Table of contents

tensorflow — TensorFlow + Jupyter

A GPU-ready TensorFlow environment with JupyterHub, SSH, and standard data-science packages. Ships in two flavors for newer and older NVIDIA hardware.

Image tags

  • docker.io/manvarharsh/tensorflow:cuda12 — CUDA 12.4 + TensorFlow 2.18 (RTX 40xx, L4, L40S, H100)
  • docker.io/manvarharsh/tensorflow:cuda11 — CUDA 11.8 + TensorFlow 2.13 (RTX 30xx, Ampere, Volta, Turing)

What’s in this image

  • Base: nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04 (or 11.8 variant)
  • Python 3.10 (conda)
  • TensorFlow 2.18 (cuda12) or 2.13 (cuda11)
  • NumPy, Pandas, Matplotlib, scikit-learn
  • JupyterHub with the Podstack authenticator
  • NVDashboard GPU monitoring
  • OpenSSH server

Default ports

PortService
22SSH
8000JupyterHub

Use cases

  • TensorFlow / Keras model training
  • TFRecord-based pipelines
  • TPU-compatible code paths
  • General notebook-driven TF experimentation

Environment variables

VariableDescription
ENABLE_SSHEnable SSH server
ENABLE_JUPYTERHUBEnable JupyterHub on port 8000
PODSTACK_API_URLBackend URL for JupyterHub token validation
SSH_PUBLIC_KEYPublic key for SSH

Persistence

Mount at /data. Notebooks under /data/notebooks/, models under /data/models/, datasets under /data/datasets/.

See also