experiment-env-installer

What problem does it solve?

This Skill solves the common, time-consuming pain points of setting up, reproducing, and maintaining Python environments for machine learning experiments across diverse compute backends, eliminating version conflicts, CUDA/ROCm driver mismatches, and inconsistent dependency setups that cause avoidable training failures and reproducibility issues.

Core Features & Use Cases

• Cross-Backend uv-Based Setup: Enforces lockfile-driven dependency management using uv for reproducible installs across local workstations, cloud GPU instances, and HPC clusters, with explicit rules against Conda, pip, and Poetry to avoid environment drift.
• Backend-Specific PyTorch Compatibility Guidance: Provides clear instructions for installing the correct PyTorch build for NVIDIA CUDA, AMD ROCm, legacy CUDA hosts, and CPU-only systems, including version compatibility checks and override examples for edge cases.
• Pre-Flight Verification & Troubleshooting: Includes step-by-step pre-launch checks for GPU visibility, import validation, and environment fingerprinting, plus a comprehensive troubleshooting guide for common install failures, ABI mismatches, and stale environment issues.
• Use Case Example: If you are migrating a training project from a local NVIDIA GPU workstation to an AMD ROCm-based HPC cluster, this Skill guides you through adjusting the PyTorch index URL, verifying driver compatibility, running smoke tests, and capturing an environment fingerprint to ensure the setup works before launching large-scale jobs.

Quick Start

Use the experiment-env-installer skill to configure a reproducible Python environment for your ML project on your current compute backend, verify GPU or ROCm compatibility, and complete all pre-flight checks before launching training runs.