seanpoyner/smolcode-coder-go-3b-tools
seanpoyner/smolcode-coder-go-3b-tools is a 3.1 billion parameter LoRA fine-tune of Qwen2.5-Coder-1.5B-Instruct, developed by seanpoyner. This model is specifically trained to emit native function calls, enabling agentic coding loops with small language models. It excels at driving tool-use agents by providing structured tool outputs, making it suitable for applications requiring precise function calling from compact models.
Loading preview...
Model Overview
seanpoyner/smolcode-coder-go-3b-tools is a 3.1 billion parameter LoRA fine-tune of the Qwen2.5-Coder-1.5B-Instruct base model, developed by seanpoyner. Its primary purpose is to enable small language models (SLMs) to effectively drive agentic coding loops by generating native <tool_call> function calls, rather than plain-text JSON descriptions. This model was specifically built for smolcode, an SLM-optimized agentic coding assistant, as part of the Hugging Face "Build Small" hackathon.
Key Capabilities
- Native Tool Call Generation: Unlike many small coder models that output tool calls as plain-text JSON, this model is fine-tuned to emit the native
<tool_call>{"name": ..., "arguments": ...}</tool_call>format, which is directly parseable by runtimes like Ollama and llama.cpp. - Agentic Loop Integration: This capability is crucial for closing the loop in agentic systems, allowing the model to directly interact with and drive external tools.
- Efficient for Small Models: It addresses a critical gap for tiny (≤2B, Tiny-Titan-class) models, enabling them to perform complex tool-use tasks that typically require larger models.
Training Details
The model was fine-tuned using bf16 LoRA on attention and MLP projections, with assistant-only loss focused solely on tool calls and final answers. Training data included NousResearch/hermes-function-calling-v1 for breadth and synthetic smolcode tool-use trajectories for sharpness on specific tools. The training process ensured byte-identical targets by rendering all data through the same apply_chat_template(tools=...) used during inference.
Use Cases
This model is ideal for developers building agentic coding assistants or other applications where a small, efficient language model needs to reliably generate structured tool calls to interact with external systems. It's particularly suited for environments where computational resources are limited but precise tool interaction is required.