To run your own container in AWS Sagemaker (Training Job) it’s required to have mamba. So how would you use pixi instead?
It’s quite straight-forward, first add micromamba as part of your dependencies in pixi.toml. Then you update your docker to something like this:
FROM --platform=linux/amd64 ghcr.io/pixi
COPY pixi.toml pixi.lock ./
ENV PIP_DEFAULT_TIMEOUT=360
ENV UV_HTTP_TIMEOUT=360
RUN pixi install --frozen # use .lock-file
# === IMPORTANT PART ===
# Link micromamba as mamba, to support AWS sagemaker
# Replace <default> with the environment name you use
# 1. link environments `micromamba` to `mamba`
RUN ln -s /.pixi/envs/default/bin/micromamba /.pixi/envs/default/bin/mamba
# 2. Add pixi to PATH
ENV PATH="/.pixi/envs/default/bin:${PATH}"
# 3. Add MAMBA_ROOT_PREFIX
ENV MAMBA_ROOT_PREFIX=/.pixi
# This isn't used by AWS but nice-to-have for local testing
# RUN pixi shell-hook -e default > /shell-hook.sh
# RUN echo 'exec "$@"' >> /shell-hook.sh
# ENTRYPOINT ["/bin/bash", "/shell-hook.sh"]This’ll create a Docker image that’s usable inside AWS Sagemaker and let you keep using pixi throughout your whole dev-cycle, remote and local!
Perhaps there’s a easier way, if you know one please share in comments below! 👇
That’s it for now,
~Hampus Londögård