can we using INT8 if there is a customer/plugin layer?

There is an attention “Currently only FP32 precision is supported by the plugin layer.” in “TensorRT 3.0 User Guide”(DU-08602-001_v3.0 | November 2017) document “3.8. SamplePlugin - Implementing A Custom Layer” section, what does it mean, we can’t use INT8 if we use a custome/plugin layer?


Currently, only FP32 is officially supported by TensorRT 3.
We are checking the possibility to enable FP16 support for plugin API. But no concrete schedule yet.

Thanks and sorry for the inconvenience.

Okay, so we can’t use INT8 if we use plugin tech. in TensorRT 3? We will make it as a check note.