INT8.input_type() and INT32.input_type()

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Hello,

I tested the data types of TRT:

>>> type(trt.infer.DataType.INT8.input_type())
<class 'numpy.float32'>
>>> type(trt.infer.DataType.INT32.input_type())
<class 'numpy.float32'>
>>> type(trt.infer.DataType.HALF.input_type())
<class 'numpy.float16'>
>>> type(trt.infer.DataType.FLOAT.input_type())
<class 'numpy.float32'>

Why the INT8.input_type() and INT32.input_type() indicate float32?
Is this intended results?

Thanks!

2

Hello,

This is an issue with TRT4. We are sorry for any inconvenience this datatype issue is causing. It is addressed in future versions of TensorRT.

Please pay attention to our announcement for the information on future releases.