The article also discusses the changes in Version 3 of the package, including new features and breaking changes. It provides information on how to install and use the package, including examples of code. The article also provides information on where to get papers for use with the package, including Zotero and Paper Scraper. It also discusses options for PDF reading, typewriter view, and caching. The article ends with a FAQ section addressing differences between Paper QA and other similar tools, the use of different LLMs, the source of documents, and saving or loading the Docs class.
Key takeaways:
- Paper QA is a minimal package for doing question and answering from PDFs or text files. It uses OpenAI Embeddings with a vector DB called FAISS to embed and search documents.
- The package uses a process of embedding documents and queries into vectors, searching for top k passages in documents, creating summaries of each passage relevant to the query, and generating an answer with the prompt.
- Version 3 of Paper QA includes new features such as memory in query, support for adding from URLs and file objects, customizable prompts, and consistent use of dockey and docname for better tracking with external databases.
- The Docs class in Paper QA can be pickled and unpickled, allowing users to save the embeddings of the documents and load them later.