Chat GPT has taken over the world by storm, as over 100 million people with access to a computer, have stumbled upon it. I was one of those people. The model has had an enormous impact in my life. I’ve used it to proofread essays of mine to helping me study for spanish tests. Yet, there have still been places in which I’ve found the model struggles. One of those is API calls.
What is Gorilla ?
Gorilla is a fine-tuned large language model (LLM) created in collaboration from researchers from UC Berkely and Microsoft, to satisfy the niche of API Calls. This model outperforms Chat GPT in referencing and making API Calls. Researchers created this model with the goal of creating a “one-stop-shop for all APIs, (that) LLMs can interact with.” There were two steps to creating the model. The first was the collection of the models. The researcher created APIBench, a dataset consisting of models from TorchHub, TensorHub, and HuggingFace. With this dataset, they finetuned Gorilla, teaching it how to access the best API for the job, and how to create a satisfactory prompt for the API.
When compared the two other best suited model for the niche, GPT-4 and Claude, Gorilla created a much better result. GPT-4 referenced an imaginary model, while Claude chose one not suited for the prompt. Gorilla successfully, chose a great API, and wrote a successful prompt for it.
Try Gorilla Here: https://lnkd.in/gMA5tcCG