Hello @AastaLLL ,
Thanks for your help.
I have successfully reflashed the OS using the SDK Manager, but I am still facing an issue with the Ollama version being outdated. Is it possible to update the container manually?
After the upgrade, when trying to pull models like llama3.1 or gemma2:2b, I get the following error message:
root@ubuntu:/# ollama run llama3.1
pulling manifest
Error: pull model manifest: 412:
The model you are attempting to pull requires a newer version of Ollama.
Please download the latest version at:
https://ollama.com/download
I also tried pulling the gemma2:2b model with the same result:
root@ubuntu:/# ollama pull gemma2:2b
pulling manifest
Error: pull model manifest: 412:
The model you are attempting to pull requires a newer version of Ollama.
Please download the latest version at:
https://ollama.com/download
When I check the current Ollama version, it is still outdated(but a bit newer than the version before):
root@ubuntu:/# ollama -v
ollama version is 5f7b4a5
Jetpack version is the latest version available in the SDK Manager
cat /etc/nv_tegra_release
# R36 (release), REVISION: 3.0, GCID: 36923193, BOARD: generic, EABI: aarch64, DATE: Fri Jul 19 23:24:25 UTC 2024
# KERNEL_VARIANT: oot
TARGET_USERSPACE_LIB_DIR=nvidia
TARGET_USERSPACE_LIB_DIR_PATH=usr/lib/aarch64-linux-gnu/nvidia
matthias@ubuntu:~$
A bit offtopic but I wanted to answer your question why I don’t like the SDK Manager
Why I Don’t Prefer the SDK Manager and Jetson Linux
I find the SDK Manager to be slow and offering no significant advantage. When compared to operating systems like DietPi or Armbian, the difference is huge. With DietPi or Armbian, I can easily update firmware and even bootloaders directly via the command line interface (CLI), which is much more efficient. In contrast, the SDK Manager feels overly complex and cumbersome to use. For example, there is no straightforward way to transfer the OS from an SD card to an NVMe drive.
Jetson Linux, as well, feels too bloated. A much more minimalistic approach, similar to DietPi, would be more appealing. DietPi offers a lightweight, customizable experience, allowing you to install just the software you need, whereas Jetson Linux comes with a lot of unnecessary features that may not be relevant to all users.
The SDK Manager also lacks options to differentiate between different installation profiles, like a minimal installation without a GUI or the exclusion of standard software like LibreOffice. Instead, it forces users to install a standard set of packages, many of which may not be necessary for all users.
Additionally, the fact that I need to reflash the entire OS to update the Jetpack version is rather impractical. In systems like DietPi, updating everything can be done with a single command. You can also easily make system backups and restore them when needed, which is not as straightforward in the Jetson ecosystem.
End of Support for Jetson Nano
I would reconsider the end of support for the Jetson Nano. It remains the most affordable device in the Jetson lineup, but it offers an outdated software experience. Despite this, the Nano makes a lot of sense for introducing students to the Nvidia Jetson ecosystem, providing a cost-effective entry point for learning AI and edge computing.
While the Jetson Orin Nano is 40 times faster and more advanced, it doesn’t mean that supporting the Jetson Nano should be abandoned. It would cost Nvidia very little to continue supporting their older device, especially when it still holds value in education and hobbyist communities. Even though I would have bought the Orin Nano anyway, keeping support for the Jetson Nano would benefit those looking for a cheaper alternative while still contributing to Nvidia’s ecosystem.