https://mer.dev/posts/i-vit-integer-only-quantization-for-efficient-vision-transformer-inference/