SynCodonLM

5-30 minutesPaperCodeProduction Ready

Codon language model

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_ready_cpu_offline_lm
Storage: <5
Strategy: shared python312_slim.sif + tool ext3 overlay
Next: Validate user-submitted CDS/protein/species settings from the web UI and consider GPU/CUDA Torch container for larger optimization jobs.

SOP validation used upstream SynCodonLM revision 93a05baf09b929d94213ce49e4a1296bfba384c7 at /media/nik/seagate_nik/bio_server/tools/syncodonlm. Public non-gated HF model jheuschkel/SynCodonLM-V2 is cached locally at /media/nik/seagate_nik/bio_server/tools/syncodonlm/models/SynCodonLM-V2; no token needed. Worker Slurm job 354 and API Slurm job 357 completed on CPU inside shared python312_slim SIF with shared CPU Torch site-packages. Tiny benchmark CDS ATGTCCACCGGGCGGTGA and protein MSKG produced finite 768-d embedding and optimized CDS ATGAGCAAAGGT with length exactly 3x protein length. Artifacts include embedding_summary.json, optimized_cds.txt, results.csv, summary.json, stdout/stderr, and Slurm logs.