ProGen2 Finetuning

PaperCodeNeed >5GB

Finetune generative protein language model

Readiness

Readiness: Need >5GB
Smoke: not_tested
Artifacts: not_applicable
Production: need_gt5gb_cuda_pytorch_hf_runtime_pending
Storage: 5-20GB
Strategy: shared CUDA PyTorch/Transformers SIF + model/cache overlay; avoid python312_slim CPU path
Next: Build/reuse shared CUDA PyTorch + Transformers SIF/overlay, stage/cache hugohrban/progen2-small and tiny training data, run direct Slurm finetune/sampling smoke, then wire runner.

No obvious large model/database dependency identified; validate upstream docs before install. 2026-07-05 SOP batch recheck: backend inventory found no staged ProGen2 finetuning source/model/cache/SIF/artifacts/runner. Upstream uses Hugging Face progen2-small (~605 MB model) but finetuning requires CUDA PyTorch/Transformers plus dataset/cache staging, so python312_slim CPU is not the right production path. Reclassified from Smoke Pending to Need >5GB for shared GPU PLM runtime/materialization.