- name
- codonfm-score
- description
- Validate, prepare, or run public CodonFM Encodon masked-codon variant scoring and review compatibility of its scoring workflows. Use only when the user explicitly requests CodonFM or Encodon, or that context is already established in the conversation. Do not select this skill for a generic variant-scoring request without that context; ask for the variant and intended analysis first.
- metadata
- author: "NVIDIA BioNeMo <[email protected]>"
Score variants with public Encodon
Run general masked-codon mutation_prediction only. This produces a research
signal, not a clinical diagnosis or an expression-direction prediction.
Instructions
First confirm CodonFM or Encodon context in the user's request or established conversation. If that context is missing, ask for any missing variant details and the intended analysis before choosing a model or inspecting model-specific files. The presence of this skill or source files alone does not establish the user's intent.
For source reviews and command preparation, inspect the supplied source and
metadata without installing the ML runtime. Use an available Python 3
interpreter with standard-library zipfile, json, and csv; do not assume
the python alias or unzip exists. Read archive members directly with
ZipFile.namelist() and ZipFile.read() where possible. If extraction is
needed, use a fresh directory from tempfile.mkdtemp() or mktemp -d and
preserve existing checkouts and scratch directories. Check whether rg is
available; use grep or Python if it is absent. Read the source sections
needed for the requested command or compatibility question.
Check whether the request is executable in public v1 before installing or downloading anything. For synonymous-codon aggregation or Decodon, inspect the parser and model configuration, explain the missing feature, and finish. Do not implement the missing workflow, search private code, or keep retrying unsupported commands.
Resolve the variant CSV, checkpoint, and output directory from the request and available files. Validate inputs before inference. Execution requires the project's ML dependencies and a compatible NVIDIA GPU. If a required resource is unavailable, return the validated inputs where possible and a command with the missing prerequisite identified. When scoring is requested and resources are ready, execute and verify the score arrays. A request for preparation ends with the inputs and command. If variants are missing, report the required schema; do not invent variants or silently switch to a public dataset.
Default to the public 80M checkpoint for demonstrations:
nvidia/NV-CodonFM-Encodon-80M-v1, revision
399ca9fe17b57941a7bebc6788033919b417413c, file
NV-CodonFM-Encodon-80M-v1.safetensors and sibling .
Reuse an existing checkpoint or download it when needed for the requested work.
Preserve an explicitly requested model size.