Gorilla: LLMs Connected to APIs Explained
LLMs have been supercharged by connecting them with external tools. An external tool could be a search engine, code executor, calculator, calendar, email, CRM, and many others! Although GPT-4 is fairly strong at formatting API requests zero-shot (without additional training), Gorilla shows that specialized training can outperform it significantly! In addition to the accuracy performance, this is also achievable with a much cheaper 7 billion parameter model, derived by fine-tuning the Meta AI LlaMA-2 7B checkpoint!! There are all sorts of interesting details about this paper covered in the video, from the APIBench dataset to Self-Instruct training data generation, Retrieval-Aware Training, and the miscellaneous details of Gorilla! I hope you enjoy the paper summary video! As always I am more than happy to answer any questions or discuss any ideas you have related to the content in the video! P.S. Please stay tuned for Weaviate Gorilla! https://www.youtube.com/watch?v=LkV5DTRNxAg