Skip to content

kaggle images

Mirrors the Kaggle notebook environment. Package versions are pinned to match a Kaggle notebook pip freeze so code developed locally behaves identically when submitted to competitions. Three variants: NVIDIA GPU (with RAPIDS), CPU-only, and Apple Silicon.


kaggle-nvidia

Uses a custom PyTorch 2.10.0 wheel with Pascal support; Kaggle's own stock cu128 wheel has no Pascal kernels, so torch on Kaggle's P100s silently fails. Same version, but it actually works. Includes the RAPIDS stack as found in Kaggle GPU sessions.

RAPIDS hardware requirement

RAPIDS (cuDF/cuML/CuPy) requires Volta (sm_70) or newer. It is not functional on P100/GTX 10xx hardware; the same limitation applies on Kaggle itself.

Component Version
Base image nvidia/cuda:12.8.1-cudnn-runtime-ubuntu24.04
Python 3.12
CUDA 12.8.1
PyTorch 2.10.0 (custom wheel, sm_60-sm_100)
TensorFlow 2.20.0
cuDF 26.2.1
cuML 26.2.0
CuPy 14.0.1

ML packages: scikit-learn 1.6.1, xgboost 3.2.0, lightgbm 4.6.0, catboost 1.2.10, transformers 5.0.0, keras 3.13.2

Other packages: numpy 2.0.2, pandas 2.3.3, polars, jupyterlab 3.6.8, optuna, shap, opencv, albumentations, kaggle, kagglehub

DockerHub: gperdrizet/kaggle-nvidia


kaggle-cpu

Component Version
Base image python:3.12-slim
Python 3.12
PyTorch 2.10.0 (CPU)
TensorFlow 2.20.0

ML packages: scikit-learn 1.6.1, xgboost 3.2.0, lightgbm 4.6.0, catboost 1.2.10, transformers 5.0.0, keras 3.13.2

Other packages: numpy 2.0.2, pandas 2.3.3, polars, jupyterlab 3.6.8, optuna, shap, opencv, albumentations, kaggle, kagglehub

DockerHub: gperdrizet/kaggle-cpu


kaggle-mac

Native linux/arm64 image for Apple Silicon (M1/M2/M3). No Rosetta emulation. GPU acceleration is not available inside Docker on macOS.

Component Version
Base image python:3.12-slim
Platform linux/arm64
Python 3.12
PyTorch 2.10.0 (CPU, ARM64)
TensorFlow 2.20.0

ML packages: scikit-learn 1.6.1, xgboost 3.2.0, lightgbm 4.6.0, catboost 1.2.10, transformers 5.0.0, keras 3.13.2

Other packages: numpy 2.0.2, pandas 2.3.3, polars, jupyterlab 3.6.8, optuna, shap, opencv, albumentations, kaggle, kagglehub

DockerHub: gperdrizet/kaggle-mac