ProFam

5-30 minutesCodeProduction Ready

Protein family language model

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_pass
Storage: <5
Strategy: shared PyTorch CPU/GPU SIF + ProFam checkpoint overlay
Next: Production Ready. Optional hardening: expose generation mode and per-residue/diversity scoring controls after separate validation.

No obvious large model/database dependency identified; validate upstream docs before install. 2026-07-05 SOP batch recheck: backend inventory found no staged ProFam source/checkpoint/SIF/artifacts/runner. ProFam is public and has CPU/PyTorch install path with a 251M-parameter checkpoint; suitable for shared PyTorch SIF smoke after staging. Kept Smoke Pending. 2026-07-05 18:24 +08:00 backend re-audit: no staged ProFam source/checkpoint/SIF/artifacts/runner found; plausible checkpoint path is not cached. Kept Smoke Pending. 2026-07-06 21:01 +08:00 implementation attempt: installed ProFam 0.1.8 no-deps into /media/nik/seagate_nik/bio_server/envs/profam_py312_site and inspected api.py/download_checkpoint.py entry points. Missing dependency fill for accelerate/biotite/numba/python-dotenv/wandb was started with --no-deps to avoid a Torch/CUDA pull, but the download stalled before biotite completed; import still fails on missing biotite and no official checkpoint smoke was submitted. 2026-07-06 21:34 +08:00 continuation: retried lightweight no-deps dependency fill, but pip broke during biotite wheel download with BrokenPipeError; ProFam import still cannot pass and no checkpoint smoke was submitted. 2026-07-06 22:18 +08:00 continuation: ProFam remains blocked at lightweight dependency overlay; previous no-deps pip fill broke during biotite download and no checkpoint/API smoke was attempted this pass. 2026-07-07 00:14 +08:00 continuation: predownloaded and staged biotite stack, numpy 1.26.4, numba 0.66.0, and llvmlite 0.48.0 into profam_py312_site; import now reaches shared torch 2.4.1+cpu and ProFam's Llama/Transformers layer, but fails because the shared palm_py312_site Transformers generation module is incompatible (cannot import GenerationMixin). No official checkpoint smoke submitted. 2026-07-07 00:58 +08:00 continuation: staged clean ProFam dependency layer at /media/nik/seagate_nik/bio_server/envs/profam_deps_py312_site with Transformers 4.46.3, Accelerate 0.34.2, SciPy 1.13.1, Lightning/TorchMetrics, and reused shared torch 2.4.1+cpu. Downloaded official Hugging Face judewells/ProFam-1 checkpoint to /media/nik/seagate_nik/bio_server/models/profam/profam-1/checkpoints/last.ckpt (1,511,154,085 bytes). Direct Slurm job 482 at /media/nik/seagate_nik/bio_server/runs/profam/smoke_20260707_005748 loaded the real checkpoint, applied a ProFam tokenizer special-token compatibility initialization from package config, and produced ScoringResult score -0.06820218 with summary.json and results.csv. Promoted to Runner Adapter Pending; runner adapter must preserve the pinned deps layer and tokenizer special-token initialization. 2026-07-08T11:31:35+08:00: Production Ready promotion: backend runner API job b612dd9fb4f9 / Slurm job 538 completed cleanly using shared python312_slim.sif with profam_deps/profam/palm overlays and official ProFam-1 checkpoint. Preserved tokenizer special-token compatibility shim, captured model/tqdm chatter to artifacts, produced zero-byte slurm.out/slurm.err, and finite ScoringResult score -0.068202175.