ctranslate2 2.17.0

Fast inference engine for Transformer models

CTranslate2

CTranslate2 is a C++ and Python library for efficient inference with Transformer models. The project implements a custom runtime that applies many performance optimization techniques such as weights quantization, layers fusion, batch reordering, etc., to accelerate and reduce the memory usage of Transformer models on CPU and GPU. The following model types are currently supported:

• Encoder-decoder models: Transformer base/big, M2M-100, BART, mBART
• Decoder-only models: GPT-2

Compatible models should be first converted into an optimized model format. The library includes converters for multiple frameworks:

• OpenNMT-py
• OpenNMT-tf
• Fairseq
• Marian
• Transformers

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

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: layer fusion, 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/ARM64 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 execution
  Multiple batches can be processed 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 simple APIs in Python and C++ to cover most integration needs.
• Configurable and interactive decoding
  Advanced decoding features allow autocompleting a partial sequence and returning alternatives at a specific location in the sequence.

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

Installation and usage

CTranslate2 can be installed with pip:

pip install ctranslate2

The Python module is used to convert models and can translate or generate text with few lines of code:

translator = ctranslate2.Translator(translation_model_path)
translator.translate_batch(tokens)

generator = ctranslate2.Generator(generation_model_path)
generator.generate_batch(start_tokens)

See the documentation for more information and examples.

Benchmarks

We translate the En->De test set newstest2014 with multiple models:

• OpenNMT-tf WMT14: a base Transformer trained with OpenNMT-tf on the WMT14 dataset (4.5M lines)
• OpenNMT-py WMT14: a base Transformer trained with OpenNMT-py on the WMT14 dataset (4.5M lines)
• OPUS-MT: a base Transformer trained with Marian on all OPUS data available on 2020-02-26 (81.9M lines)

The benchmark reports the number of target tokens generated per second (higher is better). The results are aggregated over multiple runs. See the benchmark scripts for more details and reproduce these numbers.

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.

CPU

	Tokens per second	Max. memory	BLEU
OpenNMT-tf WMT14 model
OpenNMT-tf 2.25.0 (with TensorFlow 2.8.0)	342.4	2600MB	26.93
OpenNMT-py WMT14 model
OpenNMT-py 2.2.0 (with PyTorch 1.9.1)	458.8	1734MB	26.77
- int8	500.1	1612MB	26.72
CTranslate2 2.13.1	1217.8	1068MB	26.77
- int16	1530.9	989MB	26.82
- int8	1787.3	773MB	26.89
- int8 + vmap	2179.2	789MB	26.62
OPUS-MT model
Marian 1.11.0	756.7	13819MB	27.93
- int16	723.6	10393MB	27.65
- int8	857.3	8169MB	27.27
CTranslate2 2.13.1	993.5	901MB	27.92
- int16	1290.4	921MB	27.51
- int8	1486.5	748MB	27.71

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

GPU

	Tokens per second	Max. GPU memory	Max. CPU memory	BLEU
OpenNMT-tf WMT14 model
OpenNMT-tf 2.25.0 (with TensorFlow 2.8.0)	1285.7	2666MB	2364MB	26.93
OpenNMT-py WMT14 model
OpenNMT-py 2.2.0 (with PyTorch 1.9.1)	1220.9	3082MB	3900MB	26.77
FasterTransformer 4.0	2950.8	5868MB	2436MB	26.77
- float16	6499.3	3916MB	2423MB	26.83
CTranslate2 2.13.1	3747.1	1264MB	676MB	26.77
- int8	5306.4	976MB	561MB	26.83
- float16	5367.8	816MB	607MB	26.78
- int8 + float16	6158.7	688MB	563MB	26.80
OPUS-MT model
Marian 1.11.0	2221.5	3128MB	1932MB	27.92
- float16	2832.7	2986MB	1713MB	27.93
CTranslate2 2.13.1	3136.3	1200MB	481MB	27.92
- int8	4634.4	1008MB	519MB	27.89
- float16	4708.7	816MB	560MB	27.85
- int8 + float16	5320.3	720MB	515MB	27.81

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

Additional resources

• Documentation
• Forum
• Gitter