00mjk / CTranslate2

Fast inference engine for Transformer models

Home Page:https://opennmt.net

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

CI PyPI version Gitter Forum


CTranslate2 is a fast and full-featured inference engine for Transformer models. It aims to provide comprehensive inference features and be the most efficient and cost-effective solution to deploy standard neural machine translation systems on CPU and GPU. It currently supports Transformer models trained with:

The project is production-oriented and comes with backward compatibility guarantees, but it also includes experimental features related to model compression and inference acceleration.

Table of contents

  1. Key features
  2. Quickstart
  3. Installation
  4. Converting models
  5. Translating
  6. Environment variables
  7. Building
  8. Testing
  9. Benchmarks
  10. Frequently asked questions

Key features

  • Fast and efficient execution on CPU and GPU
    The execution is significantly faster and requires less resources than general-purpose deep learning frameworks on supported models and tasks thanks to many advanced optimizations: padding removal, batch reordering, in-place operations, caching mechanism, etc.
  • Quantization and reduced precision
    The model serialization and computation support weights with reduced precision: 16-bit floating points (FP16), 16-bit integers (INT16), and 8-bit integers (INT8).
  • Multiple CPU architectures support
    The project supports x86-64 and AArch64 processors and integrates multiple backends that are optimized for these platforms: Intel MKL, oneDNN, OpenBLAS, Ruy, and Apple Accelerate.
  • Automatic CPU detection and code dispatch
    One binary can include multiple backends (e.g. Intel MKL and oneDNN) and instruction set architectures (e.g. AVX, AVX2) that are automatically selected at runtime based on the CPU information.
  • Parallel and asynchronous translations
    Translations can be run efficiently in parallel and asynchronously using multiple GPUs or CPU cores.
  • Dynamic memory usage
    The memory usage changes dynamically depending on the request size while still meeting performance requirements thanks to caching allocators on both CPU and GPU.
  • Lightweight on disk
    Quantization can make the models 4 times smaller on disk with minimal accuracy loss. A full featured Docker image supporting GPU and CPU requires less than 500MB (with CUDA 10.0).
  • Simple integration
    The project has few dependencies and exposes translation APIs in Python and C++ to cover most integration needs.
  • Interactive decoding
    Advanced decoding features allow autocompleting a partial translation and returning alternatives at a specific location in the translation.

Some of these features are difficult to achieve with standard deep learning frameworks and are the motivation for this project.

Supported decoding options

The translation API supports several decoding options:

  • decoding with greedy or beam search
  • random sampling from the output distribution
  • translating with a known target prefix
  • returning alternatives at a specific location in the target
  • constraining the decoding length
  • returning multiple translation hypotheses
  • returning attention vectors
  • approximating the generation using a pre-compiled vocabulary map
  • replacing unknown target tokens by source tokens with the highest attention
  • biasing translations towards a given prefix (see section 4.2 in Arivazhagan et al. 2020)
  • scoring existing translations

See the Decoding documentation for examples.


The steps below assume a Linux OS and a Python installation (3.6 or above).

1. Install the Python package:

pip install --upgrade pip
pip install ctranslate2

2. Convert a Transformer model trained with OpenNMT-py, OpenNMT-tf, or Fairseq:

a. OpenNMT-py

pip install OpenNMT-py

wget https://s3.amazonaws.com/opennmt-models/transformer-ende-wmt-pyOnmt.tar.gz
tar xf transformer-ende-wmt-pyOnmt.tar.gz

ct2-opennmt-py-converter --model_path averaged-10-epoch.pt --output_dir ende_ctranslate2

b. OpenNMT-tf

pip install OpenNMT-tf

wget https://s3.amazonaws.com/opennmt-models/averaged-ende-ckpt500k-v2.tar.gz
tar xf averaged-ende-ckpt500k-v2.tar.gz

ct2-opennmt-tf-converter --model_path averaged-ende-ckpt500k-v2 --output_dir ende_ctranslate2 \
    --src_vocab averaged-ende-ckpt500k-v2/wmtende.vocab \
    --tgt_vocab averaged-ende-ckpt500k-v2/wmtende.vocab \
    --model_type TransformerBase

c. Fairseq

pip install fairseq

wget https://dl.fbaipublicfiles.com/fairseq/models/wmt16.en-de.joined-dict.transformer.tar.bz2
tar xf wmt16.en-de.joined-dict.transformer.tar.bz2

ct2-fairseq-converter --model_path wmt16.en-de.joined-dict.transformer/model.pt \
    --data_dir wmt16.en-de.joined-dict.transformer \
    --output_dir ende_ctranslate2

3. Translate tokenized inputs with the Python API:

import ctranslate2

translator = ctranslate2.Translator("ende_ctranslate2/", device="cpu")

# The OpenNMT-py and OpenNMT-tf models use a SentencePiece tokenization:
translator.translate_batch([["▁H", "ello", "▁world", "!"]])

# The Fairseq model uses a BPE tokenization:
translator.translate_batch([["H@@", "ello", "world@@", "!"]])


Python package

Python packages are published on PyPI:

pip install ctranslate2


  • OS: Linux (x86-64, AArch64), macOS (x86-64), Windows (x86-64)
  • Python version: >= 3.6
  • pip version: >= 19.3
  • (optional) CUDA version: 11.x
  • (optional) GPU driver version: >= 450.80.02

Docker images

The opennmt/ctranslate2 repository contains images with prebuilt libraries and clients:

docker pull opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2

The library is installed in /opt/ctranslate2 and a Python package is installed on the system.


  • Docker
  • (optional) GPU driver version: >= 450.80.02

Manual compilation

See Building.

Converting models

The core CTranslate2 implementation is framework agnostic. The framework specific logic is moved to a conversion step that serializes trained models into a simple binary format.

The following frameworks and models are currently supported:

OpenNMT-tf OpenNMT-py Fairseq
Transformer (Vaswani et al. 2017)
+ relative position representations (Shaw et al. 2018)

If you are using a model that is not listed above, consider opening an issue to discuss future integration.

The Python package includes a conversion API and conversion scripts:

  • ct2-opennmt-py-converter
  • ct2-opennmt-tf-converter
  • ct2-fairseq-converter

The conversion should be run in the same environment as the selected training framework.

Integrated model conversion

Models can also be converted directly from the supported training frameworks. See their documentation:

Quantization and reduced precision

The converters support reducing the weights precision to save on space and possibly accelerate the model execution. See the Quantization documentation.

Adding converters

Each converter should populate a model specification with trained weights coming from an existing model. The model specification declares the variable names and layout expected by the CTranslate2 core engine.

See the existing converters implementation which could be used as a template.


The examples use the English-German OpenNMT model converted in the Quickstart. This model requires a SentencePiece tokenization.

With the translation client

echo "▁H ello ▁world !" | docker run -i --rm -v $PWD:/data \
    opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2 --model /data/ende_ctranslate2 --device cpu

To translate on GPU, use docker run --gpus all and set the option --device cuda.

See docker run --rm opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2 --help for additional options.

With the Python API

import ctranslate2
translator = ctranslate2.Translator("ende_ctranslate2/", device="cpu")
translator.translate_batch([["▁H", "ello", "▁world", "!"]])

See the Python reference for more advanced usages.

With the C++ API

#include <iostream>
#include <ctranslate2/translator_pool.h>

int main() {
  const size_t num_translators = 1;
  const size_t num_threads_per_translator = 4;
  ctranslate2::TranslatorPool translator(num_translators,

  const std::vector<std::vector<std::string>> batch = {{"▁H", "ello", "▁world", "!"}};
  const std::vector<ctranslate2::TranslationResult> results = translator.translate_batch(batch);

  for (const auto& token : results[0].output())
    std::cout << token << ' ';
  std::cout << std::endl;
  return 0;

See the TranslatorPool class for more advanced usages such as asynchronous translations.

Environment variables

Some environment variables can be configured to customize the execution:

  • CT2_CUDA_ALLOCATOR: Select the CUDA memory allocator. Possible values are: cub_caching, cuda_malloc_async (requires CUDA >= 11.2). The default allocator depends on the CUDA version:
    • CUDA >= 11.2: cuda_malloc_async
    • CUDA < 11.2: cub_caching
  • CT2_CUDA_ALLOW_FP16: Allow using FP16 computation on GPU even if the device does not have efficient FP16 support.
  • CT2_CUDA_CACHING_ALLOCATOR_CONFIG: Tune the CUDA caching allocator (see Performance).
  • CT2_FORCE_CPU_ISA: Force CTranslate2 to select a specific instruction set architecture (ISA). Possible values are: GENERIC, AVX, AVX2. Note: this does not impact backend libraries (such as Intel MKL) which usually have their own environment variables to configure ISA dispatching.
  • CT2_TRANSLATORS_CORE_OFFSET: If set to a non negative value, parallel translators are pinned to cores in the range [offset, offset + inter_threads]. Requires compilation with -DOPENMP_RUNTIME=NONE.
  • CT2_USE_EXPERIMENTAL_PACKED_GEMM: Enable the packed GEMM API for Intel MKL (see Performance).
  • CT2_USE_MKL: Force CTranslate2 to use (or not) Intel MKL. By default, the runtime automatically decides whether to use Intel MKL or not based on the CPU vendor.
  • CT2_VERBOSE: Configure the logs verbosity:
    • -3 = off
    • -2 = critical
    • -1 = error
    • 0 = warning (default)
    • 1 = info
    • 2 = debug
    • 3 = trace

When using Python, these variables should be set before importing the ctranslate2 module, e.g.:

import os
os.environ["CT2_VERBOSE"] = "1"

import ctranslate2


Docker images

The Docker images build the C++ shared libraries, the translation client, and the Python package. The docker build command should be run from the project root directory, e.g.:

docker build -t opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2 -f docker/Dockerfile .

See the docker/ directory for available images.

Build options

The project uses CMake for compilation. The following options can be set with -DOPTION=VALUE:

CMake option Accepted values (default in bold) Description
BUILD_CLI OFF, ON Compiles the translation clients
BUILD_TESTS OFF, ON Compiles the tests
CMAKE_CXX_FLAGS compiler flags Defines additional compiler flags
CUDA_DYNAMIC_LOADING OFF, ON Enables the dynamic loading of CUDA libraries at runtime instead of linking against them. Requires CUDA >= 11.
ENABLE_CPU_DISPATCH OFF, ON Compiles CPU kernels for multiple ISA and dispatches at runtime (should be disabled when explicitly targeting an architecture with the -march compilation flag)
ENABLE_PROFILING OFF, ON Enables the integrated profiler (usually disabled in production builds)
OPENMP_RUNTIME INTEL, COMP, NONE Selects or disables the OpenMP runtime (INTEL: Intel OpenMP; COMP: OpenMP runtime provided by the compiler; NONE: no OpenMP runtime)
WITH_CUDA OFF, ON Compiles with the CUDA backend
WITH_DNNL OFF, ON Compiles with the oneDNN backend (a.k.a. DNNL)
WITH_MKL OFF, ON Compiles with the Intel MKL backend
WITH_ACCELERATE OFF, ON Compiles with the Apple Accelerate backend
WITH_OPENBLAS OFF, ON Compiles with the OpenBLAS backend
WITH_RUY OFF, ON Compiles with the Ruy backend

Some build options require external dependencies:

  • -DWITH_MKL=ON requires:
  • -DWITH_DNNL=ON requires:
  • -DWITH_ACCELERATE=ON requires:
  • -DWITH_OPENBLAS=ON requires:
  • -DWITH_CUDA=ON requires:

Multiple backends can be enabled for a single build. When building with both Intel MKL and oneDNN, the backend will be selected at runtime based on the CPU information.

Example (Ubuntu)

Install Intel MKL (optional for GPU only builds)

Use the following instructions to install Intel MKL:

wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo sh -c 'echo "deb https://apt.repos.intel.com/oneapi all main" > /etc/apt/sources.list.d/oneAPI.list'
sudo apt-get update
sudo apt-get install intel-oneapi-mkl-devel

See the Intel MKL documentation for other installation methods.

Install CUDA (optional for CPU only builds)

See the NVIDIA documentation for information on how to download and install CUDA.


Under the project root, run the following commands:

git submodule update --init --recursive
mkdir build && cd build
make -j4

(If you did not install one of Intel MKL or CUDA, set its corresponding flag to OFF in the CMake command line.)

These steps should produce the cli/translate binary. You can try it with the model converted in the Quickstart section:

$ echo "▁H ello ▁world !" | ./cli/translate --model ende_ctranslate2/ --device auto
▁Hallo ▁Welt !



To enable the tests, you should configure the project with cmake -DBUILD_TESTS=ON. The binary tests/ctranslate2_test runs all tests using Google Test. It expects the path to the test data as argument:

./tests/ctranslate2_test ../tests/data


# Install the CTranslate2 library.
cd build && make install && cd ..

# Build and install the Python wheel.
cd python
pip install -r install_requirements.txt
python setup.py bdist_wheel
pip install dist/*.whl

# Run the tests with pytest.
pip install -r tests/requirements.txt
pytest tests/test.py

Depending on your build configuration, you might need to set LD_LIBRARY_PATH if missing libraries are reported when running tests/test.py.



For a fair comparison, we restrict the benchmark to toolkits compatible with the pretrained English-German Transformer model from OpenNMT-py or OpenNMT-tf.

We translate the test set newstest2014 and report the number of target tokens generated per second. The results are aggregated over multiple runs (see the benchmark scripts for more details).

Please note that the results presented below are only valid for the configuration used during this benchmark: absolute and relative performance may change with different settings.


Tokens per second Max. memory BLEU
OpenNMT-tf 2.24.0 (with TensorFlow 2.7.0) 335.9 2679MB 26.93
OpenNMT-py 2.2.0 (with PyTorch 1.9.1) 462.3 1650MB 26.77
- int8 500.6 1527MB 26.72
CTranslate2 2.11.0 1218.6 1069MB 26.77
- int16 1593.0 973MB 26.84
- int8 1872.7 854MB 26.88
- int8 + vmap 2312.3 726MB 26.65

Executed with 8 threads on a c5.metal Amazon EC2 instance equipped with an Intel(R) Xeon(R) Platinum 8275CL CPU.


Tokens per second Max. GPU memory Max. CPU memory BLEU
OpenNMT-tf 2.24.0 (with TensorFlow 2.7.0) 1424.2 2670MB 2377MB 26.93
OpenNMT-py 2.2.0 (with PyTorch 1.9.1) 1373.2 3082MB 3965MB 26.77
FasterTransformer 4.0 3263.5 5868MB 2633MB 26.77
- float16 6750.7 3916MB 2613MB 26.83
CTranslate2 2.11.0 3906.8 1264MB 673MB 26.77
- int8 5586.3 976MB 562MB 26.90
- float16 5805.0 816MB 599MB 26.78
- int8 + float16 6409.9 688MB 560MB 26.88

Executed with CUDA 11 on a g4dn.xlarge Amazon EC2 instance equipped with a NVIDIA T4 GPU (driver version: 470.82.01).

Model size

The table below compares the model size on disk of the pretrained Transformer models which are "base" Transformers without shared embeddings and a vocabulary of size 32k:

Model size
OpenNMT-py 542MB
OpenNMT-tf 367MB
CTranslate2 364MB
- int16 187MB
- float16 182MB
- int8 100MB
- int8 + float16 95MB

Frequently asked questions

How does it relate to the original CTranslate project?

The original CTranslate project shares a similar goal which is to provide a custom execution engine for OpenNMT models that is lightweight and fast. However, it has some limitations that were hard to overcome:

  • a strong dependency on LuaTorch and OpenNMT-lua, which are now both deprecated in favor of other toolkits;
  • a direct reliance on Eigen, which introduces heavy templating and a limited GPU support.

CTranslate2 addresses these issues in several ways:

  • the core implementation is framework agnostic, moving the framework specific logic to a model conversion step;
  • the call to external libraries (Intel MKL, cuBLAS, etc.) occurs as late as possible in the execution to not rely on a library specific logic.

What is the state of this project?

The implementation has been generously tested in production environment so people can rely on it in their application. The project versioning follows Semantic Versioning 2.0.0. The following APIs are covered by backward compatibility guarantees:

  • Converted models
  • Python converters options
  • Python symbols:
    • ctranslate2.Translator
    • ctranslate2.converters.FairseqConverter
    • ctranslate2.converters.OpenNMTPyConverter
    • ctranslate2.converters.OpenNMTTFConverter
  • C++ symbols:
    • ctranslate2::models::Model
    • ctranslate2::TranslationOptions
    • ctranslate2::TranslationResult
    • ctranslate2::Translator
    • ctranslate2::TranslatorPool
  • C++ translation client options

Other APIs are expected to evolve to increase efficiency, genericity, and model support.

Why and when should I use this implementation instead of PyTorch or TensorFlow?

Here are some scenarios where this project could be used:

  • You want to accelarate standard translation models for production usage, especially on CPUs.
  • You need to embed translation models in an existing C++ application without adding large dependencies.
  • Your application requires custom threading and memory usage control.
  • You want to reduce the model size on disk and/or memory.

However, you should probably not use this project when:

  • You want to train custom architectures not covered by this project.
  • You see no value in the key features listed at the top of this document.

What hardware is supported?


  • x86-64 processors supporting at least SSE 4.1
  • AArch64 processors

On x86-64, prebuilt binaries are configured to automatically select the best backend and instruction set architecture for the platform (AVX, AVX2, or AVX512). In particular, they are compiled with both Intel MKL and oneDNN so that Intel MKL is only used on Intel processors where it performs best, whereas oneDNN is used on other x86-64 processors such as AMD.


  • NVIDIA GPUs with a Compute Capability greater or equal to 3.5

The driver requirement depends on the CUDA version. See the CUDA Compatibility guide for more information.

What are the known limitations?

The current approach only exports the weights from existing models and redefines the computation graph via the code. This implies a strong assumption of the graph architecture executed by the original framework.

What are the future plans?

There are many ways to make this project better and even faster. See the open issues for an overview of current and planned features.

What is the difference between intra_threads and inter_threads?

  • intra_threads is the number of OpenMP threads that is used per translation: increase this value to decrease the latency of CPU translations.
  • inter_threads is the maximum number of translations executed in parallel: increase this value to increase the throughput. Even though the model data are shared, this execution mode will increase the memory usage as some internal buffers are duplicated for thread safety.

The total number of computing threads launched by the process is inter_threads * intra_threads.

On GPU, translations executed in parallel are using separate CUDA streams. Depending on the workload and GPU specifications this may or may not improve the translation throughput. For better parallelism on GPU, consider running the translation on multiple GPUs. See the option device_index that accepts multiple device IDs.

Do you provide a translation server?

The OpenNMT-py REST server is able to serve CTranslate2 models. See the code integration to learn more.

How do I generate a vocabulary mapping file?

The vocabulary mapping file (a.k.a. vmap) maps source N-grams to a list of target tokens. During translation, the target vocabulary will be dynamically reduced to the union of all target tokens associated with the N-grams from the batch to translate.

It is a text file where each line has the following format:

src_1 src_2 ... src_N<TAB>tgt_1 tgt_2 ... tgt_K

If the source N-gram is empty (N = 0), the assiocated target tokens will always be included in the reduced vocabulary.

See here for an example on how to generate this file. The file can then be passed to the converter script to be included in the model directory (see option --vocab_mapping) and can be used during translation after enabling the use_vmap translation option.

ezoic increase your site revenue


Fast inference engine for Transformer models


License:MIT License


Language:C++ 74.5%Language:Python 11.2%Language:Cuda 10.7%Language:CMake 1.8%Language:Shell 1.1%Language:Dockerfile 0.6%Language:C 0.1%