seanpoyner/smolcode-coder-rust-3b-tools

Overview

This model, seanpoyner/smolcode-coder-rust-3b-tools, is a 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 emit native <tool_call> function calls, which is crucial for driving agentic coding loops. This addresses a limitation where base Qwen-Coder models typically describe tool calls as plain-text JSON, breaking agentic tool-use workflows.

Key Capabilities

• Native Tool Call Emission: Specifically trained to output <tool_call>{"name": ..., "arguments": ...}</tool_call> format, compatible with runtimes like Ollama and llama.cpp.
• Agentic Workflow Integration: Designed to seamlessly integrate with agentic coding assistants, such as smolcode.
• Optimized for Small Models: Brings advanced tool-use capabilities to tiny (≤2B parameter) models, making them more effective in constrained environments.

Training Details

The model was trained using bf16 LoRA (r=16, α=32) with assistant-only loss, focusing on tool calls and final answers. The training data combined NousResearch/hermes-function-calling-v1 for breadth and synthetic smolcode tool-use trajectories for sharpness, all processed through the exact apply_chat_template(tools=...) used during inference to ensure byte-identical training targets. Version 2 specifically rectifies template mismatches found in v1, ensuring robust free-generation tool-call parsing.

Ideal Use Cases

• Agentic Coding Assistants: Perfect for systems requiring small, efficient models to interact with external tools and APIs for code generation and execution.
• Resource-Constrained Environments: Enables sophisticated tool-use capabilities on hardware where larger models are impractical.
• Automated Code Generation: Facilitates more reliable and structured interaction with coding tools within automated development workflows.