contrib.tensorrt

Module to enable the use of TensorRT optimized graphs.

Functions

get_use_fp16()

Get an environment variable which describes if TensorRT is currently running in FP16

init_tensorrt_params(sym, arg_params, aux_params)

Set weights in attributes of TensorRT nodes

set_use_fp16(status)

Set an environment variable which will enable or disable the use of FP16 precision in

mxnet.contrib.tensorrt.get_use_fp16()[source]

Get an environment variable which describes if TensorRT is currently running in FP16 :return: Boolean, true if TensorRT is running in FP16, False for FP32

mxnet.contrib.tensorrt.init_tensorrt_params(sym, arg_params, aux_params)[source]

Set weights in attributes of TensorRT nodes :param sym: Symbol, the symbol graph should contains some TensorRT nodes :param arg_params: arg_params :param aux_params: aux_params :return arg_params, aux_params: remaining params that are not in TensorRT nodes

mxnet.contrib.tensorrt.set_use_fp16(status)[source]

Set an environment variable which will enable or disable the use of FP16 precision in TensorRT Note: The mode FP16 force the whole TRT node to be executed in FP16 :param status: Boolean, True if TensorRT should run in FP16, False for FP32