matthewfeickert

# SCM syntax highlighting
pixi.lock linguist-language=YAML linguist-generated=true

matthewfeickert

# SCM syntax highlighting
pixi.lock linguist-language=YAML linguist-generated=true

matthewfeickert

# GitHub syntax highlighting
pixi.lock linguist-language=YAML linguist-generated=true

matthewfeickert

# GitHub syntax highlighting
pixi.lock linguist-language=YAML linguist-generated=true

matthewfeickert

# GitHub syntax highlighting
pixi.lock linguist-language=YAML linguist-generated=true

matthewfeickert

pixi resolution order for composing features to make environments

$ pixi --version
pixi 0.30.0

For testing purposes, I'm trying to create lock files for an ecosystem of tools across multiple Python versions. To avoid having to repeat describing the dependencies, I'd like to use pixi's features to define the tools once in one feature (with lower bounds only as this would be part of nightly/weekly solves) and then have the Python version exist in a separate feature. The goal here is to be able to create environments quickly by composing features together like

matthewfeickert

FROM mambaorg/micromamba:1.5.9-bookworm-slim

USER mambauser

SHELL [ "/bin/bash", "-c" ]

# Use C.UTF-8 locale to avoid issues with unicode encoding
ENV LC_ALL=C.UTF-8
ENV LANG=C.UTF-8

matthewfeickert

# build products
output/
example
# pixi environments
.pixi
*.egg-info

matthewfeickert

Minimal PyTorch CUDA environment with pixi

Based off of the pixi docs example

Steps

0. Install pixi
   • curl -fsSL https://pixi.sh/install.sh | bash
1. pixi run detect-gpu

• If this is your first time doing anything in the environment pixi will run pixi install for you first.

matthewfeickert

Use VS Code to run notebooks on remote servers in environments built on top of LCG views

(Note that this whole thing is a bit cursed as LCG views are a terrible way to distribute Python environments)

Setup

Get a working VS Code server on the remote

Requires: