abtranslate

ABTranslate

A translator wrapper for open-source translation models with batch optimization. Currently supports ArgosTranslate models and is designed for scalable deployment.

Features

• ✅ Optimized translation for list or pandas.Series of text
• 🛡️ Supports masking of specific patterns to exclude them from translation
• ⚙️ Thread-tuned for multi-threaded deployment with ctranslate2
• 📦 Compatible with ArgosTranslate model packages

Installation

pip install abtranslate

Usage

from abtranslate import load_argostranslate_model

# Load model from local path
model_package = load_argostranslate_model(model_path, package_dir=package_extraction_dir)

# Load translator with optional optimizations
translator = model_package.load_translator(optimized_config=True, lazy_load=False)

# Define translation quality settings (based on CTranslate2)
QUALITY_CONFIG = {
    "beam_size": 4,
    "num_hypotheses": 1,
    "replace_unknowns": True,
}

# Translate a batch of texts (e.g. pandas Series or list)
translated_texts = translator.translate_batch(df["input_text"], translation_config=QUALITY_CONFIG)

🔗 Refer to CTranslate2 Translate Batch Parameters for more configuration options.

Main Functions

• load_argostranslate_model(model_path: str, package_dir: str) -> ModelPackage: Loads and prepares an ArgosTranslate model for usage.

• ModelPackage.load_translator(translator_config: dict, optimized_config: bool, lazy_load: bool) -> Translator: Returns a translator instance with optional optimizations.

• Translator.translate_batch(text_list: List[str] | pd.Series, translation_config: dict, return_type: any) -> List[str]: Performs high-performance batch translation with support for multi-threading and tuning.

License

MIT

Author

Ichsan Takwa GitHub