ルミナイ株式会社 (Lluminai) テックブログ

このリポジトリは、ルミナイ株式会社がZennで公開しているテックブログの記事で利用したソースコードやサンプルコードを管理するものです。

📝 Zenn テックブログ

記事はこちらで公開しています。

• ルミナイ株式会社 Tech Blog: https://zenn.dev/p/lluminai_tech

📂 ディレクトリ構成

各記事に対応するディレクトリ内に、関連するソースコードが格納されています。

/
├── article-slug-1/      # 記事1のサンプルコード
│   ├── src/
│   └── README.md
├── article-slug-2/      # 記事2のサンプルコード
│   ├── src/
│   └── README.md
└── ...

🚀 使い方

各ディレクトリのREADME.mdに、コードの概要や実行方法を記載していますので、そちらを参照してください。

📄 ライセンス

このリポジトリ内のソースコードは、MIT License のもとで公開されています。

🏢 会社情報

• [ルミナイ株式会社 公式サイト] https://lluminai.co.jp/

PyTorch & Ollama GPU Development Environment with Docker

This is a template for building a PyTorch and Ollama development environment that utilizes NVIDIA GPUs with Docker Compose.

The goal is to create an isolated and reproducible environment for machine learning projects, providing a persistent Ollama server for LLM inference alongside a complete PyTorch development setup.

---

🎯 Features

• Reproducible: The Dockerfile ensures that the same environment can be built every time.
• Portable: This environment can be reproduced on any machine that runs Docker.
• GPU Enabled: Ready for model training (PyTorch) and inference (Ollama) using NVIDIA GPUs.
• Persistent Server: Starts an Ollama server automatically, with model data saved in a persistent Docker volume.
• Efficient: Start and stop the entire environment with simple Docker Compose commands.

---

🔧 Environment

• Ubuntu 22.04
• NVIDIA CUDA Toolkit 12.1
• Python 3.11
• PyTorch (with CUDA support)
• Ollama Server

---

✅ Prerequisites

Before you begin, ensure you have the following installed and configured on your host machine:

• An NVIDIA GPU
• An up-to-date NVIDIA Driver
• Docker
• A Docker environment configured for GPU access:
  • (For Windows) WSL 2 integration enabled in Docker Desktop
  • (For Linux) NVIDIA Container Toolkit

---

🚀 Getting Started

1. Clone or Create the Files

Ensure your Dockerfile, docker-compose.yml, and this README.md are in the same directory.

2. Build and Run the Container

Run the following command. This will build the image from the Dockerfile and start the container in the background. The CMD in the Dockerfile ensures the Ollama server starts automatically.

docker compose up --build -d

3. Verify the Ollama Server (from Host)

The server is now running and accessible from your host machine (the PC running Docker) at http://localhost:11434.

You can test this by opening a new terminal (not inside the container) and running:

curl http://localhost:11434/

If it's working, you should see "Ollama is running".

4. Enter the Container for Development

The server is running in the background. To get an interactive shell inside the same running container for development (e.g., to run Python scripts or download models), execute:

docker compose exec app /bin/bash

5. Using Ollama (Inside Container)

Once you are inside the container (after running the exec command), you can use the Ollama CLI:

# List currently downloaded models
ollama list

# Pull a model (e.g., Llama 3)
ollama pull llama3

# Run a model
ollama run llama3 "Why is the sky blue?"

6. Verify GPU Access for PyTorch (Inside Container)

While still inside the container, start Python and run the following code to check if PyTorch is correctly recognizing the GPU.

# Start python
python3.11

# Inside the Python interpreter
import torch
print(f"PyTorch version: {torch.__version__}")
print(f"CUDA available: {torch.cuda.is_available()}")
if torch.cuda.is_available():
    print(f"GPU Name: {torch.cuda.get_device_name(0)}")

7. Stopping the Environment

To stop the Ollama server and the container, run:

docker compose down

(Your downloaded models will be saved in the ollama_data volume and will be available next time you run docker compose up -d).