Hugging Face Overview
ToolACE-8B: Advanced Function Calling Model
ToolACE-8B, developed by Team-ACE, is an 8 billion parameter model fine-tuned from LLaMA-3.1-8B-Instruct. Its core strength lies in tool usage and function calling, demonstrating state-of-the-art performance on the Berkeley Function-Calling Leaderboard (BFCL), where it competes with advanced GPT-4 models.
Key Capabilities
- Superior Function Calling: Excels at interpreting natural language queries and generating precise function calls, as evidenced by its top rankings on BFCL-v3.
- ToolACE Dataset: Trained using the proprietary ToolACE dataset, an automatically generated, agentic pipeline designed for Accurate, Complex, and divErse tool-learning data.
- Comprehensive API Pool: Leverages a self-evolution synthesis process to curate a vast API pool of 26,507 diverse APIs.
- Robust Data Generation: Utilizes multi-agent interplay and a formalized thinking process to generate high-quality dialogs for tool learning.
- Dual-Layer Verification: Ensures data accuracy through a combination of rule-based and model-based checks.
Ideal Use Cases
- Automated Agent Development: Building intelligent agents that require precise and complex tool interaction.
- Function Calling Applications: Implementing robust function calling capabilities in various applications, from data retrieval to task automation.
- Complex Query Interpretation: Handling user queries that necessitate multiple, chained, or conditional tool invocations.
More details on the underlying methodology can be found in the ICLR2025 paper: "ToolACE: Winning the Points of LLM Function Calling".