Not compatible with new GPUs (RTX 5070 ti)

[Iliocostalis]

Was someone able to get the CUDA version running on the RTX 50xx GPUs?
I tried a lot of things, but in the end I need CUDA >12.x or else my GPU isn't supported. But the project needs torch <2.x and I couldn't find this combination.

[qinmingdong]

torch2.5.1cu124（），This is the maximum supported version, and any higher versions are no longer supported，This is the maximum supported version, and any higher versions are not supported. I tried Python 3.8-3.11 (requirements. txt needs to be modified to be compatible with Python 3.11 and Torch 2.5.1 CU124), but they are all supported. Python 3.12 and above have not been tested
........................................................................................................
Environment Python 3.11 graphics card 4060ti 16G
cmd command：

D:\python\pycharm-project\remove-watermark\lama>activate lama311
(lama311) D:\python\pycharm-project\remove-watermark\lama>python %cd%\bin\predict.py model.path=%cd%\big-lama indir=%cd%\LaMa_test_images outdir=%cd%\output_images

info:
[2025-11-01 09:43:33,108][saicinpainting.training.trainers.base][INFO] - BaseInpaintingTrainingModule init done
D:\python\pycharm-project\remove-watermark\lama\saicinpainting\training\trainers_init_.py:27: FutureWarning: You are using torch.load with weights_only=False (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for weights_only will be flipped to True. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via torch.serialization.add_safe_globals. We recommend you start setting weights_only=True for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
state = torch.load(path, map_location=map_location)
[2025-11-01 09:43:35,030][saicinpainting.training.data.datasets][INFO] - Make val dataloader default from D:\python\pycharm-project\remove-watermark\lama\LaMa_test_images/
23%|███████████████████▏ | 11/47 [00:06<00:13, 2.60it/s]