wesjos/Qwen3-4B-toolcall
wesjos/Qwen3-4B-toolcall is a 4 billion parameter language model, fine-tuned from unsloth/Qwen3-4B-unsloth-bnb-4bit, specifically optimized for tool-calling capabilities and mathematical reasoning. It demonstrates improved performance on GPQA, GSM8K, and ToolBench benchmarks compared to its base model. This model is designed for applications requiring enhanced function calling and problem-solving skills.
Loading preview...
Overview
wesjos/Qwen3-4B-toolcall is a 4 billion parameter language model, fine-tuned from the unsloth/Qwen3-4B-unsloth-bnb-4bit base model. It has been trained on a combination of datasets including interstellarninja/tool-calls-single-reasoning, mlabonne/FineTome-100k, and unsloth/OpenMathReasoning-mini, with a focus on enhancing its tool-calling and mathematical reasoning abilities.
Key Capabilities
- Improved Tool-Calling: Demonstrates significant gains in tool-calling accuracy, with its ToolBench Act.EM score increasing from 0.1667 to 0.3182.
- Enhanced Mathematical Reasoning: Shows better performance on mathematical tasks, with its GSM8K AverageAccuracy improving from 0.5 to 0.68.
- General Reasoning Boost: Achieves higher scores on general reasoning benchmarks like GPQA, with the overall AveragePass@1 increasing from 0.24 to 0.3333.
Good for
- Function Calling Applications: Ideal for scenarios where the model needs to accurately identify and call external tools or APIs based on user prompts.
- Mathematical Problem Solving: Suitable for tasks requiring numerical reasoning and solving mathematical problems.
- Complex Reasoning Tasks: Can be applied to use cases demanding improved logical and general reasoning capabilities.