How to Install PyTorch and TensorFlow on a VPS

PyTorch and TensorFlow are the two dominant machine learning frameworks, used for building, training, and running neural network models. This guide covers installing both (or either) on a CPU-only VPS.

Choosing Between PyTorch and TensorFlow

Both are capable, well-maintained frameworks — PyTorch is generally favored in research and has become the more common choice for many newer open-source models; TensorFlow remains widely used, particularly in some production deployment contexts. If you don't have a specific reason to prefer one, check which framework your specific model/project actually requires.

Step 1 — Set Up a Python Virtual Environment

python3 -m venv ml-env
source ml-env/bin/activate

Using a dedicated virtual environment avoids dependency conflicts with other Python projects on the same server — see How to Install Python on Ubuntu & Debian for the base setup.

Step 2 — Install PyTorch (CPU Version)

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

The CPU-specific index avoids downloading unnecessary GPU-related components, resulting in a smaller install.

Step 3 — Verify PyTorch Installation

python3 -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"

cuda.is_available() correctly returns False on a CPU-only server — this is expected, not an error.

Step 4 — Install TensorFlow (CPU Version)

pip install tensorflow-cpu

Step 5 — Verify TensorFlow Installation

python3 -c "import tensorflow as tf; print(tf.__version__)"

Running a Simple Test Script (PyTorch)

import torch

x = torch.rand(3, 3)
print(x)
print(x @ x.T)  # basic matrix operation

Installing Additional Common Libraries

pip install numpy pandas scikit-learn matplotlib

Managing Memory for Larger Models

CPU-based deep learning can be memory-intensive — monitor usage with htop and consider setting explicit batch size limits in your code to avoid out-of-memory errors on smaller VPS plans.

Using a requirements.txt for Reproducibility

pip freeze > requirements.txt

pip install -r requirements.txt

Ensures consistent package versions if you redeploy this environment on another server later.

Common Errors

"Killed" message with no other error during training — almost always an out-of-memory condition where the OOM killer terminated the process; reduce batch size or model complexity, or check available RAM with free -h.

Installation is extremely slow — some ML package installations involve large downloads; ensure adequate disk space and a stable connection, and be patient with the initial install.

Continue Reading

Browse more articles in AI & Machine Learning on a VPS.

Databáze řešení

Kategorie

Kategorie