TensorRT 2 int8 inference for multi-label models

Hi everyone,
I have read the TensorRT guides and the sample_int8 code, they all seem to focus on classification models that produce a single label for each input image. Now how can we perform int8 inference for models that give multiple labels for each input image? My question is more focused on the calibration part as the building and executing seems similar to FP32 otherwise.