Prot T5 XL Embeddings

1-5 minutesPaperCodeNeed >20GB

Generate Prot T5 Embeddings

Readiness

Readiness: Need >20GB
Smoke: not_tested
Artifacts: not_applicable
Production: need_gt20gb_full_prot_t5_xl_model_cache
Storage: >20
Strategy: large Hugging Face model cache phase for full Prot-T5 XL; half-encoder requires explicit scoped approval
Next: Defer full Prot-T5 XL to >20GB model-storage phase, or explicitly approve half-encoder scope and run a one-sequence embedding smoke before adapter wiring.

Prot-T5 XL Hugging Face full repo Rostlab/prot_t5_xl_uniref50 is public/ungated but ~45.1 GB with multiple ~11.28 GB checkpoints. Preferred practical embeddings target is Rostlab/prot_t5_xl_half_uniref50-enc (~2.42 GB encoder-only). No weights downloaded and no Slurm smoke run; GPU/Singularity available. License needs final review on HF metadata, though ProtTrans repo is MIT. 2026-07-05 SOP triage: literal full Rostlab/prot_t5_xl_uniref50 target is public but approximately 45 GB and no model cache/runtime/runner exists. Reclassified from Smoke Pending to Need >20GB. Half-encoder is a separate smaller scoped substitution and was not assumed.