 1.0 156 5.3846 3.36 0.56 3.27 20.0 60.6
5.257 2.0 312 5.0424 5.44 0.88 5.35 20.0 60.56
4.743 3.0 468 4.8294 9.21 1.8 8.93 20.0 62.91
4.3832 4.0 624 4.7240 9.88 2.19 9.6 20.0 62.65
4.1166 5.0 780 4.6861 11.61 2.86 11.13 20.0 63.71
3.91 6.0 936 4.6692 12.27 3.11 11.76 20.0 64.07
3.7569 7.0 1092 4.6805 12.93 3.38 12.28 20.0 64.61
3.6454 8.0 1248 4.6877 13.2 3.43 12.45 20.0 64.88

Framework versions

  • Transformers 4.19.4
  • Pytorch 1.11.0+cu113
  • Datasets 2.2.2
  • Tokenizers 0.12.1
Downloads last month
7
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train ahmeddbahaa/arabert2arabert-finetuned-ar-wikilingua