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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 14, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The seanpoyner/smolcode-coder-cpp-1.5b-tools is a 1.5 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, particularly for applications like smolcode, with a context length of 32768 tokens.

Loading preview...

Overview

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

Key Capabilities

  • Native Tool Calling: Unlike standard Qwen-Coder models that output tool calls as plain-text JSON, this fine-tune generates <tool_call> formatted output, which is directly parseable by runtimes like Ollama and llama.cpp.
  • Agentic Coding: Designed to drive agentic coding assistants, specifically built for the smolcode project.
  • Efficient Training: Utilizes bf16 LoRA fine-tuning with assistant-only loss, focusing on tool calls and final answers.
  • Robust Data: Trained on a combination of NousResearch/hermes-function-calling-v1 for breadth and synthetic smolcode tool-use trajectories for specific tool sharpness.

What Makes This Different?

Many small coder models struggle with native tool call generation, often outputting JSON that requires additional parsing. This model directly addresses that gap, making it highly effective for integrating tool-use capabilities into resource-constrained agentic systems. The v2 iteration specifically resolves template mismatches, ensuring high free-generation tool-call parse rates.

Good For

  • Developers building agentic coding assistants that require native tool-calling capabilities.
  • Applications where a small, efficient model (1.5B parameters) needs to interact with external tools seamlessly.
  • Projects leveraging the Qwen2.5 chat template for tool-enabled interactions.

Finetunes
1 models