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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The seanpoyner/smolcode-coder-git-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 designed to emit native function calls, enabling agentic coding loops for small language models. It excels at integrating tool use within coding assistants, addressing the limitation of base models that output plain-text JSON for tool descriptions. With a context length of 32768 tokens, it is optimized for driving agentic coding workflows.

Loading preview...

Overview

This model, seanpoyner/smolcode-coder-git-3b-tools, is a LoRA fine-tune of the Qwen2.5-Coder-1.5B-Instruct base model, specifically engineered to enable agentic coding loops. It addresses a critical limitation in smaller Qwen-Coder models by teaching them to emit native <tool_call> function calls, rather than plain-text JSON, which is essential for seamless integration with runtime environments like Ollama and llama.cpp.

Key Capabilities

  • Native Tool Call Emission: Generates <tool_call> formatted responses for tool use, crucial for agentic workflows.
  • Agentic Coding Support: Designed to drive agentic coding assistants, particularly for the smolcode project.
  • Efficient Fine-tuning: Utilizes bf16 LoRA with assistant-only loss, focusing training on tool calls and final answers.
  • Template Consistency: Trained and served using the same apply_chat_template(tools=...) to ensure byte-identical training targets and inference prompts.

Training Details

  • Base Model: Qwen/Qwen2.5-Coder-1.5B-Instruct.
  • Data: Leveraged NousResearch/hermes-function-calling-v1 for breadth and synthetic smolcode tool-use trajectories for specific sharpness.
  • Methodology: 3 epochs of training with a full 2048 sequence length, ensuring robust learning of tool-calling patterns.

Good For

  • Developers building agentic coding assistants that require precise tool-call formatting.
  • Use cases where small, efficient models need to interact with external tools via structured function calls.
  • Scenarios demanding a reliable and consistent tool-use mechanism from a compact language model.

Finetunes
1 models