OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Equiformer Pytorch

Equiformer Pytorch is a Python library implementing an SE(3)/E(3)-equivariant attention neural network in PyTorch. It provides model interfaces for feature tokens, 3D coordinates and masks, with optional edge features and sparse-neighbor adjacency inputs.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Equiformer Pytorch — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you need a PyTorch implementation to explore equivariant attention over coordinate-based inputs, particularly in molecular modeling or protein-structure research.

Who it is for

  • Researchers prototyping equivariant neural networks for molecular or protein-structure research.
  • PyTorch developers exploring attention models for structured scientific data.

Documented uses

  • Construct an equivariant attention model and pass in batched features, coordinates and masks.
  • Represent molecular relationships with edge features such as bond types.
  • Restrict attention to a supplied adjacency-defined neighborhood.
  • Explore reversible networks as a way to decouple memory use from model depth.
Before you choose

Requirements & limits

The public repository is MIT-licensed and documents installation with pip. The README describes the code as a work in progress.

  • The documented interface expects the developer to define model dimensions, degrees, heads and depth.
  • The repository shows distinct input patterns for basic coordinate features, typed edges and sparse adjacency matrices.
  • The README warns that its illustrative 48-layer reversible configuration is unstable at higher depths.
  • The project is a model implementation rather than a documented hosted inference API.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install and run a basic Equiformer example

Create a basic Equiformer model and pass token features, coordinates, and a mask through it.

Before you start

  • Python environment with PyTorch available.
  • The repository documents installing the package with pip.
  • The example uses CUDA tensors and calls .cuda().

Steps

  1. Install the package with the command below.

    Documented command
    pip install equiformer-pytorch

  2. Import torch and Equiformer, then construct a model with the example settings: num_tokens=24, dim=(4, 4, 2), dim_head=(4, 4, 4), heads=(2, 2, 2), num_degrees=3, and depth=4; the documented example also sets num_linear_attn_heads=0, attend_self=True, reduce_dim_out=True, and `l2_dist_attention=Fals

  3. Create token features with shape (1, 128), coordinates with shape (1, 128, 3), and a Boolean mask with shape (1, 128), as in the example, and move the model and tensors to CUDA.

  4. Call model(feats, coors, mask) and inspect the returned type0 and type1 outputs.

Expected outcome

The example returns an invariant `type0` output of shape `(1, 128)` and an equivariant `type1` output of shape `(1, 128, 3)`.

Guide limitations

  • The documented example requires CUDA and does not describe a CPU execution path.
  • The repository labels the implementation as work in progress.
  • The repository does not document pretrained weights or a protein-folding inference workflow.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
lucidrains/equiformer-pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

Equiformer v2 — A separate repository identified in the README as a follow-up implementation using a transformer-like framework and further equivariant-network scaling developments.