quantization_config.json

{
  "format": "quanto",
  "weights": "int8",
  "description": "ERNIE-Image INT8 release package with component-wise precision manifest."
}