praveenseb/product_review_generator

This model is a fine-tuned version of distilgpt2 on a sample of amazon_us_reviews dataset. The sample was drawn from 'Apparel_v1_00' subset.

Model description

This model can auto generate review text for apparel products on providing product title, review rating (1-5) and review headline as an input prompt.

The input prompt should be in the format <|BOS|>product_title<|SEP|>product_rating<|SEP|>review_title<|SEP|>. For example, <|BOS|>Columbia Women's Benton Springs Full-Zip Fleece Jacket<|SEP|>5<|SEP|>Awesome jacket!<|SEP|>. You can find the complete code in my GitHub repository.

Intended uses & limitations

This model is only intended to demonstrate the text generation capabilities of transformer-based models. Do not use it commercially or for any real-life purpose . The model is trained specifically on 'Apparel_v1_00' dataset. So, using non-apparel product titles in the input prompt may yield inconsistent results.

Training procedure

Code used for training can found in my GitHub repository.

Training hyperparameters

The following hyperparameters were used during training:

• optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'ExponentialDecay', 'config': {'initial_learning_rate': 0.0002, 'decay_steps': 1000, 'decay_rate': 0.95, 'staircase': True, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-07, 'amsgrad': False, 'weight_decay_rate': 0.0}
• training_precision: float32

Training results

Train Loss	Epoch
0.7579	0
0.6720	1

Framework versions

• Transformers 4.27.3
• TensorFlow 2.11.0
• Datasets 2.10.1
• Tokenizers 0.13.2