Unity Coder 7B

Unity Coder 7B

Answers Unity game development questions and writes C# scripts, XR code and editor tools, as a fine-tune of Qwen2.5-Coder-7B-Instruct. For Unity developers who want a local coding assistant.

Made by vishnuOI on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Unity Coder 7B is a fine-tune of Qwen2.5-Coder-7B-Instruct for Unity game development in C#. It was trained with QLoRA on 1,687 Unity C# instruction pairs over three epochs. The record lists about 7.6B parameters.

Who it suits: It suits Unity developers who want help with scripts, XR developers, and tool builders who write editor extensions.

Industry | Who uses it | How they use it | Example
Game development | Unity developers | Ask for C# scripts and fixes | How to detect a collision between two objects
XR and VR | XR developers | Get help with OpenXR and the XR Interaction Toolkit | Ask how to set up a grab interaction
Tools | Editor tool authors | Draft editor scripts and tooling | Ask for an editor script that automates a repeated task

Anyone who works outside Unity or C# can skip it, since the tuning targets that one area.

Where to use it:

• Python with Transformers
• A GPU or a machine with enough memory for the weights
• Windows, macOS or Linux, where the files download with Python

Use case | Who | What they get back
Unity C# scripting | Unity developers | Code for MonoBehaviour, ScriptableObjects and coroutines
XR development | XR developers | Guidance on OpenXR and the XR Interaction Toolkit
Physics, animation and UI | Game developers | Help with physics, animation and UI Toolkit
Performance work | Developers | Optimization patterns

How to set it up:

• Python with pip, plus Transformers and PyTorch
• Memory for about 15.2 GB in BF16, or about 3.8 GB at 4-bit as an estimate

1. Install the Hugging Face hub client.
2. Download the model files, or let Transformers fetch them by name.
3. Run the usage snippet below.

pip install -U huggingface_hub
hf download vishnuOI/unity-coder-7b

How to use it: Load the model and tokenizer, build a chat prompt with a system message and your question, then call generate and decode the reply.

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "vishnuOI/unity-coder-7b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are an expert Unity game developer specializing in C# scripting and XR development."},
    {"role": "user", "content": "How do I detect collision between two objects in Unity?"},
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, do_sample=True)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Tool | What it does
from_pretrained | Loads the tokenizer and the model in bfloat16
apply_chat_template | Turns the system and user messages into one prompt
generate | Writes the reply, up to 512 new tokens at temperature 0.7

Example: You send the snippet's question about detecting a collision between two objects in Unity, with a system message that names the model an expert Unity developer. It replies with a written answer, and you test the code in your own project.

For newcomers and experts: Think of it as a coding helper that already knows Unity vocabulary. Run the snippet as written, then swap in your own question.

The tuning used QLoRA with 4-bit NF4 weights, rank 16 and alpha 32. The dataset of 1,687 pairs is small, so expect it to cover common Unity tasks better than rare ones.

Start from Qwen2.5-Coder-7B-Instruct
Fine-tune with QLoRA on Unity C# pairs
Load the result with Transformers
Ask a Unity question and read the reply

Good to know:

• Training: 1,687 instruction pairs and 3 epochs, which is a small set
• Size: about 7.6B parameters, and the record lists English
• License: CC BY 4.0
• Cost to run: none named
• Data it reaches: the prompts you send to the model, with no web or account access named
• Not for: work outside Unity and C#, which the card does not target

On Hugging Face, 13 November 2026

• Parameters: 7.6B
• Weights: about 15.2 GB in BF16; about 3.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 146
• Likes: 4
• License: cc-by-4.0
• Languages: en
• Based on: Qwen/Qwen2.5-Coder-7B-Instruct
• Quantized versions: 2
• Tags: safetensors, qwen2, unity, unity3d, game-development, csharp, openxr, code, finetuned

Links

• Model: https://huggingface.co/vishnuOI/unity-coder-7b
• Files: https://huggingface.co/vishnuOI/unity-coder-7b/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download vishnuOI/unity-coder-7b

Then follow the usage in its model card. Nothing in this kit runs by itself.