BanglaTranslation: Pretraining Vanilla Transformer

• Tech Stack: Transformers, Pandas, Numpy, huggingface
• Github URL: Project Link

Pre-trained a Vanilla Transformer model for English-to-Bangla translation, implementing the complete architecture from scratch.

Explored and integrated two variants of the attention mechanism — Multi-Head Attention and Multi-Query Attention — within a unified encoder-decoder framework.