TorchSim/torch-sim

Torch-native, batchable, atomistic simulations.

What it solves

TorchSim is designed to accelerate atomistic simulations, specifically when using Machine Learning Interatomic Potentials (MLIPs). Traditional tools often struggle with the computational overhead of MLIPs, and TorchSim provides a way to run these simulations significantly faster by leveraging GPU acceleration and efficient memory management.

How it works

By rewriting the core primitives of atomistic simulation directly in PyTorch, the engine allows for automatic batching of multiple systems and native GPU support. This enables the simulation of many atomic systems simultaneously, which can lead to speedups of up to 100x compared to traditional frameworks like ASE.

Who it’s for

It is intended for researchers and scientists in materials science and chemistry who use MLIP models (such as MACE, Fairchem, SevenNet, and MatterSim) to perform molecular dynamics or structure relaxation.

Highlights

  • High-speed MLIP support: Native integration with a wide range of popular MLIP models.
  • Automatic Batching: Ability to run multiple simulations simultaneously to maximize GPU utilization.
  • Comprehensive Simulation Tools: Includes MD integration schemes (NVE, NVT, NPT), relaxation methods (FIRE, gradient descent), and Swap Monte Carlo algorithms.
  • Ecosystem Integration: Works with common tools like ASE, Pymatgen, and Phonopy.
  • Differentiable Simulation: Supports differentiable simulation and elastic properties calculation.

Related

  • Project
    mit-han-lab/torchquantum

    A PyTorch-based framework for simulating quantum computations on classical hardware, enabling the training of parameterized quantum circuits and hybrid classical-quantum models.

  • Project
    torchmd/torchmd-net

    A PyTorch-based library providing state-of-the-art neural network potentials for molecular simulations and the tools to train them.

  • Project
    IBM/aihwkit

    A Python toolkit for simulating and exploring in-memory computing devices to accelerate AI workloads by eliminating the von Neumann bottleneck.

  • Project
    pytorch/ao

    A PyTorch-native library for model optimization that provides quantization and sparsity tools to accelerate training and inference while reducing memory usage.

  • Project
    pytorch/torchtitan

    torchtitan is a PyTorch‑native library for large‑scale generative‑AI training. It bundles modern parallelism (FSDP2, tensor, pipeline, context), memory‑saving tricks (activation checkpointing, Float8/MXFP8 quantization), distributed checkpointing, and extensive logging. Ready‑to‑use for Llama 3.1 (8 B‑405 B) and extensible to new models, it runs on single‑node or multi‑node clusters via simple CLI/config files, and is released under a BSD‑3‑Clause license.