Efficient Ternary Weight Embedding Model: Bridging Scalability and Performance