Readiness
- Readiness
- Need >20GB
- Smoke
- passed
- Artifacts
- passed
- Production
- need_gt20gb_mber_runtime_weights_gpu_benchmark_pending
- Storage
- >20GB
- Strategy
- large shared SIF/database/model bundle phase
- Next
- Budget/provision >20GB runtime space, build mBER Docker/SIF or shared runtime plus required AF2/NanoBodyBuilder2/ESM2 weights, run the smallest VHH benchmark; if the 8GB RTX 5050 OOMs, reclassify to Production Blocked with larger-VRAM requirement.
No obvious large model/database dependency identified; validate upstream docs before install. Worker evidence 2026-07-01: source https://github.com/manifoldbio/mber-open revision dbabbf8d09a4b35dd20dc11495c611209a51a561 staged at /media/nik/seagate_nik/bio_server/tools/mber/source, current footprint 73MB. Slurm probes 279/280 on gpu partition: import mber succeeds with PYTHONPATH=src:protocols/src, mber_protocols discoverable, but mber-vhh CLI fails before benchmark with ModuleNotFoundError colabdesign. Missing runtime modules include torch, jax, colabdesign, ImmuneBuilder, anarci, transformers. Upstream weights AF2/NanoBodyBuilder2/ESM2 are about 9GB; optional ESMFold about 16GB. Backend GPU RTX 5050 Laptop has 8151MiB VRAM, below upstream 32GB guidance. Seagate venv symlink issue noted; use micromamba on /home, pip target, SIF, or overlay. 2026-07-07 SOP reclassification: agent audit confirmed this is not a simple production-blocked runner issue. Required upstream weights are about 9GB before optional ESMFold (~16GB), and the CUDA/JAX/PyTorch/ColabDesign/ImmuneBuilder/ANARCI/Transformers stack is missing; moved to Need >5GB shared runtime/assets lane. 2026-07-17T01:18:00+08:00 SOP Need >5GB next-10 audit: remains Need >5GB. Runner audit is not_configured and backend inventory found no mBER source/runtime/assets. Existing notes require CUDA/JAX/PyTorch/ColabDesign/ImmuneBuilder/ANARCI/Transformers plus ~9GB required weights; optional ESMFold may push scope >20GB but required path remains 5-20GB. 2026-07-19T22:38:00+08:00 SOP Need >5GB active install batch: Official mBER source staged at /media/nik/seagate_nik/bio_server/repos/mber-open commit dbabbf8 (6.0MB). Upstream install uses conda environment.yml, pip install -e protocols, and download_weights.sh. Required weights are about 9GB for AlphaFold2/NanoBodyBuilder2/ESM2 before optional ESMFold; README recommends a 32GB GPU, with small VHH targets maybe fitting below 16GB. Current backend has RTX 5050 8GB, so do not promote until a tiny benchmark proves it fits or a larger GPU route is configured. 2026-07-19T22:45:00+08:00 SOP Need >5GB active install batch reclassification: mBER source is staged at /media/nik/seagate_nik/bio_server/repos/mber-open commit dbabbf8, but practical production deployment belongs in Need >20GB. Required public weights are about 9GB, while upstream Docker/runtime guidance puts the container plus required weights around 31GB before optional ESMFold; optional ESMFold would add about 16GB more. Current backend free space was 35GB and GPU is 8GB RTX 5050, below upstream 32GB recommendation, so production validation should be budgeted as a >20GB install and then benchmarked on the smallest VHH case or larger GPU route.