Can I modify the dockerfile or config file for the transfer learning toolkit provided by nvidia?

Hello, I’d like to train with the tlt provided by nvidia.
But there was an error in the tlt-train command, as shown in the following link.

Currently, the CUDA of the Docker is version 10, so can I modify it arbitrarily?

Please Answer me.

Sorry, I am afraid not.

Then, could you tell me the nvidia graphics card that works perfectly in the current tlt version? I beg you.

See CUDA GPUs - Compute Capability | NVIDIA Developer
I think sm_7.5 or a bit lower will be ok. Such as GeForce RTX 2080, GeForce GTX 1080 Ti, T4, V100, etc.

Thank you. I will refer to it

Is it possible to apply for the rtx2080 super?

Yes, see my comments above.

I wonder, is there any difference in architecture between rtx2080 and rtx2080 super? sorry…

The rtx2080 and rtx2080 super are powered by the same NVIDIA Turing™ architecture.

See more in