# marian-nmt/marian

Fast Neural Machine Translation in C++

Repository: https://github.com/marian-nmt/marian
Canonical: https://ross.abutalabs.com/products/marian
Homepage: https://marian-nmt.github.io
Language: C++
License: NOASSERTION
License Family: other
Topics: neural-machine-translation, cuda, gpu, fast
Last push: 2023-08-25T22:31:09+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 3793, "days_push": 1104, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1468, forks 250 (observed 2026-08-28T04:04:48.756313+00:00)

## What it is
Marian is an efficient Neural Machine Translation framework written in pure C++ with minimal dependencies, supporting fast multi-GPU training and GPU/CPU translation of deep RNN and Transformer models. Developed primarily by the Microsoft Translator team, it is the engine behind production translation services and is MIT licensed.

## Use cases
- train a neural machine translation model on multiple GPUs
- run fast NMT inference on GPU or CPU from C++
- train a deep RNN or transformer translation model
- deploy a production machine translation engine
- translate text with a custom-trained NMT model
- reproduce research experiments with Nematus-compatible models

## When to choose
- you need fast, production-proven NMT training and decoding with minimal dependencies in C++
- you want multi-GPU training of state-of-the-art translation architectures like deep RNN and transformer
- you need a permissively licensed (MIT) toolkit with proven deployment at scale (Microsoft Translator)
- you want GPU-accelerated translation inference without a Python dependency stack

## When to avoid
- you need a Python-first ML framework or broad ecosystem integrations (PyTorch, Hugging Face, Fairseq)
- your task is general-purpose machine learning rather than translation
- you require cutting-edge large language model features or very active recent development, as the project's release cadence has slowed
- you only need lightweight CPU translation of small models, where simpler decoders may suffice

## Facets
- artifact type: framework
- maturity: maintenance
- function: deep-learning, machine-learning, nlp, gpu-computing
- domain: deep-learning, machine-learning, artificial-intelligence
- platform: windows, cpp
- tags: neural-machine-translation, nmt, cuda, transformer, rnn, translation, training, inference, microsoft-translator, c-plus-plus, natural-language-processing, linux, macos, gpu

## Member repositories
- marian-nmt/marian (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:48.756313+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 2026-08-30T04:34:57.118268+00:00, confidence not recorded.
  - readme: https://github.com/marian-nmt/marian (fetched 2026-08-28T04:04:48.756313+00:00, sha 1c7a7e75851c)
  - homepage: https://marian-nmt.github.io (fetched 2026-08-29T11:43:00.615459+00:00, sha 0a65b98f5568)
- Data as of 2026-08-30T08:39:29.467469+00:00.