---
id: fair-esm
version: "2.0.0"
license: MIT
license_treatment: permissive
maintenance: abandoned
---

# fair-esm — Evolutionary Scale Modeling (esm): Pretrained language models for proteins. From Facebook AI Research.

License: permissive · Maintenance: abandoned · Downloads: 170.2K/mo

## What it is and what it does

Fair-esm is a collection of pre-trained transformer models for protein sequence analysis from Facebook AI Research. It includes ESM-2 (a general-purpose protein language model available in multiple sizes), ESMFold (end-to-end structure prediction), ESM-1v (specialized for variant effect prediction), MSA Transformer (for multiple sequence alignment analysis), and ESM-IF1 (for inverse folding and sequence design). The models are trained on large-scale protein sequence datasets and can generate embeddings, predict 3D structures from sequences, assess the functional impact of mutations, and design new sequences for given protein structures.

The package is designed for researchers and practitioners working with protein analysis. It requires PyTorch (not declared as a dependency) and downloads pre-trained weights on first use. The repository is no longer actively maintained—the last commit was February 2024 and the latest PyPI release is from November 2022—so users should expect no updates, bug fixes, or support for new PyTorch versions.

Use it for:

- Generate protein embeddings for downstream machine learning tasks like function prediction or homology detection
- Predict 3D protein structures directly from amino acid sequences without multiple sequence alignments
- Assess how amino acid substitutions affect protein function using zero-shot variant effect prediction
- Design new protein sequences for a given 3D structure using inverse folding
- Extract contact predictions and structural features from protein sequences for comparative analysis

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Provides pre-trained transformer protein language models (ESM-2, ESMFold, ESM-1v, MSA Transformer, ESM-IF1) for protein structure prediction, embedding generation, variant effect prediction, and inverse folding directly from sequence.

Yes, with conditions. The models are scientifically rigorous and widely cited, and the MIT license is permissive. However, the repository is archived and unmaintained since February 2024, so there will be no bug fixes or compatibility updates. Install only if you can tolerate a frozen codebase and are willing to manage PyTorch compatibility yourself. For active development or production systems requiring ongoing support, consider whether a maintained fork or alternative is available.

## Install

pip install fair-esm
uv add fair-esm
poetry add fair-esm

## Installing fair-esm

Before you install: Installation is straightforward with no runtime dependencies. However, the repository is archived and unmaintained since February 2024, with the latest PyPI release from November 2022. PyTorch is a prerequisite but not declared as a package dependency.

License in practice: MIT license permits commercial and private use with minimal restrictions, making it suitable for research and production applications without licensing friction.

Quickstart:

pip install fair-esm

import esm
model, alphabet = esm.pretrained.esm2_t36_3B_UR50D()
# Use model for inference on protein sequences

PyTorch must be installed separately; the package does not declare it as a dependency. Pre-trained model weights are downloaded on first use and require sufficient disk space.

Verify before relying:

- Exact disk space and memory requirements for different model variants (8M, 35M, 150M, 700M, 3B, 15B parameters)
- Whether PyTorch dependency is intentionally omitted or an oversight in packaging
- Current state of pre-trained weight URLs and download reliability given repository archival

## Package facts

- License: MIT (permissive)
- Python support: unspecified
- Install friction: low
- Maintenance: abandoned
- Downloads: 170.2K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

protein language model, protein structure prediction, ESM transformer, protein embeddings, variant effect prediction, inverse protein folding, sequence to structure, protein-science, deep-learning, structure-prediction

[View on SkillFed](https://skillfed.io/packages/fair-esm) · [View on PyPI](https://pypi.org/project/fair-esm/)