OmegaFold

1-5 minutesPaperCodeNeed >5GB

Predict protein structure without MSA

Readiness

Readiness: Need >5GB
Smoke: not_tested
Artifacts: not_applicable
Production: need_gt5gb_omegafold_pytorch_checkpoint_runtime
Storage: >5
Strategy: large shared PyTorch/CUDA OmegaFold SIF plus mirrored release1.pt checkpoint phase
Next: Stage OmegaFold source, mirror release1.pt into shared storage, build or assign compatible PyTorch/CUDA SIF, then run tiny FASTA Slurm fold before adapter wiring.

OmegaFold source identified as HeliXonProtein/OmegaFold main rev 313c873ad190b64506a497c926649e15fcd88fcd, latest tag v1.1.0 cd8b5adb5e71672f13e6226b5c7fb010560a07de. Upstream first run downloads release1.pt from helixon S3 into ~/.cache/omegafold_ckpt/model.pt; S3 HEAD/connect timed out during triage, exact size not verified. No local OmegaFold env/container/cache found and no Slurm smoke submitted. 2026-07-05 SOP triage: backend inventory found no OmegaFold source/runtime/cache. Upstream needs a compatible PyTorch/CUDA runtime and release1.pt checkpoint; prior S3 check timed out. Reclassified from Smoke Pending to Need >5GB for large shared model/runtime phase.