Error with Torch on Jetpack 5.1.2: AttributeError in gemma2 Model

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Orin
,
1

Hello,

I am currently using Jetpack 5.1.2 and installed torch through torch-2.1.0a0+41361538.nv23.06-cp38-cp38-linux_aarch64.whl.(PyTorch for Jetson)
However, when I try to run the gemma2 model, I get the following error:

AttributeError: module ‘torch._dynamo’ has no attribute ‘mark_static_address’

The known solution to this error is to upgrade torch, but if I upgrade torch, I cannot use CUDA with torch. I am not sure what to do. Please help.

Also, are there any plans to support official 2.1 rather than the alpha version of Torch 2.1 in 5.1.2?

3

Hi,

Do you need the JetPack 5 environment?
If not, you can upgrade to JetPack 6 which has PyTorch v2.1 prebuilt package already.

We won’t provide further packages for JetPack 5 since we now focus more on the latest JetPack 6.0.
You can still build v2.1 from the source and the steps can be found in the below topic.

Thanks.

5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.