text_encoder/z_image_int8-quantization.json

{
 "format": "INT8",
 "scale": "per-row",
 "per_tensor_relative_error": 0.058675,
 "per_tensor_weights_lost": 0.096222,
 "matrices_measured": 253
}