OpenAI Jukebox Server Setup

This document outlines the steps to set up and troubleshoot OpenAI Jukebox on an Ubuntu 22.04 server with NVIDIA GPU.

---

1. Operating System

• System: Ubuntu 22.04 LTS
• Steps Taken:
  • Confirmed OS compatibility for NVIDIA drivers and CUDA using:

    lsb_release -a

  • No changes needed.

---

2. NVIDIA Driver

• Installed Driver: 535.183.01
• Steps Taken:
  • Verified the driver compatibility with nvidia-smi. Output confirmed CUDA runtime version compatibility (12.2 supported, fine for CUDA 11.1):

    nvidia-smi

  • Skipped driver installation during CUDA Toolkit setup as the current driver was newer and compatible.

---

3. CUDA Toolkit

• Installed Version: CUDA 11.1
• Steps Taken:
  1. Downloaded and installed CUDA 11.1 (Ubuntu 20.04 version) because Ubuntu 22.04 is not officially supported for this version:

     wget https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run
     sudo sh cuda_11.1.1_455.32.00_linux.run

  2. Selected "Install Toolkit Only" and skipped driver installation.
  3. Added CUDA paths to .bashrc:

     echo 'export PATH=/usr/local/cuda-11.1/bin:$PATH' >> ~/.bashrc
     echo 'export LD_LIBRARY_PATH=/usr/local/cuda-11.1/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc
     source ~/.bashrc

  4. Verified the installation:

     nvcc --version

---

4. Python Environment

• Python Version: 3.8
• Steps Taken:
  1. Created and activated a new Conda environment:

     conda create --name jukebox_env python=3.8 -y
     conda activate jukebox_env

  2. Installed PyTorch with CUDA 11.1:

     pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html

---

5. Jukebox Installation

• Steps Taken:
  1. Set up the local repository:
     • Avoided recloning since the repository was already configured.
     • Verified the directory structure and existing files.
  2. Installed jukebox as an editable module:

     pip install -e .

---

6. Issues and Resolutions

1. CUDA Compatibility:

   • Installed CUDA 11.1 for compatibility with PyTorch 1.8.0.
   • Skipped unnecessary driver installation to avoid conflicts with the already installed NVIDIA driver.

2. NumPy and Numba Conflict:

   • Error: AttributeError: module 'numpy' has no attribute 'long'.
   • Resolution: Downgraded NumPy to version 1.19.5 for compatibility with Numba:

     pip install numpy==1.19.5

3. Module Import Error:

   • Error: ModuleNotFoundError: No module named 'jukebox'.
   • Resolution: Installed jukebox as an editable package with pip install -e ..

---

7. Verifications

1. Confirmed GPU and CUDA availability in PyTorch:

   python -c "import torch; print(torch.cuda.is_available()); print(torch.cuda.get_device_name(0))"