This GitHub repository hosts a comprehensive project that **fine-tunes the DistilGPT-2 **model for generating coherent and creative stories from prompts. The repository includes model selection code using Hugging Face's Transformers library, specifying the "distilgpt2" model, and configuring essential hyperparameters for training.
The training process is well-documented in the given report, with key settings like the number of training epochs, learning rate, batch sizes, and more. Additionally, data preprocessing steps, such as text cleaning and bracket removal, are incorporated, ensuring high-quality input data for the model.
This repository provides a complete solution for training and inferencing a story generation model.
**NOTE: Finetunned models can be found here: You can download and use: **
https://drive.google.com/drive/folders/109xaME-XbXxAACZmXkI-X4fKUhHYdJlu?usp=sharing
Two tunned models:
- "final_model" (trained with 30K examples for 10 epochs) (Best)
- "1_epoch_final_model" (train with 273K examples for 1 epoch only)