Overview
driaforall/Tiny-Agent-a-1.5B is a 1.5 billion parameter model from the Dria-Agent-a series, fine-tuned on Qwen2.5-Coder for efficient deployment on edge devices. A key innovation is its Pythonic function calling mechanism, which allows the model to interact with tools using Python code blocks. This approach facilitates complex, multi-step reasoning and actions within a single turn, moving beyond the limitations of traditional JSON-based function calling.
Key Capabilities
- One-shot Parallel Multiple Function Calls: Executes multiple synchronous processes in a single chat turn.
- Free-form Reasoning: Provides natural language reasoning traces alongside actions embedded in Python code.
- On-the-fly Complex Solution Generation: Generates custom Python programs with conditionals and sequential logic for intricate tasks.
- Quantization Aware Training: Designed to minimize performance degradation when quantized for edge device deployment.
Performance
Evaluated on the Dria-Pythonic-Agent-Benchmark (DPAB), Tiny-Agent-a-1.5B achieves a 73% score for Pythonic function calling, outperforming many larger models, including Qwen2.5-Coder-3B-Instruct and Qwen2.5-Coder-7B-Instruct, in this specific capability.
Good For
- Applications requiring advanced tool use and complex, multi-step automation.
- Edge device deployments where model size and efficient function calling are critical.
- Developers seeking a flexible, code-centric approach to agentic AI tasks.