Hi everyone,
it is possible to run the imagenet example in TensorRT with FP32 precision instead of FP16? If so, how can I change it.
I hank you for the help.
Hi everyone,
it is possible to run the imagenet example in TensorRT with FP32 precision instead of FP16? If so, how can I change it.
I hank you for the help.
Hi maycondouglasd, please change tensorNet.cpp:56 to the following code:
mOverride16 = true;
This will disable FP16 all together. If you want to disable FP16 on a specific runtime model instance, there is the ImageNet::DisableFP16() function.