This survey explores how binary neural networks can dramatically reduce the computational and memory requirements of large language models while maintaining acceptable performance.

• Binary quantization can reduce model size by up to 32x compared to full-precision models
• Current techniques include Binary Representation Learning and Binary Reasoning Enhancement
• Challenges remain in preserving model capabilities during extreme compression
• Binary LLMs offer potential for edge device deployment and resource-constrained environments

This engineering breakthrough enables more efficient model deployment, reduced inference costs, and broader accessibility of LLM technology across computing environments.

Binary Neural Networks for Large Language Model: A Survey