# laekov/fastmoe

A fast MoE impl for PyTorch

Repository: https://github.com/laekov/fastmoe
Canonical: https://ross.abutalabs.com/products/fastmoe
Homepage: https://fastmoe.ai
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2025-02-10T06:04:33+00:00

## Health v2 (maintenance only)
Score: 26/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 6, release rhythm 8, longevity 100
- inputs: {"age_days": 2047, "days_push": 569, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1859, forks 206 (observed 2026-08-28T04:05:45.262722+00:00)

## What it is
FastMoE is a PyTorch library providing efficient Mixture of Experts (MoE) layers with custom C/CUDA operators. It supports distributed expert parallelism and one-key conversion of Transformer models (e.g., Megatron-LM) into MoE models.

## Use cases
- scale up transformer MLP layers to mixture of experts
- train MoE models in PyTorch
- run expert parallel training across GPUs
- convert Megatron-LM models to MoE
- implement sparse expert routing in deep learning models

## When to avoid
- you need MoE on CPU or without CUDA
- you use a framework other than PyTorch
- you want a maintained high-level API without custom CUDA builds

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-training, gpu-computing
- domain: deep-learning, large-language-models, machine-learning
- platform: python
- tags: mixture-of-experts, pytorch, distributed-training, transformer, cuda, gpu, linux, docker

## Member repositories
- laekov/fastmoe (main) score 26

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:45.262722+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:16:01.013109+00:00, confidence not recorded.
  - readme: https://github.com/laekov/fastmoe (fetched 2026-08-28T04:05:45.262722+00:00, sha 6dee94d44deb)
- Data as of 2026-08-30T08:39:29.467469+00:00.