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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

seanpoyner/smolcode-coder-cpp-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 for small language models. It excels at integrating with runtimes like Ollama and llama.cpp for tool-use scenarios, making it ideal for agentic coding assistants.

Loading preview...

Overview

This model, smolcode-coder-cpp-3b-tools, is a LoRA fine-tune of the Qwen2.5-Coder-1.5B-Instruct base model, developed by seanpoyner. Its primary innovation lies in teaching the model to emit native <tool_call> function calls, a crucial feature for enabling agentic coding loops in small language models (SLMs).

Key Capabilities & Differentiators

  • Native Tool Calling: Unlike standard small Qwen-Coder models that output plain-text JSON for tool calls, this fine-tune generates the native <tool_call> format, which is directly parsed by runtimes like Ollama and llama.cpp.
  • Agentic Coding: Designed to drive agentic coding assistants, specifically built for the smolcode project.
  • Optimized Training: Trained using bf16 LoRA with assistant-only loss, focusing on tool calls and final answers. The training data includes NousResearch/hermes-function-calling-v1 and synthetic smolcode tool-use trajectories, ensuring byte-identical training targets to inference prompts.
  • Qwen2.5 Chat Template: Utilizes the standard Qwen2.5 chat template with tools= for seamless integration.

Why Use This Model?

This model addresses a critical limitation in small coder models by enabling robust, native tool-use. If your application requires a compact, efficient model capable of driving agentic workflows through structured function calls, this fine-tune provides the necessary capability without the overhead of larger models. It's particularly well-suited for scenarios where direct integration with tool-parsing runtimes is essential for an effective agentic loop.