Installation Guide

Installation Guide#

We recommend installing mTopic in a fresh environment to avoid dependency conflicts.

Prerequisites#

Create environment#

conda create -n mtopic python=3.11 -y
conda activate mtopic
python -m pip install -U pip

Install PyTorch#

mTopic requires PyTorch, which must be installed before installing mTopic.

PyTorch wheels already include the CUDA runtime, so a local CUDA toolkit is not required.

NVIDIA GPU (recommended)

pip install torch --index-url https://download.pytorch.org/whl/cu124

This installs the CUDA 12.4 build. If your GPU is not detected or you encounter CUDA/driver errors, install a build compatible with your system instead (e.g., cu121 or cu118).

You can check your NVIDIA driver and CUDA compatibility with:

nvidia-smi

Browse all available compute platforms (cpu, cu118, cu121, cu124, cu126, etc.) at https://download.pytorch.org/whl/.

CPU-only (no GPU)

pip install torch --index-url https://download.pytorch.org/whl/cpu

Install mTopic#

git clone https://github.com/TabakaLab/mTopic.git
pip install mTopic/.

Verification#

After installation, verify that PyTorch works correctly:

python -c "import torch; print('torch:', torch.__version__); print('cuda available:', torch.cuda.is_available())"

  • True - GPU is active

  • False - running on CPU

Jupyter notebook kernel#

To create a Jupyter notebook kernel, run:

python -m ipykernel install --user --name mtopic --display-name "Python (mTopic)"