iromu/Qwen3-0.6B-tools
The iromu/Qwen3-0.6B-tools model is a 0.8 billion parameter Qwen3-based language model developed by iromu, fine-tuned specifically for reliable OpenAI-style tool calling. It excels at structured tool/function calling and agent-style multi-step interactions, making it suitable for small-footprint on-device or edge deployments. This model was fine-tuned from Qwen/Qwen3-0.6B using supervised fine-tuning with LoRA on a specialized tool-calling dataset.
Loading preview...
Overview
iromu/Qwen3-0.6B-tools is a specialized 0.8 billion parameter model based on the Qwen3 architecture, developed by iromu. It has been fine-tuned to reliably emit OpenAI-style tool calls, making it highly effective for agentic workflows. The model was trained using Supervised Fine-Tuning (SFT) with LoRA (dim 32, alpha 32, dropout 0.05) on the sft_tools split of the r0b0tlab/qwen3.8-max-glm5.2-kimi-k3-distillation dataset, which features assistant turns with tool_calls.
Key Capabilities
- Structured Tool/Function Calling: Designed to accurately generate tool calls in a structured format.
- Agent-Style Interactions: Supports multi-step interactions typical of AI agents.
- Small Footprint: Optimized for efficient deployment in resource-constrained environments.
Performance Highlights
Validation on a held-out split demonstrates significant improvement in tool-calling capabilities compared to its base model. The Qwen3-0.6B-tools model achieved a 99.3% rate of tool call emission, with 83.2% of names matching and 63.9% exact argument matches, representing a +63.1 percentage point increase in exact argument matches over the base Qwen/Qwen3-0.6B model.
Good For
- Applications requiring reliable function calling.
- Building AI agents that interact with external tools.
- On-device or edge deployments where model size and efficiency are critical.