Create a text summarization model by fine-tuning a pre-trained model with a custom dataset

This project aims to create a high-quality custom dataset using PaLM API and create a high-quality text summarization model by fine-tuning a pre-trained model with the custom dataset. The custom dataset will be a CSV file with 10K rows and 4 columns: id, prompt, document, and summary. The document should vary in different writing styles such as news articles, blog posts, legal papers and so on, and the length of a summary will be between 1/3 and 1/4 of the length of the original document. This will make the custom dataset to be far superior to most publicly available datasets like CNN-Daily-Mail or XSUM, which typically generate 1-2 sentence summaries. Then I will create a few text summarization models with different pre-trained models, improving each fine-tuned model’s accuracy by implementing strategy driven from bias/variance and error analysis. After creating a few fine-tuned models, I will pick the best performing one based on the evaluation metric. Based on this, I will work on creating a tutorial covering the whole process, such as creating a custom dataset using PaLM API, fine-tuning pre-trained models using Keras API, and selecting the best performing one based on a evaluation metric. As a result, by creating this comprehensive tutorial, the project will encourage other developers to explore/use Keras API and PaLM API in their projects.