Received my Tachyon last week and flashed it with headless Linux, following the “Setting up your Tachyon” guide.
Currently, I don’t have any sensors, screens or anything else connected. I have tried to use the various methods to deploy a Linux container to the device, but with no success. This is mainly due to lack of experience on my side.
I do have experience with small computers like the NanoPi. For the NanoPi, I have built Python code that runs as a service and collects various inputs, from USB-ports and API’s. The solution consolidates the information and sends it to an InfluxDB instance. This is all about home energy: solar, EV charging, etc.
To get started with the Tachyon, I think I’ll try to reuse my Python experience. Are there ways to access to AI accelerator from Python code? Something similar to NVIDIA’s CUDA. Any examples?
The easiest way to get started with AI/ML models is using Qualcomm AI hub. The most common way to deploy these models is using Docker containers, but it has the ability to export ONNX models, which can be run from Python.
Have exported a model from the Qualcomm AI hub. Now, I have a tflite object on my laptop’s disk. Probably a TensorFlow model optimized for the target. No idea how I can make a docker or podman container from it, and also I have difficulties installing podman and docker on the Tachyon. The list of devices shown by the AI hub does not show anything that matches how the Tachyon is operating. Not the right chipset, not Linux as target os, nothing about containers. How is that simple?
Was hoping to be able to install a torch library or something CUDA-like that makes it possible to access the AI accelerator from Python code running on the Tachyon, for example to do benchmarking with matrix multiplications.