More information about "safety restricted mode"


I can’t found a clear documentation about “safety restricted mode” of TensorRT. What type of operation the network can performs in this mode?

I have tested a network using trtexec and without --safe the network work. With --safe the program gives me this error:
`[08/07/2020-19:51:21] [E] [TRT] (Unnamed Layer* 140) [Matrix Multiply]: MatrixMultiplyLayer not supported in safety restricted mode.

[08/07/2020-19:51:21] [E] [TRT] Layer (Unnamed Layer* 140) [Matrix Multiply] failed validation

It’s seems that TensorRT can’t perform matrix multiplication in safe mode. There is a documentation in which is listed usable operation in this mode?



TensorRT Version: 7.0.0
GPU Type: 2080
Nvidia Driver Version: 440.100
CUDA Version: 10.2
CUDNN Version: 7.6.5
Operating System + Version: Ubuntu 18.04
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

Please attach or include links to any models, data, files, or scripts necessary to reproduce your issue. (Github repo, Google Drive, Dropbox, etc.)

Steps To Reproduce

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered

Hi @pippomuzzo,
Your query has been noted. Please allow me sometime to check on this.


1 Like

There really isn’t any safety restricted mode in TensorRT 7.0. Safety restricted mode is part of TensorRT released with DriveOS and is documented in the DriveOS SDK.

Thank You for reply.
So what is the purpose of the --safe argument of trtexec present in the TensorRT standard release (not DriveOS)?

In the documentation of trtexec ( it is wrote
" Specifying the --safe parameter turns the safety mode switch ON . By default, the --safe parameter is not specified; the safety mode switch is OFF . The layers and parameters that are contained within the --safe subset are restricted if the switch is set to ON . The switch is used for prototyping the safety restricted flows until the TensorRT safety runtime is made available."

It seems that there is a “safety restricted mode” in the TensortRT standard release. But it can be a my wrong interpretation.