# TensorRT backend for ONNX on jetson nano

**URL:** <https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980>\
**Category:** Jetson Nano\
**Tags:** tensorrt\
**Created:** [May 20, 2019, 8:23am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980 "2019-05-20T08:23:05Z")\
**Posts on this page:** 20\
**Page:** 1

<div class="post-metadata">

**Author:** ![Walter\_LIU](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@Walter\_LIU](https://forums.developer.nvidia.com/u/Walter_LIU)\
**Post date:** [May 20, 2019, 8:23am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/1 "2019-05-20T08:23:05Z")

</div>

HI，expert

I have Installationed TensorRT backend for ONNX on my jetson nano. But I can’t pass the onnx\_backend\_test.py .

ONNX backend tests can be run as follows:

```auto
hgnan@jetson-nano:~/nas/onnx-tensorrt$ python onnx_backend_test.py
s(Unnamed Layer* 0) [Unary]
(4, 5)
.sssssssssss(Unnamed Layer* 0) [ElementWise]
(4, 5)
.sssssssssssssssssssssssssssssssssssssssssssssssssssssssss(Unnamed Layer* 0) [Shuffle]
(Unnamed Layer* 1) [Pooling]
(Unnamed Layer* 2) [Shuffle]
(3, 31)
...............

.Start downloading model vgg19 from https://s3.amazonaws.com/download.onnx/models/opset_9/vgg19.tar.gz
Done
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:604] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:81] The total number of bytes read was 574674712
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:604] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:81] The total number of bytes read was 574674712

(Unnamed Layer* 35) [Activation]
(Unnamed Layer* 36) [Pooling]
(Unnamed Layer* 37) [Shuffle]
(Unnamed Layer* 38) [Shuffle]
(Unnamed Layer* 39) [Fully Connected]
(Unnamed Layer* 40) [Shuffle]
(Unnamed Layer* 41) [Activation]
(Unnamed Layer* 42) [Shuffle]
(Unnamed Layer* 43) [Shuffle]
(Unnamed Layer* 44) [Fully Connected]
(Unnamed Layer* 45) [Shuffle]
(Unnamed Layer* 46) [Activation]
(Unnamed Layer* 47) [Shuffle]
(Unnamed Layer* 48) [Shuffle]
(Unnamed Layer* 49) [Fully Connected]
(Unnamed Layer* 50) [Shuffle]
(Unnamed Layer* 51) [Shuffle]
(Unnamed Layer* 52) [Softmax]
(Unnamed Layer* 53) [Shuffle]
(1000,)
Killed

```

Do you have an idea how to fix this? :)

---

<div class="post-metadata">

**Author:** ![dusty\_nv](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/dusty_nv/32/14043_2.png) [@dusty\_nv](https://forums.developer.nvidia.com/u/dusty_nv)\
**Post date:** [May 20, 2019, 12:03pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/2 "2019-05-20T12:03:37Z")

</div>

Hi Walter, can you try running sudo tegrastats in the background during this test, and keeping an eye on the memory usage? Perhaps this test program is consuming all the memory available. The “killed” message is typically an indicator that the system is out of memory.

If that’s the case, you can try running your Nano headless (without display attached) to save memory, or mount a swap file.

---

<div class="post-metadata">

**Author:** ![Walter\_LIU](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@Walter\_LIU](https://forums.developer.nvidia.com/u/Walter_LIU)\
**Post date:** [May 20, 2019, 4:50pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/3 "2019-05-20T16:50:54Z")

</div>

Hi, Dusty\_nv

Thanks your support.  
I try running “sudo python onnx\_backend\_test.py” in headless and swapon mode , but the test failed at the same stage. I have closed necessary system service to save memory.  
jtop as follows:

```auto
jtop - Raffaello Bonghi
NVIDIA Jetson NANO/TX1 - Jetpack 4.2 [L4T 32.1.0]
CPU1 [|||| schedutil - 18%] 204MHz
CPU2 [|||| schedutil - 16%] 204MHz
CPU3 [|| schedutil - 9%] 204MHz
CPU4 [|||| schedutil - 16%] 204MHz

Mem [|3.9GB/4.0GB] (lfb 2x512MB)
EMC [||| 5%] 204MHz
Imm [0.0GB/0.3GB] (lfb 252MB)
Swp [| 2.9GB/8.2GB] (cached 120MB)

GPU [0%] 76MHz
Dsk [# 27.1GB/58.4GB]
 APE: 25MHz [Sensor] [Temp] [Power] [Cur/Avr]
 Board info: AO 44.00C POM_5V_CPU 301 mW/555 mW
   Name: NANO/TX1 PMIC 100.00C POM_5V_IN 1852mW/3074mW
   JP: 4.2 [L4T 32.1.0] thermal 38.00C POM_5V_GPU 0 mW/754 mW
 NV Power: MAXN - 0 GPU 37.50C
                             PLL 35.50C
                             CPU 38.00C

```

I can successfully load mobilenetv2-1.0.onnx and resnet18v1.onnx ，but load yolov3.onnx is fail.

```auto
hgnan@jetson-nano:~/nas/onnx-tensorrt$ cat trt-backend.py
import onnx
import onnx_tensorrt.backend as backend
import numpy as np

model = onnx.load("yolov3.onnx")
engine = backend.prepare(model, device='CUDA:0')
input_data = np.random.random(size=(32, 3, 224, 224)).astype(np.float32)
output_data = engine.run(input_data)[0]
print(output_data)
print(output_data.shape)
hgnan@jetson-nano:~/nas/onnx-tensorrt$ python3 trt-backend.py
Traceback (most recent call last):
  File "trt-backend.py", line 6, in <module>
    engine = backend.prepare(model, device='CUDA:0')
  File "/home/hgnan/nas/onnx-tensorrt/onnx_tensorrt/backend.py", line 217, in prepare
    super(TensorRTBackend, cls).prepare(model, device, **kwargs)
  File "/home/hgnan/.local/lib/python3.6/site-packages/onnx/backend/base.py", line 74, in prepare
    onnx.checker.check_model(model)
  File "/home/hgnan/.local/lib/python3.6/site-packages/onnx/checker.py", line 86, in check_model
    C.check_model(model.SerializeToString())
onnx.onnx_cpp2py_export.checker.ValidationError: Nodes in a graph must be topologically sorted, however input 'y3:01' of node:
input: "y3:01" output: "TFNodes/yolo_evaluation_layer_1/Shape_3:0" name: "TFNodes/yolo_evaluation_layer_1/Shape_3" op_type: "Shape"
 is not output of any previous nodes.
hgnan@jetson-nano:~/nas/onnx-tensorrt$

```

---

<div class="post-metadata">

**Author:** ![AastaLLL](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/aastalll/32/14043_2.png) [@AastaLLL](https://forums.developer.nvidia.com/u/AastaLLL)\
**Post date:** [May 27, 2019, 4:38am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/4 "2019-05-27T04:38:19Z")

</div>

Hi,

We have a TensorRT sample for YOLOv3 with .onnx format.

Would you mind to check it first?  
/usr/src/tensorrt/samples/python/yolov3\_onnx

Thanks.

---

<div class="post-metadata">

**Author:** ![sh2222](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sh2222](https://forums.developer.nvidia.com/u/sh2222)\
**Post date:** [May 29, 2019, 10:57am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/5 "2019-05-29T10:57:48Z")

</div>

The conversion of the YoloV3-608 to ONNX does not work because the python script yolov3\_to\_onnx.py fails with the following errors.  
It would be great if you could fix this because I like to convert the ONNX model to TensorRT.

Layer of type yolo not supported, skipping ONNX node generation.  
Layer of type yolo not supported, skipping ONNX node generation.  
Layer of type yolo not supported, skipping ONNX node generation.  
graph YOLOv3-608 (  
%000\_net[FLOAT, 64x3x608x608]  
) initializers (

.  
.  
.  
Building Something …  
.  
.  
.  
\_conv\_weights, %106\_convolutional\_conv\_bias)  
return %082\_convolutional, %094\_convolutional, %106\_convolutional  
}  
Traceback (most recent call last):  
File “yolov3\_to\_onnx.py”, line 812, in   
main()  
File “yolov3\_to\_onnx.py”, line 805, in main  
onnx.checker.check\_model(yolov3\_model\_def)  
File “/home/sh/.local/lib/python2.7/site-packages/onnx/checker.py”, line 86, in check\_model  
C.check\_model(model.SerializeToString())  
onnx.onnx\_cpp2py\_export.checker.ValidationError: Op registered for Upsample is depracted in domain\_version of 10

==\> Context: Bad node spec: input: “085\_convolutional\_lrelu” input: “086\_upsample\_scale” output: “086\_upsample” name: “086\_upsample” op\_type: “Upsample” attribute { name: “mode” s: “nearest” type: STRING }

---

<div class="post-metadata">

**Author:** ![AastaLLL](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/aastalll/32/14043_2.png) [@AastaLLL](https://forums.developer.nvidia.com/u/AastaLLL)\
**Post date:** [June 3, 2019, 4:44am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/6 "2019-06-03T04:44:27Z")

</div>

Hi,

This sample required onnx=1.4.1.  
The latest ONNX(1.5.0) is deprecated Upsample layer which causes the error.

Could you try to update your environment to onnx v1.4.1 and try it again?

```auto
pip3 uninstall onnx
pip3 install onnx==1.4.1 --user

```

Thanks.

---

<div class="post-metadata">

**Author:** ![sh2222](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sh2222](https://forums.developer.nvidia.com/u/sh2222)\
**Post date:** [June 3, 2019, 10:49am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/7 "2019-06-03T10:49:27Z")

</div>

Hi,

Unfortunately, the script shows the same error (Bad node spec …) even after installing the ONNX version 1.4.1:

```auto
pip3 show onnx
Name: onnx
Version: 1.4.1
Summary: Open Neural Network Exchange
Home-page: https://github.com/onnx/onnx
Author: bddppq
Author-email: jbai@fb.com
License: UNKNOWN
Location: /home/sh/.local/lib/python3.6/site-packages
Requires: numpy, typing, typing-extensions, six, protobuf
Required-by:

```

PS:  
Thanks for your great work!

---

<div class="post-metadata">

**Author:** ![sojohans](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sojohans](https://forums.developer.nvidia.com/u/sojohans)\
**Post date:** [June 3, 2019, 2:58pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/8 "2019-06-03T14:58:38Z")

</div>

Hi

I tried to do :

pip3 uninstall onnx  
pip3 install onnx==1.4.1 --user

But got the error (below). Any hints?

Building wheels for collected packages: onnx  
WARNING: Building wheel for onnx failed: [Errno 13] Permission denied: ‘/home/soren/.cache/pip/wheels/c5’  
Failed to build onnx  
Installing collected packages: onnx  
Running setup.py install for onnx … error  
ERROR: Complete output from command /usr/bin/python3 -u -c ‘import setuptools, tokenize; **file** =’“'”‘/tmp/pip-install-ynga2nui/onnx/setup.py’“'”‘;f=getattr(tokenize, ‘"’“‘open’”’“‘, open)( **file** );code=f.read().replace(’”‘"’\r\n’“'”‘, ‘"’"’\n’“'”‘);f.close();exec(compile(code, **file** , ‘"’“‘exec’”’"‘))’ install --record /tmp/pip-record-wevgeg\_d/install-record.txt --single-version-externally-managed --compile --user --prefix=:  
ERROR: fatal: not a git repository (or any of the parent directories): .git  
running install  
running build  
running build\_py  
running create\_version  
running cmake\_build  
– Build type not set - defaulting to Release  
– The C compiler identification is GNU 7.4.0  
– The CXX compiler identification is GNU 7.4.0  
– Check for working C compiler: /usr/bin/cc  
– Check for working C compiler: /usr/bin/cc – works  
– Detecting C compiler ABI info  
– Detecting C compiler ABI info - done  
– Detecting C compile features  
– Detecting C compile features - done  
– Check for working CXX compiler: /usr/bin/c++  
– Check for working CXX compiler: /usr/bin/c++ – works  
– Detecting CXX compiler ABI info  
– Detecting CXX compiler ABI info - done  
– Detecting CXX compile features  
– Detecting CXX compile features - done  
CMake Error at CMakeLists.txt:217 (message):  
Protobuf compiler not found  
Call Stack (most recent call first):  
CMakeLists.txt:248 (relative\_protobuf\_generate\_cpp)

```
-- Configuring incomplete, errors occurred!
See also "/tmp/pip-install-ynga2nui/onnx/.setuptools-cmake-build/CMakeFiles/CMakeOutput.log".
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/tmp/pip-install-ynga2nui/onnx/setup.py", line 328, in <module>
    'backend-test-tools = onnx.backend.test.cmd_tools:main',
  File "/usr/lib/python3/dist-packages/setuptools/ __init__.py", line 129, in setup
    return distutils.core.setup(**attrs)
  File "/usr/lib/python3.6/distutils/core.py", line 148, in setup
    dist.run_commands()
  File "/usr/lib/python3.6/distutils/dist.py", line 955, in run_commands
    self.run_command(cmd)
  File "/usr/lib/python3.6/distutils/dist.py", line 974, in run_command
    cmd_obj.run()
  File "/usr/lib/python3/dist-packages/setuptools/command/install.py", line 61, in run
    return orig.install.run(self)
  File "/usr/lib/python3.6/distutils/command/install.py", line 589, in run
    self.run_command('build')
  File "/usr/lib/python3.6/distutils/cmd.py", line 313, in run_command
    self.distribution.run_command(command)
  File "/usr/lib/python3.6/distutils/dist.py", line 974, in run_command
    cmd_obj.run()
  File "/usr/lib/python3.6/distutils/command/build.py", line 135, in run
    self.run_command(cmd_name)
  File "/usr/lib/python3.6/distutils/cmd.py", line 313, in run_command
    self.distribution.run_command(command)
  File "/usr/lib/python3.6/distutils/dist.py", line 974, in run_command
    cmd_obj.run()
  File "/tmp/pip-install-ynga2nui/onnx/setup.py", line 203, in run
    self.run_command('cmake_build')
  File "/usr/lib/python3.6/distutils/cmd.py", line 313, in run_command
    self.distribution.run_command(command)
  File "/usr/lib/python3.6/distutils/dist.py", line 974, in run_command
    cmd_obj.run()
  File "/tmp/pip-install-ynga2nui/onnx/setup.py", line 190, in run
    subprocess.check_call(cmake_args)
  File "/usr/lib/python3.6/subprocess.py", line 291, in check_call
    raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['/usr/bin/cmake', '-DPYTHON_INCLUDE_DIR=/usr/include/python3.6m', '-DPYTHON_EXECUTABLE=/usr/bin/python3', '-DBUILD_ONNX_PYTHON=ON', '-DCMAKE_EXPORT_COMPILE_COMMANDS=ON', '-DONNX_NAMESPACE=onnx', '-DPY_EXT_SUFFIX=.cpython-36m-aarch64-linux-gnu.so', '/tmp/pip-install-ynga2nui/onnx']' returned non-zero exit status 1.
----------------------------------------

```

ERROR: Command “/usr/bin/python3 -u -c ‘import setuptools, tokenize; **file** =’”‘"’/tmp/pip-install-ynga2nui/onnx/setup.py’“'”‘;f=getattr(tokenize, ‘"’“‘open’”’“‘, open)( **file** );code=f.read().replace(’”‘"’\r\n’“'”‘, ‘"’"’\n’“'”‘);f.close();exec(compile(code, **file** , ‘"’“‘exec’”’“‘))’ install --record /tmp/pip-record-wevgeg\_d/install-record.txt --single-version-externally-managed --compile --user --prefix=” failed with error code 1 in /tmp/pip-install-ynga2nui/onnx/

sojohans

---

<div class="post-metadata">

**Author:** ![sh2222](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sh2222](https://forums.developer.nvidia.com/u/sh2222)\
**Post date:** [June 3, 2019, 3:56pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/9 "2019-06-03T15:56:02Z")

</div>

The error log says:  
CMake Error at CMakeLists.txt:217 (message):  
Protobuf compiler not found

Try to install the Protobuf Compiler:

```auto
sudo apt-get install protobuf-compiler

```

---

<div class="post-metadata">

**Author:** ![sojohans](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sojohans](https://forums.developer.nvidia.com/u/sojohans)\
**Post date:** [June 3, 2019, 5:52pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/10 "2019-06-03T17:52:27Z")

</div>

Hi sh2222

Thanks but same error…

Sojohans

---

<div class="post-metadata">

**Author:** ![sh2222](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sh2222](https://forums.developer.nvidia.com/u/sh2222)\
**Post date:** [June 3, 2019, 5:59pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/11 "2019-06-03T17:59:31Z")

</div>

Please try:

```auto
pip3 install protobuf

```

---

<div class="post-metadata">

**Author:** ![AastaLLL](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/aastalll/32/14043_2.png) [@AastaLLL](https://forums.developer.nvidia.com/u/AastaLLL)\
**Post date:** [June 4, 2019, 2:22am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/12 "2019-06-04T02:22:47Z")

</div>

Hi, sh2222

Let me check this and update with you later.  
Thanks.

---

<div class="post-metadata">

**Author:** ![sojohans](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sojohans](https://forums.developer.nvidia.com/u/sojohans)\
**Post date:** [June 4, 2019, 6:50am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/13 "2019-06-04T06:50:46Z")

</div>

Hi AastaLLL

I did try pip3 install protobuf.

But it gave the same error when I installed onnx on jetson nano.

It is the python tensorrt example onnx\_yolo I am trying to get to work.

Sojohan

---

<div class="post-metadata">

**Author:** ![sojohans](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sojohans](https://forums.developer.nvidia.com/u/sojohans)\
**Post date:** [June 6, 2019, 7:58am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/14 "2019-06-06T07:58:25Z")

</div>

Hi AastaLLL

Did you find a solution to this?

Thanks,

sojohan

---

<div class="post-metadata">

**Author:** ![SB\_97](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@SB\_97](https://forums.developer.nvidia.com/u/SB_97)\
**Post date:** [June 6, 2019, 12:09pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/15 "2019-06-06T12:09:46Z")

</div>

The yolov3\_onnx sample worked for me once I had installed onnx 1.4.1. Before that I was getting the same check\_model error as sh222. Note that the yolov3\_to\_onnx.py script is only compatible with python2 so onnx should not be installed using pip3. You can check the version of onnx used by python2 with e.g. python2 -m pip freeze | grep onnx

---

<div class="post-metadata">

**Author:** ![sojohans](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sojohans](https://forums.developer.nvidia.com/u/sojohans)\
**Post date:** [June 6, 2019, 12:12pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/16 "2019-06-06T12:12:11Z")

</div>

Hi SB\_97

Thanks….Will try…

Sojohan

---

<div class="post-metadata">

**Author:** ![AastaLLL](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/aastalll/32/14043_2.png) [@AastaLLL](https://forums.developer.nvidia.com/u/AastaLLL)\
**Post date:** [June 10, 2019, 4:55am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/17 "2019-06-10T04:55:20Z")

</div>

Hi,

We can execute _yolov3\_to\_onnx.py_ without error.  
Here are our steps for your reference:

- Flash JetPack 4.2.

```auto
$ cp -r /usr/src/tensorrt/ .
$ cd tensorrt/samples/python/yolov3_onnx/
$ python2 -m pip install -r requirements.txt
$ python yolov3_to_onnx.py

```

Thanks.

---

<div class="post-metadata">

**Author:** ![sh2222](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@sh2222](https://forums.developer.nvidia.com/u/sh2222)\
**Post date:** [June 10, 2019, 6:26am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/18 "2019-06-10T06:26:49Z")

</div>

Installing ONNX 1.4.1 for python2 solved the problem.

BUT!

Do you have an idea how to run the 2nd step: python onnx\_to\_tensorrt.py to create the TensorRT Engine without running into a killed process due to memory issues?

---

<div class="post-metadata">

**Author:** ![SB\_97](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@SB\_97](https://forums.developer.nvidia.com/u/SB_97)\
**Post date:** [June 10, 2019, 7:38am UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/19 "2019-06-10T07:38:20Z")

</div>

> [@](#):
>
> Installing ONNX 1.4.1 for python2 solved the problem.
> 
> BUT!
> 
> Do you have an idea how to run the 2nd step: python onnx\_to\_tensorrt.py to create the TensorRT Engine without running into a killed process due to memory issues?

I solved this by using a USB flash drive as swap:

- plug in an empty USB stick (pref USB 3.0)
- sudo fdisk -l to identify the drive letter (it was /dev/sda1 on mine)
- sudo mkswap /dev/sdx1
- sudo swapon -p 32767 /dev/sdx1 (where 32767 is the highest priority)
- cat /proc/swaps to check the new swap is listed

---

<div class="post-metadata">

**Author:** ![mdegans](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/mdegans/32/13104_2.png) [@mdegans](https://forums.developer.nvidia.com/u/mdegans)\
**Post date:** [June 10, 2019, 3:18pm UTC](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980/20 "2019-06-10T15:18:17Z")

</div>

The -p flag may not do what you think. Priority only matters if there are multiple swap files (even then it usually makes sense to leave it the same, since if it is, round robin is used, and that will likely increase swap performance the more disks you add).

[https://unix.stackexchange.com/questions/84453/what-is-the-purpose-of-multiple-swap-files](https://unix.stackexchange.com/questions/84453/what-is-the-purpose-of-multiple-swap-files)

What you may be looking for is vm.swappiness. vm.swappiness is what tells the kernel how aggressively to swap.

Set at 10, the system will only start swapping when it’s almost out of ram. Meaning it performs well and then hits a wall.

A value closer to 90 will swap nearly all the time, making performance worse all the time, but the slowdown will also be more consistent.

Swappiness can be set temporarily with “sysctl vm.swappiness=10” ( recommend 10-90 ) and persistently in /etc/sysctl.conf by adding “vm.swappiness=10” at the end of the file.

[Next page](https://forums.developer.nvidia.com/t/tensorrt-backend-for-onnx-on-jetson-nano/74980.md?page=2)