README.md

---
title: CodeCraft - Qwen3-Coder-30B
emoji: 💻
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 5.23.3
app_file: app.py
pinned: false
python_version: "3.12"
---

# CodeCraft - AI Coding Assistant

Powered by **Qwen3-Coder-30B-A3B-Instruct** running on Hugging Face ZeroGPU (xlarge).
A Mixture-of-Experts model with 30B total params (3B active) — matches or exceeds much larger models on coding benchmarks.

Chat with a cutting-edge coding assistant. Supports code generation, debugging, refactoring, explanation, and general programming help across all major languages.

## Features

- 💬 **Chat interface** with syntax-highlighted code blocks
- ⚙️ **Adjustable parameters**: temperature, top-p, max tokens (up to 8192), system prompt
- 📡 **Built-in API endpoint** at `/gradio_api/call/predict` for programmatic use
- 🎨 **Syntax-highlighted output** via Gradio Markdown + code blocks
- 🔄 **256K context** — handle entire codebases in conversation
- 🛠️ **Agentic tool-use** support (function calling, ReAct, structured output)

## API Usage

Every Gradio Space exposes a REST API at `/gradio_api/call/predict`. For this Space:

```python
import requests

response = requests.post(
    "https://serotoninboi-codecraft.hf.space/gradio_api/call/predict",
    json={
        "data": [
            "write a fibonacci function in rust",
            "You are a helpful coding assistant.",
            0.3,
            0.9,
            2048
        ]
    }
)
print(response.json())
```

## Model

[Qwen/Qwen3-Coder-30B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct) — 30B MoE (3B active) code-specific LLM with 256K context, built for agentic coding tasks.