Hugging Face Overview of Arch-Agent-3B
Arch-Agent-3B, developed by katanemo, is a 3.1 billion parameter language model designed for advanced function calling and agentic applications. It is built to handle complex, multi-step tasks requiring intelligent tool selection, adaptive planning, and seamless integration with external APIs. The model maintains contextual continuity across multiple dialogue turns and can plan and execute sequences of function calls, adapting dynamically based on intermediate results.
Key Capabilities
- Multi-Turn Function Calling: Supports ongoing conversations with nested or evolving tool use.
- Multi-Step Function Calling: Decomposes goals into sub-tasks and executes a sequence of function calls.
- Agentic Capabilities: Provides advanced decision-making and workflow management for complex tasks, including tool coordination and error recovery.
Performance and Usage
Arch-Agent-3B's performance is evaluated on the Berkeley Function-Calling Leaderboard (BFCL), demonstrating strong results in comparison to other commonly used models. For evaluation, Arch-Agent models utilize YaRN scaling for Multi-Turn scenarios and are assessed with a context length of 64K. The model is integrated with the Hugging Face transformers library (version 4.51.0 or newer) and uses a specific prompt format to extract JSON output similar to OpenAI's function calling. Detailed examples for quickstart and integration are provided, showcasing how to define tools and format prompts for effective function calling.