Sensitivity-aware layer-wise mixed precision KV cache quantization framework for efficient and...