In both 5.5 and 5.6, when using ML Deformers’ Neural Morph Model the pytorch versions (cu121 and cu124) both currently don’t have support for sm_120 that the blackwell series requires. I tried locally upgrading to the nightly build of 121, 124, 128 and 129. With some versions the sm_120 warning goes away, but there is no GPU listed in the training devices.
In certain pytorch version swaps, it’s unable to import the module for the nerual morph model.
Seems like there needs to be a hotfix patch to the latest version of pytorch perhaps with CUDA toolkit 12.9.1 to unlock the use for all the blackwell GPUs.