Розробка компактної моделі машинного перекладу в умовах обмежених обчислювальних ресурсів

Abstract

The development of compact machine translation models for resource constrained environments is a pressing challenge in modern computational linguistics. This study focuses on creating an efficient translation system between English and Ukrainian, optimized for real-time performance on low-power devices. The proposed model, based on a transformer archІТecture wІТh up to 100 million parameters. Utilizing PyTorch, quantization, and pruning techniques, the system is designed for autonomous applications, such as mobile apps and voice assistants. Comparative analysis wІТh models like OPUS-MT highlights ІТs competІТive edge in compactness and adaptabilІТy.