seanpoyner/smolcode-coder-bsd-1.5b-tools

Model Overview

The seanpoyner/smolcode-coder-bsd-1.5b-tools is a 1.5 billion parameter model, fine-tuned from Qwen2.5-Coder-1.5B-Instruct with a LoRA adaptation. Its primary purpose is to enable small coder models (like those under 2 billion parameters) to emit native <tool_call> function calls, which are crucial for agentic coding loops. This addresses a common limitation where base Qwen-Coder models describe tool calls as plain-text JSON, breaking tool-use functionality in runtimes like Ollama or llama.cpp.

Key Capabilities

• Native Tool Call Emission: Specifically trained to output <tool_call>{"name": ..., "arguments": ...}</tool_call> format, ensuring compatibility with agentic tool-use systems.
• Agentic Coding Support: Designed to drive agentic coding assistants by facilitating accurate tool interaction.
• Optimized Training: Fine-tuned using NousResearch/hermes-function-calling-v1 and synthetic smolcode tool-use trajectories, with assistant-only loss focusing on tool calls and final answers.
• High Fidelity: Training data was rendered through the exact apply_chat_template(tools=...) used at inference, ensuring byte-identical training targets.

Ideal Use Cases

• SLM-Optimized Agentic Coding Assistants: Particularly suited for projects like smolcode where efficient and accurate tool-use is paramount.
• Tool-Enabled Code Generation: For applications requiring a small, efficient model to interact with external tools or APIs through structured function calls.
• Bridging Tool-Use Gaps: Useful for developers encountering issues with base models' inability to produce native tool call formats for agentic workflows.