sue-env-install

What problem does it solve?

Manually configuring runtime environments for ML/HPC scale-up experiments across different HPC sandboxes leads to configuration drift, missing dependencies, failed runs, and wasted compute hours due to inconsistent or non-compliant environment setups.

Core Features & Use Cases

• Multi-backend environment setup: Handles conda/uv environment creation for Snellius, custom Singularity SIF builds for LUMI, and coordinates RunPod-based builds for offline NM5 clusters.
• Dependency and cache management: Installs required Python packages, pre-warms HuggingFace model/checkpoint caches in correct workspace locations, and verifies GPU/import readiness before experiment launches.
• Safety and compliance guardrails: Enforces inode-safe installation paths, prevents forbidden loose environment creation on restricted backends like LUMI, and validates all environment paths against the authoritative runtime.yaml config to avoid misconfiguration. Use case example: A research team running video understanding experiments on LUMI can use this skill to automatically build a custom SIF with required processing packages, verify CUDA visibility, and update the runtime config so subsequent dryrun and fullrun jobs use the correct environment without manual intervention.

Quick Start

Use the sue-env-install skill to set up the required runtime environment for your current DeepResearch scale-up experiment on the selected HPC backend.