BioNeMo Framework

The BioNeMo Framework container has been archived and is no longer being maintained by NVIDIA. Please check out the Bionemo Recipes Github to see our NVIDIA-optimized reference architectures for biological foundation models. For any open questions, please submit a Github issue on the Github site.

What is BioNeMo Framework?

BioNeMo Framework provides versatile functionalities for developing, training and deploying large language models. BioNeMo allows users to build biomolecular models by providing access to pre-trained models, creating workflows to fit downstream task models from embeddings, and generating biomolecules that meet user-specified criteria based on the fit model. Built for supercomputing scale, the framework allows developers to easily configure and deploy distributed multi-node jobs with minimal code.

The source code for this container is available here: https://github.com/NVIDIA/bionemo-framework

This container contains prebuilt source in a ready-to-run environment.

Between BioNeMo 1.x and 2.x versions, there are implementations of the following models :

• Evo2 (v2.x)
• Geneformer (v2.x, v1.x)
• ESM-2nv (v2.x, v1.x)
• DNABERT (v1.x)
• OpenFold (v1.x)
• MolMIM (v1.x)
• EquiDock (v1.x)
• DiffDock (v1.x)
• ESM-1nv (v1.x)
• ProtT5 (v1.x)
• MegaMolBART (v1.x)
• DSMBind (v1.x)

This container also supports downstream tasks like protein secondary structure prediction, protein thermostability using Meltome Atlas, subcellular localization, retrosynthesis and fine-tuning on FLIP/PhysChem datasets.

Getting Started With BioNeMo Framework

For information about how to get started with BioNeMo refer to the documentation.

Compatible Infrastructure Software Versions

BioNeMo is only supported on x86 Linux systems. If you are not on such a system, you must use the project's Docker images to develop and execute BioNeMo code.

System Requirements:

• Docker (with GPU support, docker engine >= 19.03).
• Ubuntu 20.04 or Higher
• NVIDIA GPU, if you intend to do model training. BioNeMo is compatible with most NVIDIA GPUs, with some incompatibility:
  Tested GPUs:
  • DGX-B200, H100, A100, V100
  • L40, L40s
  • GH200 (with ARM processor)
  • GeForce RTX 5090, 4090, 2080 Ti
  • RTX A6000, A8000
• bfloat16 precision requires an Ampere generation GPU or higher.
• fp8 precision requires Hopper generation GPUs.

NVIDIA Licensing Portal

Go to the NVIDIA Licensing Portal to manage your software licenses.

License

This container is licensed under the NVIDIA AI Product Agreement. By pulling and using this container, you accept the terms and conditions of this license.