Edit model card

reviews-generator

This model is a fine-tuned version of facebook/bart-base on the amazon_reviews_multi dataset. It achieves the following results on the evaluation set:

  • Loss: 3.4990

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: 32
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • training_steps: 1000

Training results

Training Loss Epoch Step Validation Loss
3.7955 0.08 500 3.5577
3.7495 0.16 1000 3.4990

Framework versions

  • Transformers 4.18.0
  • Pytorch 1.11.0+cu113
  • Datasets 2.1.0
  • Tokenizers 0.12.1
Downloads last month
6
Hosted inference API
This model can be loaded on the Inference API on-demand.

Dataset used to train Caroline-Vandyck/reviews-generator

Evaluation results