TypeError: eager() got an unexpected keyword argument 'mode'

[KutayDemiray]

Describe the bug I am trying to train a NeRF with my image data (that I ran through colmap), following the instructions in this page. Running ns-train nerfacto --data {PROCESSED_DATA_DIR} yields TypeError: eager() got an unexpected keyword argument 'mode'. I have also tried this with mipnerf and instant-ngp and received the same error.

To Reproduce Steps to reproduce the behavior:

1. Use colmap to process data as instructed (no problems here)
2. Run ns-train nerfacto --data {PROCESSED_DATA_DIR}
3. See error

Expected behavior Model training without issues (hopefully)

Additional context I am using the following:

• Ubuntu 20.04 LTS
• CUDA 12.2
• Python 3.10.13
• Pytorch 2.1.1
• Nerfstudio 0.3.4

Here is the gist of full error: https://gist.github.com/KutayDemiray/3df91a2b60ce91edf726aa0072ca0bb7

[HeaIn]

I've made a temporary adjustment by transforming errors into warnings, achieved by adding the following code in this location.

import torch._dynamo
torch._dynamo.config.suppress_errors = True

I'm aware that this isn't the ideal solution. Hope someone with a better fix will come forward.

[lw112]

I got the same warning, I think it's because essentially cuda 12.2 is not supported. I downgraded to cuda 11.8, now the warning is gone.

[liuyi0316]

i checked my version of cuda with 11.8, but still faced the same problem