SuperWater

5-30 minutesPaperCodeNeed >5GB

Add waters to protein

Readiness

Readiness: Need >5GB
Smoke: passed
Artifacts: passed
Production: need_gt5gb
Storage: 5-20
Strategy: large shared SIF/database/model bundle phase
Next: Build a CUDA SIF/overlay with PyTorch 2.5.1, matching PyG CUDA wheels, e3nn, RDKit/OpenBabel, and ESM-2 cache; run upstream SuperWater GPU benchmark through Slurm, then wire the adapter.

2026-07-17T13:13:16+08:00 SOP Need >5GB batch: remains Need >5GB. Source/checkpoints are partially staged, but production requires a CUDA PyTorch 2.5.1 + PyG + e3nn + RDKit/OpenBabel runtime and ESM-2 cache; no real runner adapter or Slurm benchmark is validated yet.