Ct2fast Opus Mt ROMANCE En

🚀 Fast-Inference with Ctranslate2

This project uses int8 inference in C++ to speed up inference by 2x - 8x. It's a quantized version of Helsinki-NLP/opus-mt-ROMANCE-en, aiming to provide a more efficient solution for translation tasks.

🚀 Quick Start

📦 Installation

Install the necessary packages using the following command:

pip install hf-hub-ctranslate2>=1.0.0 ctranslate2>=3.13.0

Convert the model using the following command:

ct2-transformers-converter --model Helsinki-NLP/opus-mt-ROMANCE-en --output_dir /home/michael/tmp-ct2fast-opus-mt-ROMANCE-en --force --copy_files README.md generation_config.json tokenizer_config.json vocab.json source.spm .gitattributes target.spm --quantization float16

💻 Usage Examples

Basic Usage

from hf_hub_ctranslate2 import TranslatorCT2fromHfHub, GeneratorCT2fromHfHub
from transformers import AutoTokenizer

model_name = "michaelfeil/ct2fast-opus-mt-ROMANCE-en"
# use either TranslatorCT2fromHfHub or GeneratorCT2fromHfHub here, depending on model.
model = TranslatorCT2fromHfHub(
        # load in int8 on CUDA
        model_name_or_path=model_name, 
        device="cuda",
        compute_type="int8_float16",
        tokenizer=AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ROMANCE-en")
)
outputs = model.generate(
    text=["How do you call a fast Flan-ingo?", "User: How are you doing?"],
)
print(outputs)

🔧 Technical Details

The checkpoint is compatible with ctranslate2 and hf-hub-ctranslate2.

• Use compute_type=int8_float16 for device="cuda"
• Use compute_type=int8 for device="cpu"

📄 License

This is just a quantized version. Licence conditions are intended to be identical to the original huggingface repo.

📚 Documentation

Original description

opus-mt-ROMANCE-en

Benchmarks