Riva issue running on Jetson Orin nano dev kit

I have flashed the SD card for my Jetson Orin nano dev kit with the latest Jetpack 5.1.1 Image.

I have downloaded this CLI image:
ngc registry resource download-version “nvidia/riva/riva_quickstart_arm64:2.10.0”

I have followed the install instructions from the rivia embedded quick start :

I do not have any problems with the install.
The issue is when I run “$ sudo bash riva_start.sh”
All I get is this:

Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Waiting for Riva server to load all models…retrying in 10 seconds
Health ready check failed.
Check Riva logs with: docker logs riva-speech

It says check the logs but there is nothing in them

Hi @adventuredaisy

Thanks for your interest in Riva

Apologies you are facing the issue,

Jetpack 5.1 is the correct version for Riva 2.10

Request to kindly share the following information with us

  1. config.sh used
  2. complete log output of docker logs riva-speech attached as file in this thread

Quick Checks

  1. we request to kindly check if you have set the default runtime to nvidia on the Jetson platform by adding the following line in the /etc/docker/daemon.json file
    If not already changed please Restart the Docker service using sudo systemctl restart docker after editing the file.

  2. We request to run basic container with nvidia runtime to check things are fine, so kindly run docker run -it --rm --runtime nvidia ubuntu:20.04 and let us know if it works

Thanks

So I was able to get Rivia to run on the Jetson Orin nano dev kit.
I had to load the minimum of services to get it to run and it still taxes the Jetson Orin Nano to the limit.

Here is all the things I did but I really dont know which one worked

  1. /etc/docker/daemon.json

{
“default-runtime”: “nvidia”,
“runtimes”: {
“nvidia”: {
“path”: “/usr/bin/nvidia-container-runtime”,
“runtimeArgs”:
}
}
}

2.config.sh

service_enabled_asr=true
service_enabled_nlp=false
service_enabled_tts=false
service_enabled_nmt=false
and
gpus_to_use=“device=1”

  1. I did notice when the models are downloaded the
    /home/joev/riva_quickstart_arm64_v2.10.0/model_repository
    has the lock symbol on it.
    So I ran this to unlock it.

sudo chmod -R a+rwx /home/joev/riva_quickstart_arm64_v2.10.0/model_repository

I think that running all the services at one is just to taxing for the Jetson Orin nano Dev kit

I will be posting a video soon with a little more in depth explanasion

HI @adventuredaisy

Thanks for the feedback

I will check with the internal team on

  1. .whether running all services at once will burden the Jetson Orin and provide feedback

  2. I will also check on the lock symbol issue faced

Thanks

HI @adventuredaisy

I have updates from the internal team,

  1. That’s True, Jetson Orin has limited GPU Memory, so loading all the models at once is not recommended, So it is expensive/taxing and we recommended running the single/limited service

  2. Regarding models downloaded has the lock symbol, This is expected as the Models are downloaded inside of Docker Volume, so should be fine and not needed for them to be unblocked

Thanks

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.