No torch and torchaudio package available for CUDA 12.6 (Jetpack 6.2)

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

Hello,

I’m using currently a Jetson Orin Nano installed with Jetpack 6.2 with CUDA 12.6.
I need to install torch and torchaudio for Speaker Diarization and the requirements for the Jetson are quite challenging.

I have already installed the latest versions according to this post PyTorch for Jetson but that post only goes until CUDA 12.4. This means that when I install torch and torchaudio, the CUDA is not available.

Are there already compatible versions of torch and torchaudio for Jetson’s using CUDA 12.6?
Thanks.

Tiago

4

Hi,

Please visit https://pypi.jetson-ai-lab.io/jp6/cu126 to obtain the prebuilt library

Thanks

5

Hi @DavidDDD,

Thanks for the help. I did that and it worked without incompatibilities. Unfortunately I will have to downgrade everything (JetPack + CUDA + Torch + Torchaudio) to the previous (CUDA 12.2) because the performance of the applications I have was really impacted.
I’m using speechbrain to perform a realtime speaker recognition of a conversation using the mic. In the previous version, he could detect speaker changes in about 2 sec, which was totally accepted. Right now, he is taking 10 seconds, which removes completely the realtime capability.

I believe this might be related with the changes torch and torchaudio are going through now, so I will downgrade from torch 2.8 (CUDA 12.6) to torch 2.3 (CUDA 12.2).

Let me know if you have any ideas or knowledge why this might be happening.

Thanks for the support.
Tiago

6

Hi,

Please file a new topic since it’s not related to original issue.

Thanks

7

Hi @DavidDDD!
Don’t need to do it. I just had to update part of my code to the new torch. It seems he was not using the GPU even though he detected.

Thanks for the help!
Regards

8

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