How does the skill handle CUDA package license conflicts?

The skill implements explicit priority values in the Flox configuration (toml), ensuring that conflicting files like LICENSE headers are resolved correctly during environment activation.

Does this help with Python deep learning libraries?

Absolutely. It contains specific patterns for setting up PyTorch and TensorFlow within virtual environments that are correctly linked to the Flox-managed CUDA runtime and libraries.

Can I use this for legacy CUDA versions?

Yes, the skill includes templates and patterns for both current CUDA 12.x versions and legacy support for CUDA 11.x environments.

Does this Flox skill support macOS development?

CUDA is restricted to Linux (aarch64 and x86_64). However, this skill provides specific configuration patterns to allow your environments to gracefully fall back to CPU or Metal packages on Darwin/macOS.

Flox CUDA Development

Name: Flox CUDA Development
Author: flox

byflox

•

Data Science & ML

Streamlines NVIDIA CUDA setup and GPU computing environments using Flox for deep learning and high-performance development.

The Flox CUDA skill provides standardized patterns and best practices for managing complex GPU development environments within the Flox ecosystem. It helps developers navigate the complexities of CUDA toolkit installations, library dependencies like cuDNN and NCCL, and specific package versioning for Linux systems while offering robust patterns for cross-platform portability. This skill is essential for engineers and researchers building AI models, deep learning frameworks such as PyTorch or TensorFlow, and high-performance computing applications that require reproducible, isolated NVIDIA-backed environments.

Key Features

013 GitHub stars

02Pre-configured environment templates for PyTorch and TensorFlow GPU acceleration.

03Standardized conflict resolution patterns using explicit Flox package priorities.

04Dynamic GPU detection logic for cross-platform Linux (CUDA) and macOS (CPU/Metal) fallbacks.

05Modular environment architecture for separating base CUDA, math libraries, and debugging tools.

06Automated CUDA toolkit and library discovery including cuDNN, cuBLAS, and NCCL.

Use Cases

01Configuring multi-GPU development workstations with NCCL communication and Nsight debugging tools.

02Setting up reproducible GPU-accelerated environments for Deep Learning model training and inference.

03Creating portable development environments that automatically scale from local macOS CPUs to Linux GPU clusters.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add flox/flox-agentic flox-cuda

For use in Claude.ai and ChatGPT

Download Skill