nvidia/NV-GenMol-89M-v2

Model Overview

The code for using the GenMol model checkpoint is available in the official Github repository.

Description:

GenMol is a masked diffusion model¹ trained on molecular Sequential Attachment-based Fragment Embedding (SAFE) representations² for fragment-based molecule generation, which can serve as a generalist model for various drug discovery tasks, including De Novo generation​, linker design​, motif extension​, scaffold decoration/morphing​, hit generation​, and lead optimization.

This model is ready for commercial use.

License/Terms of Use:

Governing Terms: Use of this model is governed by the NVIDIA Open Model License. GenMol source code is licensed under Apache 2.0. By using GenMol, you accept the terms and conditions of this license.

Deployment Geography: Global

Use Case: GenMol is a flexible generative AI tool for small molecule design in drug discovery. Its primary use cases include: Generating novel molecules: GenMol can create new, valid molecules from scratch (de novo design). Optimizing existing molecules: It can be used to modify and improve current molecules through inference tasks like scaffold decoration, motif extension, superstructure generation, and linker design. Property optimization: When paired with external scoring models or methods, GenMol aids in designing molecules with specific desired properties, which is a critical step in developing effective drugs.

Release Date: Github 07/22/2025 via https://github.com/NVIDIA-Digital-Bio/genmol

References:

@misc{sahoo2024simpleeffectivemaskeddiffusion,  
      title={Simple and Effective Masked Diffusion Language Models},   
      author={Subham Sekhar Sahoo and Marianne Arriola and Yair Schiff and Aaron Gokaslan and Edgar Marroquin and Justin T Chiu and Alexander Rush and Volodymyr Kuleshov},  
      year={2024},  
      eprint={2406.07524},  
      archivePrefix={arXiv},  
      primaryClass={cs.CL},  
      url={https://arxiv.org/abs/2406.07524},   
}

@misc{noutahi2023gottasafenewframework,  
      title={Gotta be SAFE: A New Framework for Molecular Design},   
      author={Emmanuel Noutahi and Cristian Gabellini and Michael Craig and Jonathan S. C Lim and Prudencio Tossou},  
      year={2023},  
      eprint={2310.10773},  
      archivePrefix={arXiv},  
      primaryClass={cs.LG},  
      url={https://arxiv.org/abs/2310.10773},   
}

Model Architecture:

Architecture Type: Transformer

Network Architecture: BERT

Input:

Input Type(s): Text (Molecular Sequence), Number (Molecules to generate, SoftMax temperature scaling factor, randomness factor, diffusion step-size), Enumeration (Scoring method), Binary (Showing unique molecules only)

Input Format(s): Text: String (Sequential Attachment-based Fragment Embedding (SAFE)); Number: Integer, FP32; Enumeration: String (QED, LogP); Binary: Boolean

Input Parameters: 1D

Other Properties Related to Input: Maximum input length is 512 tokens.

Output:

Output Type(s): Text (List of molecule sequences), Number (List of scores)

Output Format: Text: Array of string (Sequential Attachment-based Fragment Embedding (SAFE)); Number: Array of FP32 (Scores)

Output Parameters: 2D

Other Properties Related to Output: Maximum output length is 512 tokens.

Software Integration:

Runtime Engine(s):
PyTorch >= 2.5.1

Supported Hardware Microarchitecture Compatibility:

NVIDIA Ampere

NVIDIA Ada Lovelace

NVIDIA Hopper

NVIDIA Grace Hopper

[Preferred/Supported] Operating System(s):

Linux

Model Version(s):

GenMol v2.0
GenMol v1.0

Training & Evaluation Dataset:

Training and Testing Dataset:

Link: SAFE-GPT GitHub, HuggingFace,

Data Collection Method by dataset: Automated

Labeling Method by dataset: Automated

Properties: 1.1B SAFE strings consist of various molecule types (drug-like compounds, peptides, multi-fragment molecules, polymers, reagents and non-small molecules).

Dataset License(s): CC-BY-4.0

Evaluation Dataset:

Link: SAFE-DRUGS GitHub, HuggingFace

Data Collection Method by dataset: Not Applicable

Labeling Method by dataset: Not Applicable

Properties: SAFE-DRUGS consists of 26 known therapeutic drugs.

Dataset License(s): CC-BY-4.0

Inference:

Engine: PyTorch

Test Hardware: A6000, A100, L40, L40S, H100

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Users are responsible for ensuring the physical properties of model-generated molecules are appropriately evaluated and comply with applicable safety regulations and ethical standards.

For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards [Insert Link to Model Card++ subcards here].

Please report security vulnerabilities or NVIDIA AI Concerns here.

Subcards

Bias

Field	Response
Participation considerations from adversely impacted groups (protected classes) in model design and testing	Not Applicable
Measures taken to mitigate against unwanted bias	Not Applicable

Explainability

Field	Response
Intended Application(s) & Domain(s)	Molecular drug discovery and design
Model Type	Molecular sequence generation
Intended Users	Developers in the academic or pharmaceutical industries who build artificial intelligence applications to perform property guided molecule optimization and novel molecule generation.
Output	Text (molecule sequence)
Describe how the model works	From the input of a molecular sequence (SAFE format) with masks (masking tokens), the neural network model (Transformer & BERT architecture) will one-by-one predict the best characters (valid tokens in the chemistry vocabulary) to replace the masking tokens, until all masks are replaced, which is an unmasking process by discrete diffusion.
Name the adversely impacted groups this has been tested to deliver comparable outcomes regardless of	Not Applicable
Technical Limitations	Model may not perform well on sequences that are highly divergent from the ZINC-15 dataset.
Verified to have met prescribed quality standards?	Yes
Performance Metrics	Validity, Uniqueness, Diversity, Central Distance, Qualified Ratio.
Potential Known Risks	The model may produce molecules that are difficult or impossible in synthesis.
Licensing & Terms of Use	Governing Terms: Use of this model is governed by the NVIDIA Open Model License. GenMol source code is licensed under Apache 2.0.

Privacy

Field	Response
Generatable or reverse engineerable personal data?	No
Personal data used to create this model?	No
How often is dataset reviewed?	Before Every Release
Is there provenance for all datasets used in training?	Yes
Does data labeling (annotation, metadata) comply with privacy laws?	Yes
Applicable Privacy Policy	NVIDIA Privacy Policy

Safety

Field	Response
Model Application(s)	Molecular drug discovery and design
Describe life critical application (if present)	Experimental drug discovery and medicine. Should not be used for life-critical use cases per NVIDIA Software License Agreement.
Use Case Restrictions	Abide by NVIDIA Open Model License. GenMol source code is licensed under Apache 2.0.
Model and Dataset Restrictions	The Principle of least privilege (PoLP) is applied limiting access for dataset generation and model development. Restrictions enforce dataset access during training, and dataset license constraints adhered to.