BoltzGen

5-30 minutesPaperCodeProduction Ready

De novo binder design for nanobodies, antibodies, proteins, peptides, and cyclotides

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_ready_api_slurm_boltzgen_inverse_folding_shared_sif
Storage: 5-20GB shared runtime/cache installed
Strategy: Production Ready via shared python312_slim.sif, boltz_py312_site plus boltzgen_py312_site, and BoltzGen cache under /media/nik/seagate_nik/bio_server/cache/boltzgen.
Next: Production Ready for bounded BoltzGen inverse-folding/protein-redesign path. Optional hardening: expose protocol/input YAML controls, support larger design batches on higher-memory GPUs, and build a compiler-enabled SIF if Triton kernels are needed.

2026-07-18T17:33:47+08:00 SOP large-tool batch: remains Need >5GB. Official BoltzGen docs provide boltzgen download/run and note weights/data artifacts can be downloaded automatically or with boltzgen download all. Backend does not yet have BoltzGen package/model artifacts or runner adapter; current runner is boltzgen-smoke placeholder. 2026-07-18T18:39:16+08:00 SOP Need >5GB active batch: Partial shared runtime staged on backend: repo /media/nik/seagate_nik/bio_server/repos/boltzgen, Python layer /media/nik/seagate_nik/bio_server/envs/boltzgen_py312_site (1.9GB), cache /media/nik/seagate_nik/bio_server/cache/boltzgen. Installed boltzgen 0.3.2 and CUDA cuequivariance/biotite dependencies over shared Boltz site; `boltzgen --help` works. `boltzgen download inverse-fold --cache ...` succeeded and `boltzgen check example/inverse_folding/1brs.yaml` produced /media/nik/seagate_nik/bio_server/runs/boltzgen/check_inverse_folding_20260718_174159/check/1brs.cif. Direct `--only_inverse_fold` run was interrupted after unauthenticated Hugging Face warning and did not yield production artifacts; full design checkpoints and Slurm/API adapter still pending. 2026-07-18T19:02:14+08:00 SOP Need >5GB active install batch: No further production validation completed in this continuation; prior partial BoltzGen install/cache/check remains staged. Needs authenticated/full model download and Slurm run. 2026-07-18T20:49:48+08:00 SOP targeted make-ready/reclassify batch: Promoted to Production Ready. Completed `boltzgen download all` into /media/nik/seagate_nik/bio_server/cache/boltzgen (about 9.0GB), validated direct inverse-folding/design run with real CIF/CSV/PDF artifacts, added real `boltzgen` runner adapter, changed registry from boltzgen-smoke to boltzgen/gpu, and validated backend runner API job a9e4ac60a4cf / Slurm 824. Verified artifacts: rank1_design.cif 144,767 bytes, results.csv 5,324 bytes, results_overview.pdf 183,466 bytes, summary.json with completed status. Adapter uses shared python312_slim.sif, disables custom kernels because the slim image lacks a C compiler, and defaults to 12G memory for the current backend node.