Looking for correct version of pytorch for RTX 4090, CUDA 12.2

vision
1

image
This is my pytorch version and my GPU is “NVIDIA GeForce RTX 4090”

When I try to train my model, Kernel dies with the below output
C:\actions-runner_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\cuda\IndexKernel.cu:93: block: [62,0,0], thread: [115,0,0] Assertion index >= -sizes[i] && index < sizes[i] && "index out of bounds" failed.
Need help !!

2

The error message points towards an indexing error and is unrelated to the used PyTorch binary.
Narrow down the failing indexing operation and fix it.

3

Yes, indeed it was.
Fixed it. Thanks.

4

How to build pytorch-cuda=12.2 ? my GPU is NVIDIA GeForce RTX 4090 and I have nvidia toolkit 12.2 installed on my windows 11.

5

You would need to build from source using your locally installed CUDA toolkit. Alternatively, you can download the nightly binaries using CUDA 12.1 Update1.

6

Thank you. I couldn’t find the nightly build for 12.2.