CodonFM Codon Language Model

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 generator 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

Generator contributors

Coding-sequence mutation generator using the CodonFM (Encodon) codon language model. Refines existing coding sequences by masking a subset of codon positions and resampling them from Encodon’s per-codon distribution. Sequence length is preserved. The generator category is "mutation".

API Reference

ConfigCodonFMGeneratorConfig Source

Configuration object for CodonFMGenerator.CodonFM (Encodon) is a codon-level masked language model. As a Proto Language generator it refines an existing coding sequence by masking a subset of codons and resampling them from the model — a mutation-category generator, so the segment must already carry a sequence (directly or from a prior optimizer stage).

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

RandomMaskingStrategy

Which codons to resample (counted in codons); fixed_positions pins codons that must survive.

number

default:"1.0"

Sampling temperature. Below 1 sharpens toward the likely codon; above 1 adds diversity.

string

default:"cuda"

GPU device to run CodonFM on (e.g. ‘cuda’ or ‘cuda:0’).

integer

default:"1"

Number of same-length sequences to process simultaneously on GPU.

array

Single-letter codes no resampled codon may encode (e.g. [‘C’] to forbid cysteine).

Usage

python

>>> from proto_language.generator import CodonFMGenerator, CodonFMGeneratorConfig
>>> from proto_language.core import Segment
>>> from proto_tools.transforms.masking import RandomMaskingStrategy
>>> gen = CodonFMGenerator(CodonFMGeneratorConfig(masking_strategy=RandomMaskingStrategy(num_mutations=3)))
>>> segment = Segment(sequence="ATGGTGAGCAAGGGC", sequence_type="dna")  # 5 codons
>>> gen.assign(segment)
>>> gen.sample()  # resamples 3 randomly masked codons

Metadata

Property	Value
Key	`codonfm`
Class	`CodonFMGenerator`
Category	`mutation`
Input Type	`starting_sequence`
Uses GPU	`True`
Supported Sequence Types	`dna`, `rna`
Allows Empty Start	`False`

Language

Language Concepts

Language Guides

Constraints

Generators

Optimizers

CodonFM Codon Language Model

API Reference

Usage

Metadata

​API Reference

​Usage

​Metadata

API Reference

Usage

Metadata