MixPE: Quantization and Hardware Co-design for Efficient LLM Inference