BioNeMo Framework provides versatile functionalities for developing, training, and deploying large language models. It 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.

Features

• Pre-Trained Models: Access to a variety of pre-trained models including Evo2, Geneformer, ESM-2nv, DNABERT, OpenFold, MolMIM, EquiDock, DiffDock, ESM-1nv, ProtT5, MegaMolBART, and DSMBind.
• Downstream Tasks: Supports tasks like protein secondary structure prediction, protein thermostability using Meltome Atlas, subcellular localization, retrosynthesis, and fine-tuning on FLIP/PhysChem datasets.
• Multi-Node Support: Enables high-performance distributed computing across multiple nodes.
• Multi-Arch Support: Compatible with various architectures, including x86 Linux systems.
• Docker Integration: Simplifies deployment with Docker images.
• Python Integration: Provides Python interfaces for Tensor Network operations.

Benefits

• Versatility: Supports a wide range of bio-based models and downstream tasks.
• High Performance: Optimized for supercomputing scale with multi-node and multi-GPU support.
• Ease of Use: Simplifies the development and deployment of large-scale models with minimal code.
• Scalability: Suitable for large-scale biomolecular simulations and model training.