The concept of Artificial General Intelligence (AGI) has been a topic of discussion for years, yet we’re still far from achieving a truly comprehensive AGI system. However, a recent project called Gorilla might just be the stepping stone we’ve been waiting for.
Gorilla: Bridging the Gap Between Language Models and APIs
Gorilla is an innovative API store designed for large language models. It allows these models to call APIs using natural language queries, ensuring both semantic and syntactic accuracy. By integrating over 1,600 APIs, Gorilla showcases a groundbreaking method of API interaction using language models, significantly reducing the occurrence of AI hallucinations.
One of Gorilla’s key features is APIBench, an extensive, curated collection of APIs that’s easy to use for training purposes. This tool addresses several critical challenges in the AI field:
- Precision in API Calls: Gorilla enhances the accuracy of external API calls when using large language models.
- Natural Language Interface: Users can query and call APIs using natural language, boosting development efficiency and accuracy.
- Reduced Errors: The APIBench feature optimizes the API calling process, minimizing mistakes and erroneous outputs.
For those interested in experiencing Gorilla firsthand, a demo is available at https://huggingface.co/spaces/gorilla-llm/gorilla-demo.
Fine-tuning and Customization
Gorilla offers two methods for fine-tuning large language models with custom API sets:
- Using “text-generation-webui” for full precision model fine-tuning (note: this doesn’t support .k-quantized models in .ggu file format).
- Employing “Llama.cpp” for fine-tuning llama-based k-quantized models (currently not supporting falcon or mpt-based models).
These options allow users to tailor the system to their specific needs, following step-by-step guidelines for configuration and implementation.
Project Structure and Resources
The Gorilla project is well-organized, with key components including:
- Evaluation scripts for model function calls
- Comprehensive API data, including community contributions
- Evaluation code and output results
- Inference code for local implementation
The project provides complete model weights, supports various API calls, and offers detailed instructions for both local execution and API hosting.
For those looking to dive deeper into the project, all resources are available at https://github.com/ShishirPatil/gorilla.
While AGI remains a distant goal, projects like Gorilla are paving the way for more sophisticated AI systems. By bridging the gap between natural language processing and API functionality, Gorilla represents a significant step forward in our journey towards true artificial general intelligence.
