Gorilla: fine-tuned LLaMA that calls APIs without errors
In one sentence UC Berkeley presents Gorilla, a retrieval-augmented fine-tuned LLaMA for accurate API calls: reduces API hallucination from 83% to 3%, outperforming GPT-4 on this task.
One of the most frustrating problems with language models is their tendency to invent nonexistent APIs or call them with wrong parameters. If you're building an agent that needs to interact with real services, this error stops everything.
Gorilla is a fine-tuned LLaMA specifically trained to call APIs accurately. Instead of relying solely on training memory, it uses a retrieval mechanism: it searches for up-to-date API documentation and uses it as context before generating the call.
The result is striking: on a benchmark of over 1600 APIs (HuggingFace, TorchHub, TensorHub), Gorilla reduces errors from 83% to 3%, even outperforming GPT-4 on this specific task despite being much smaller.
Companies
UC Berkeley, Microsoft Research
Tools
Gorilla, LLaMA, APIBench
Tags
Sources