Search quantization parameters of each layer in hyper (scale) decoder.
Suggested calibration set is at ./data/calibration_set which is a subset of validation data.
Use command line:
./scripts/quantize_model.sh
The script will quantize models from directories models/VM_base and models/VM_high, and store them to directories models/VM_base_int and models/VM_high_int.