Im a complete beginner and have purchased an Orin AGX Developer Kit. Currently I’m going through the NVidia free courses on how to get setup and started.
I have setup Jupyter and its running fine, I been playing with the Hello Camera tutorial, training it to recognise different data sets, all working fine. One part of the code that runs creates a directory in which it stores the pictures I take for training:
Set up the data directory location if not there already
How can I set it up so that it is stored on my NVME drive that I have connected to the Orin? Also, is it best to setup Jupyter to be installed on the NVME instead of the Jetsons internal HDD?
I apologise if these are basic questions, Im a complete newbie but very determined to learn, my end goal is to have my own Robot that runs a local LLM and can do fun AI tasks, like image recognition, speech, etc…
It looks like you follow our tutorial for Jetson Nano.
Since the tutorial is designed for Jetson Nano with JetPack 4, it might not work on the Orin series with JetPack 6.
For a training job on the Orin, storing data on the NVME is good since it’s usually larger.
Suppose it should work by just modifying the path to the NVME mounted location accordingly.
We do have some tutorials about LLM for the Orin series.
You can find it below:
Thank you very much for your reply. I did notice the tutorials are for the Nano but figured id give them a go, all the examples I’ve followed so far seem to work without problems on the Orin, i did come across a couple of problems which i managed to figure out by myself.
That link you provided has literally everything i was looking for in one place, my intention was to run Llama and also be able to implement GPT4. Amazing, thank you!
I’m actually considering running the OS and everything else off the nvme and just boot off that if possible. What would be the best way to format everything and start over with the nvme. Which image would i need to burn onto it for setting up? When i got the Orin i plugged it in and it started booting up automatically to the setup straight away, so i didn’t need to burn anything onto an SD card like i do with my Raspberry Pi, so not sure what I’m looking for exactly. Or would it be easier to just move everything from the Orin onto the Nvme?
I wrote that article, though I didn’t disclose that in my reply . The Orin Nano has a different architecture, it uses a SD card instead of the AGX Orin eMMC. Also, the article is to install JetPack 5, not the current JetPack 6. For JetPack 6 things are somewhat different.
Typically for a development AGX Orin, I flash the eMMC first and get a working system. This gives a backup in case something untoward happens. Some people will format the SSD at this point and use it as extra storage. There are pros and cons to the approach.
Next, if I intend to run the Orin from the SSD, I flash the SSD. Both of these are done using the SDK Manager, generally following the procedure in the video. I tend to flash a minimal system with just the OS, and then install nvidia-jetpack from the command line on the Jetson. Even though this takes more time to setup, and is rather tedious, I have found that it saves more time in the long run when development inevitably causes a catastrophic error.
You can switch from where you are booting using the UEFI boot manager when the Jetson starts up. There are ways to do all this from the command line on the host, but if you can get the SDK Manager to help it is simpler.