microsoft/Graphormer

Graphormer is a general-purpose deep learning backbone for molecular modeling.

What it solves

Graphormer is designed to accelerate research and application in AI for molecule science, specifically for tasks like material discovery and drug discovery. It provides a deep learning package for training custom models to model molecules.

How it works

Graphormer uses a Transformer-based architecture to perform graph representation learning for molecules. It integrates with the fairseq backbone and supports a wide range of data interfaces and datasets, including PyG, DGL, OGB, and OCP.

Who it’s for

Researchers and developers working in molecule modeling and AI for science, particularly those focusing on quantum prediction and molecular property prediction.

Highlights

  • Competition Winner: Won 1st place in the quantum prediction track of the Open Graph Benchmark Large-Scale Challenge (KDD CUP 2021) and the Open Catalyst Challenge.
  • Pre-trained Models: Offers pre-trained models on PCQM4M and PCQM4Mv2.
  • Broad Compatibility: Supports interfaces and datasets from PyG, DGL, OGB, and OCP.
  • Flexible Toolkit: Provides a general molecule modeling toolkit with a model zoo and flexible data interfaces.

Related

  • Project
    NovaSky-AI/SkyRL

    SkyRL is an open‑source, modular RL framework for training LLMs to perform long‑horizon, tool‑use tasks. It bundles a core training stack (with Tinker API support), an agent layer, a Gymnasium‑style environment library, and a cross‑platform backend, and provides recipes for cutting‑edge models and real‑world applications.

  • Project
    Farama-Foundation/Arcade-Learning-Environment

    A framework that allows researchers and hobbyists to develop and test AI agents for Atari 2600 games by providing a standardized interface to the Stella emulator.

  • Project
    natolambert/rlhf-book

    A comprehensive open-source guide and code repository for understanding Reinforcement Learning from Human Feedback (RLHF) and the post-training of language models.

  • Project
    mckinsey/causalnex

    A toolkit for causal reasoning and "what-if" analysis using Bayesian Networks to identify causal relationships and estimate the impact of interventions.