seanpoyner/smolcode-coder-py-1.5b-tools
The seanpoyner/smolcode-coder-py-1.5b-tools model is a 1.5 billion parameter LoRA fine-tune of Qwen2.5-Coder-1.5B-Instruct, specifically designed to emit native function calls. With a 32768 token context length, this model enables small language models to drive agentic coding loops by correctly parsing tool calls. It is optimized for integration into agentic coding assistants like smolcode, addressing the limitation of base Qwen-Coder models that describe tool calls as plain-text JSON.
Loading preview...
Overview
seanpoyner/smolcode-coder-py-1.5b-tools is a 1.5 billion parameter model, fine-tuned from Qwen2.5-Coder-1.5B-Instruct, with a 32768 token context length. Its primary purpose is to enable small language models (SLMs) to effectively utilize tools by emitting native <tool_call> function calls, which are directly parsed by runtimes like Ollama and llama.cpp. This addresses a key limitation where base Qwen-Coder models would output tool calls as plain-text JSON, breaking agentic workflows.
Key Capabilities
- Native Tool Call Emission: Generates
<tool_call>{"name": ..., "arguments": ...}</tool_call>format, crucial for agentic tool-use loops. - Agentic Coding: Specifically designed to drive agentic coding assistants, such as
smolcode. - Efficient Tool Use: Allows tiny (\u22642B parameter) models to perform complex tasks requiring external tools.
Training Details
The model was trained using bf16 LoRA (r=16, \u03b1=32) on attention and MLP projections, with assistant-only loss focused on tool calls and final answers. The training data included NousResearch/hermes-function-calling-v1 for breadth and synthetic smolcode tool-use trajectories for sharpness on specific tools. A critical aspect of training was using the same apply_chat_template(tools=...) for both training and inference to ensure byte-identical targets and prevent template mismatches.
Good For
- Developers building agentic coding assistants that require robust tool-calling capabilities from small models.
- Use cases where efficient and accurate function calling is paramount for automating coding tasks.
- Environments where resource-constrained SLMs need to interact with external tools effectively.