I have to use pytorch v1.5.0 for my code which is only compatible with Jetpack 4.4DP. (PyTorch for Jetson)
There is no option to select this version with the latest SDK Manager. I’ve try to use the cli but I can’t find the correct version that correspond to Jetpack 4.4DP.
What I’ve tried:
sdkmanager --cli downloadonly --logintype devzone --product Jetson --version 4.4DP --targetos Linux --target P2888-0004
Is it possible to install Jetpack4.4DP with this method? or is there any other ways to install pytorch v1.5.0 and torchvision v0.6.0 with other later Jetpack version?
The listing of all available JetPack releases is here:
https://developer.nvidia.com/embedded/jetpack-archive
Note that the “DP” stands for “Developer Preview”, and that the 4.4 release supersedes the 4.4DP release, so you might find anything “DP” has gone away from newer installers.
so is there other way to install 4.4 DP on AGX Xavier?
If SDKM does not want to do this, then you could download the “driver package” and “sample root filesystem” yourself, unpack the driver package without sudo, and then unpack the sample rootfs into “Linux_for_Tegra/rootfs/” with sudo. Then cd back to “Linux_for_Tegra/”, and run “sudo ./apply_binaries.sh”. This will set you up for command line flash of the DP. Actual flash would then be:
sudo ./flash.sh jetson-xavier mmcblk0p1
This is the same thing which would have been flashed by JetPack/SDKM. The difference is that now you’ll have to manually get and install the “extras”, e.g., CUDA.
FYI, I think this is the L4T for the DP, but I’m not sure if this is actually DP:
https://developer.nvidia.com/embedded/linux-tegra-r32.4.2