FLOWR

New5-30 minutesPaperCodeNeed >5GB

Generate small molecules

Readiness

Readiness: Need >5GB
Smoke: not_run
Artifacts: not_run
Production: need_gt5gb_flowr_pytorch_geometric_checkpoint_large_gpu_runtime
Storage: 5-20GB
Strategy: large FLOWR PyTorch/Lightning/Torch-Geometric/chem SIF with checkpoints and suitable GPU
Next: Build the full FLOWR SIF/overlay, stage checkpoints, and validate on a suitable GPU or prove reduced inference feasibility before adapter work.

No obvious large model/database dependency identified; validate upstream docs before install. Worker evidence 2026-07-01: source https://github.com/jule-c/flowr_root revision 1b1e6d9e0b5deeb8c7cdf2f729bf56935262fba5 staged at /media/nik/seagate_nik/bio_server/tools/flowr. Real Slurm sanity probe job 284 on cpu completed exit 0 in 3s. Artifacts under /media/nik/seagate_nik/bio_server/tools/flowr/artifacts/sop_probe_20260701_164543: SOP_REPORT.md, probe_stdout.log, slurm logs, flowr_sanity.py, flowr_slurm_probe.sh. Sanity: Python compile passed; RDKit read examples/cdk2_ligands.sdf with 16 valid molecules and 3D conformers; examples/cdk2_protein.pdb has 9011 atom records and 555 residue records. Full inference blocked: no checkpoint, incomplete runtime stack (missing torch/lightning/torch_geometric/prolif), shared SIF lacks rdkit/torch/numpy/Bio/gemmi/openbabel, upstream recommends >=40GB VRAM while backend RTX 5050 has 8GB. 2026-07-07 17:10 +0800: SOP Production Blocked audit: moved to Need >5GB because only compile/RDKit fixture sanity passed; full inference needs checkpoints, a PyTorch/Lightning/Torch-Geometric/ProLIF chem stack, and upstream recommends >=40GB VRAM while backend GPU has about 8GB.