wesjos/Qwen3-4B-toolcall
The wesjos/Qwen3-4B-toolcall model is a 4 billion parameter language model, fine-tuned from unsloth/Qwen3-4B-unsloth-bnb-4bit. It specializes in tool-calling, mathematical reasoning, and general problem-solving, demonstrating improved performance on GPQA, GSM8K, and ToolBench benchmarks. With a 32768 token context length, this model is optimized for tasks requiring external tool utilization and complex reasoning.
Loading preview...
Model Overview
wesjos/Qwen3-4B-toolcall is a 4 billion parameter language model, building upon the unsloth/Qwen3-4B-unsloth-bnb-4bit base model. It has been specifically fine-tuned using a combination of datasets including interstellarninja/tool-calls-single-reasoning, mlabonne/FineTome-100k, and unsloth/OpenMathReasoning-mini to enhance its capabilities in tool-calling and mathematical reasoning.
Key Capabilities & Performance
This fine-tuned model shows notable improvements across several benchmarks:
- GPQA (General Problem Answering): The overall AveragePass@1 score increased from 0.24 to 0.3333, indicating better general reasoning.
- GSM8K (Grade School Math 8K): AverageAccuracy improved significantly from 0.50 to 0.68, demonstrating enhanced mathematical problem-solving.
- ToolBench: The Act.EM (Action Exact Match) score for tool-calling tasks rose from 0.1667 to 0.3182, highlighting its improved ability to correctly utilize external tools.
These metrics indicate that the model is particularly well-suited for applications requiring robust reasoning and effective tool interaction.
Use Cases
- Tool-calling applications: Ideal for scenarios where the LLM needs to interact with external APIs or functions.
- Mathematical problem-solving: Suitable for tasks requiring accurate numerical and logical reasoning.
- Complex question answering: Benefits from its improved general reasoning capabilities.