Hello all,
I am trying to deploy more than 1 deep learning models, mainly to understand how things work. So, i.e. I am trying to deploy both mobilenet and deeplab.
I do follow the docs (https://docs.modalai.com/voxl-tflite-server/), however I am unable to make both of models to work in the same time.
These are my approaches:
Approach 1:
Changed the file /etc/modalai/voxl-tflite-server.conf and set
'allow_multiple': true and changed the output_pipe_prefix.
I run the command voxl-confure-tflite. I selected:
No factory reset,
enable voxl-tfliteserver and
camera front stereo and
selected the mobilenet.
The service is running and I can see it through portal.
Consequently, I edited again the /etc/modalai/voxl-tflite-server.conf, to change the prefix. Did the same, but selected the deeplab model. Regarding cameras, I tried both the same (stereo front) and different (hires,tracking).
The deeplab is running, by replacing the mobilenet, every time.
Approach 2:
I created distinct /etc/modalai/voxl-tflite-server.conf files, named
/etc/modalai/voxl-tflite-server-mobilenet.conf and /etc/modalai/voxl-tflite-server-deeplab.conf,
each one with relevant configurations (allow multiple, different prefix, correct model path for each model, and input_pipe /run/mpa/tracking and /run/mpa/stereo/ accordingly).
I created a service for each, by copying and altering the file /etc/systemd/system/voxl-tflite-server.service, by adding the corresponding path to each conf.
Changed the description, the PIDFile (i.e. /run/voxl-tflite-server-mobilenet.pid) and the ExecStart line to point the corresponding file
(i.e. /bin/bach -c "/usr/bin/voxl-tflite-server --config /etc/modalai/voxl-tflite-server-mobilenet.conf").
I enabled and started the service. Same for both.
Still they are not working simultaneously. To be precise, although I was able to run via this way a service, now they are failing no matter what I do.
Our goal is to finally run multiple models simultaneously both pre-installed and custom models.
Could someone please help me on how to make it run multiple models, what I am doing wrong, or what is the correct actions?