CodonFM Fitness

License: CodonFM (Encodon) uses Apache-2.0 for code and Custom (NVIDIA Open Model License) for model weights. Please refer to the code license and model weights license for full terms.

This constraint is open source. Any third-party models, product names, or trademarks referenced are the property of their respective owners, and Proto is not affiliated with them.

Tools Used Tools Used Source Source Cite Cite

Go to Tool Page

evo-design/proto-language

View source

@articlecodonfm2025

@article{codonfm2025,
  title={Learning the language of codon translation with CodonFM},
  author={Darabi, Sajad and Cao, Fan and Naghipourfar, Mohsen and Rabi, Sara and Sethia, Ankit and Gion, Kyle and Grewal, Jasleen and Cohen, Jonathan and Greenleaf, William and Goodarzi, Hani and Sundaram, Laksshman},
  year={2025},
  publisher={NVIDIA},
  url={https://research.nvidia.com/labs/dbr/assets/data/manuscripts/nv-codonfm-preprint.pdf}
}

Copy citation

Constraint contributors

Score proposals by Encodon’s upstream non-padding-token mean log-likelihood. This follows predict_fitness exactly: the mean includes visible codons and the CLS/SEP special tokens, while excluding padding.

API Reference

ConfigCodonFMFitnessConstraintConfig Source

Configuration for the CodonFM coding-sequence fitness constraint.

enum

default:"encodon_80m"

Encodon checkpoint: encodon_80m | encodon_600m | encodon_1b | encodon_1b_cdwt.Options: encodon_80m, encodon_600m, encodon_1b, encodon_1b_cdwt

enum

default:"max"

Whether to maximize (‘max’) or minimize (‘min’) the coding-sequence fitness.Options: max, min

number

default:"-2.0"

Upstream non-padding-token mean log-likelihood mapped to energy 0.5 (tunable; negative).

number

default:"1.0"

Positive float32-safe width of the sigmoid transform; larger is gentler.

string

default:"cuda"

CUDA device used for CodonFM inference; the upstream xFormers path is GPU-only.

integer

default:"1"

Number of sequences scored per CodonFM GPU batch.

ReturnsConstraintOutput

One result per proposal. score is in [0.0, 1.0] (lower is better in the requested direction); metadata carries the raw fitness.

Usage

python

from proto_language.core import Constraint
from proto_language.constraint import codonfm_fitness_constraint, CodonFMFitnessConstraintConfig

constraint = Constraint(
    inputs=[segment],
    function=codonfm_fitness_constraint,
    function_config=CodonFMFitnessConstraintConfig(
        # Configure parameters here
    ),
)

scores = constraint.evaluate()

Metadata

Property	Value
Key	`codonfm-fitness`
Function	`codonfm_fitness_constraint`
Category	`sequence_scoring`
Mode	`discrete`
Uses GPU	`True`
Supported Types	`dna`, `rna`

Language

Language Concepts

Language Guides

Constraints

Generators

Optimizers

CodonFM Fitness

API Reference

Usage

Metadata

​API Reference

​Usage

​Metadata

API Reference

Usage

Metadata