Edit model card

test

This model is a fine-tuned version of Ameer05/tokenizer-repo on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 1.6109
  • Rouge1: 54.9442
  • Rouge2: 45.3299
  • Rougel: 50.5219
  • Rougelsum: 53.6475

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: 4
  • total_train_batch_size: 32
  • 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
No log 0.91 5 2.3705 53.62 44.3835 49.6135 52.693
No log 1.91 10 1.9035 47.478 37.0934 39.7935 45.1881
No log 2.91 15 1.7990 54.2488 45.0782 49.8421 52.7564
No log 3.91 20 1.7125 55.7903 46.7554 52.2733 54.9389
2.4456 4.91 25 1.6421 52.2279 43.4391 49.6955 51.2915
2.4456 5.91 30 1.6102 55.8598 47.3293 53.1337 54.8596
2.4456 6.91 35 1.6164 53.7902 44.6622 49.5045 52.2304
2.4456 7.91 40 1.6015 51.5597 42.0333 47.9639 50.1154
1.239 8.91 45 1.6067 53.0301 43.7214 49.0227 51.8109
1.239 9.91 50 1.6109 54.9442 45.3299 50.5219 53.6475

Framework versions

  • Transformers 4.15.0
  • Pytorch 1.9.1
  • Datasets 2.0.0
  • Tokenizers 0.10.3
Downloads last month
11