medium / README.md
Peter's picture
update model card README.md
d12c704
|
raw
history blame
1.72 kB
metadata
tags:
  - generated_from_trainer
metrics:
  - rouge
model-index:
  - name: medium
    results: []

medium

This model is a fine-tuned version of prithivida/parrot_paraphraser_on_T5 on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6025
  • Rouge1: 81.6007
  • Rouge2: 75.1196
  • Rougel: 81.4213
  • Rougelsum: 81.4956
  • Gen Len: 32.4286

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
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 3.0

Training results

Training Loss Epoch Step Validation Loss Rouge1 Rouge2 Rougel Rougelsum Gen Len
No log 1.0 63 0.5775 65.0748 58.8985 64.5731 63.6249 19.0
No log 2.0 126 0.5806 74.3055 69.2025 73.4922 73.0941 17.8571
No log 3.0 189 0.6025 71.3808 66.0359 70.1235 69.4614 18.0

Framework versions

  • Transformers 4.15.0
  • Pytorch 1.10.1+cu113
  • Datasets 1.17.0
  • Tokenizers 0.10.3