ShishirPatil/gorilla

Features

• LLM Tool Calling - Provides a foundational framework for training and evaluating models to translate natural language into executable API calls.
• Natural Language Tool Interfaces - Translates natural language queries into precise API invocations to allow language models to control external tools.
• Tool Call Accuracy Validators - Measures the accuracy, latency, and cost of language models when triggering external tools.
• Fine-Tuning Pipelines - Provides workflows for adapting large language models to improve tool-use and retrieval-augmented generation performance.
• Function Calling Fine-Tuning - Provides training procedures to adapt models for accurate tool-use and function-calling patterns using curated datasets.
• Model Fine-Tuning and Adaptation - Provides processes for fine-tuning language models to improve accuracy and relevance in retrieval-augmented generation tasks.
• API Generation Benchmarks - Evaluates the ability of models to generate correct function calls within real-world multi-turn agentic workflows.
• Model Action Runtimes - Implements a specialized runtime that ensures safety guarantees when executing actions produced by a language model.
• Model Training API Indexes - Provides a structured repository of API definitions used to train models in translating natural language into executable code.
• Training-Ready API Indexes - Organizes and maintains searchable API definitions to improve the training and inference of function calling models.
• Sandboxed Execution Environments - Implements a secure, isolated runtime for executing model-generated API calls and CLI commands with validation and undo capabilities.
• Sandboxed Execution Environments - Provides isolated runtime environments to run generated code and API calls securely with state recovery.
• Function Calling Benchmarks - Assesses model performance on single-turn and multi-step calling tasks using curated datasets and latency metrics.
• LLM Evaluation - Provides a suite of metrics and datasets to measure the accuracy, cost, and latency of model-driven function calling.
• Function Call Benchmarking - Measures model accuracy and latency across sequences of interdependent function calls and agentic workflows.
• Tool Calling Evaluation Frameworks - Measures the accuracy, cost, and latency of language models when triggering external tools across multiple languages.
• Human Approval Gates - Implements security gates that require human approval before executing model-generated infrastructure changes.
• Language Model Fine-Tuning - Adapts language models for retrieval-augmented generation to improve domain-specific question answering accuracy.
• Retrieval-Augmented Training - Implements training methods that enhance question answering accuracy using retrieval-augmented techniques and chain-of-thought reasoning.
• Parallel Function Calling - Allows models to trigger multiple independent API invocations within a single response for complex task execution.
• API Documentation Indices - Maintains a searchable index of API arguments and versions to provide accurate data for model training and inference.
• Natural Language Automation - Translates natural language instructions into executable command line actions for infrastructure management.
• Execution Sandboxes - Provides a secure runtime environment that restricts code execution to prevent system damage during action execution.
• API Definition Indices - Organizes comprehensive collections of API definitions to facilitate more accurate model training and evaluation.
• Agent Performance Benchmarks - Provides a standardized ranking system and prompt hub to compare the performance of AI agents.
• Action Result Validators - Ships a secure runtime that validates the results of executed code and API calls and reverts unintended changes.
• Application Services - API store for managing LLM interactions.

Star history