README.md

---
title: Small Code Assistant
emoji: 🧑‍💻
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.29.0
app_file: app.py
pinned: false
short_description: Streaming 3B coding model on ZeroGPU
python_version: "3.12"
startup_duration_timeout: 30m
---

# Small Code Assistant

A streaming chat demo for code generation, running [Qwen/Qwen2.5-Coder-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct)
(6.2 GB bf16) on ZeroGPU.

## Why this model

The demo was originally scoped to [CohereLabs/North-Mini-Code-1.0](https://huggingface.co/CohereLabs/North-Mini-Code-1.0),
which is a 30.5B MoE (~61 GB bf16, 49 safetensors shards) — far beyond a free Space.
Qwen2.5-Coder-3B is the closest small stand-in that fits the brief (under 10 GB).

To swap models, change `MODEL_ID` at the top of `app.py`. North-Mini-Code itself is
served remotely by Cohere (see the [official demo](https://huggingface.co/spaces/akhaliq/North-Mini-Code-1.0))
and would need a Cohere API key rather than local weights.

## API

Open `/api` or use the MCP endpoint exposed via `demo.launch(mcp_server=True)`.