# OpenNMT/CTranslate2

Fast inference engine for Transformer models

Repository: https://github.com/OpenNMT/CTranslate2
Canonical: https://ross.abutalabs.com/products/ctranslate2
Homepage: https://opennmt.net/CTranslate2
Language: C++
License: MIT
License Family: permissive
Topics: neural-machine-translation, cpp, mkl, quantization, cuda, thrust, opennmt, deep-neural-networks, openmp, onednn, intrinsics, avx2, avx, parallel-computing, gemm, neon, transformer-models, machine-translation, deep-learning, inference
Last push: 2026-08-16T16:34:36+00:00

## Health v2 (maintenance only)
Score: 96/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 98, release rhythm 91, longevity 100
- inputs: {"age_days": 2536, "days_push": 17, "days_rel": 61, "gap_med": 29.5, "n_releases_24m": 11}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4644, forks 520 (observed 2026-08-28T04:08:55.498223+00:00)

## What it is
CTranslate2 is a C++ and Python library for fast, memory-efficient inference of Transformer models on CPU and GPU. It uses quantization, layer fusion, and other optimizations to accelerate encoder-decoder, decoder-only, and encoder-only models like NLLB, Llama, T5, and Whisper.

## Use cases
- run machine translation models fast in production
- serve whisper speech recognition with low memory
- run llama or mistral inference on cpu
- quantize transformer models to int8 or fp16
- embed transformer inference in a c++ application
- convert huggingface transformers models to an optimized format
- speed up text generation compared to pytorch

## When to choose
- you need fast transformer inference in production, especially on cpu
- you want to reduce model memory footprint via quantization
- you need to embed models in a c++ application with minimal dependencies
- you require fine control over threading and memory usage

## When to avoid
- you want to train or fine-tune custom architectures
- your model architecture is not among the supported types
- you just need a quick prototype where pytorch suffices

## Facets
- artifact type: library
- maturity: stable
- function: llm-inference, machine-learning, deep-learning, speech-recognition, nlp, sdk
- domain: machine-learning, deep-learning, large-language-models, speech-processing, performance
- platform: windows, python, cpp, cross-platform
- tags: transformer-inference, quantization, model-conversion, cpu-optimization, int8, beam-search, opennmt, whisper, llama, natural-language-processing, linux, macos, gpu, docker

## Member repositories
- OpenNMT/CTranslate2 (main) score 96

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:55.498223+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 2026-08-29T18:19:32.614932+00:00, confidence not recorded.
  - readme: https://github.com/OpenNMT/CTranslate2 (fetched 2026-08-28T04:08:55.498223+00:00, sha 974968e211f1)
  - homepage: https://opennmt.net/CTranslate2 (fetched 2026-08-29T09:04:10.390941+00:00, sha bf8fb38c69d4)
  - site_page: https://opennmt.net/CTranslate2/quickstart.html (fetched 2026-08-29T09:04:10.399753+00:00, sha 4d71e5b67b1c)
  - site_page: https://opennmt.net/CTranslate2/installation.html (fetched 2026-08-29T09:04:10.401593+00:00, sha 591483a3d8b0)
  - site_page: https://opennmt.net/CTranslate2/decoding.html (fetched 2026-08-29T09:04:10.403318+00:00, sha efcf95c94e93)
  - site_page: https://opennmt.net/CTranslate2/faq.html (fetched 2026-08-29T09:04:10.404955+00:00, sha d878e39b3e51)
- Data as of 2026-08-30T08:39:29.467469+00:00.