'compute index ranges weights" not implemented for "Half

[thefirstLeonliao]

No description provided.

[elsung]

I have the same issue with my mac m2 ultra! basically this wrapper doesnt work at all with MacOS / MPS. i've tried the solution that some others have posted saying to run it in fp32, that also breaks. PC users can sorta make this work with 1 24GB GPU, but Mac users could really use some guidance here given the much higher upperlimit for the ram on macs.

details:

when running at fp16
it wont even render any frames, and we get the error listed in this issue:
"compute_index_ranges_weights" not implemented for 'Half'

If running at fp32
it will render all of the frames in the toonCrafter Interpolation node, but we get an error as it finishes and it says:
Assertion failed: (0 <= mpsAxis && mpsAxis < 4 && "Runtime canonicalization must simplify reduction axes to minor 4 dimensions."), function getKernelAxes, file GPUReductionOps.mm, line 31.

Note: i tested on python 3.12.2 and also 3.12.0, neither works. also am running the nightly-cpu build of torch
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu

also had tried to roll back for torch like some people have said to do, but got the same results as above.

[jdavid82]

same problem :( M4 Pro chip with 24GB unified memory

[rohitsainier]

same here

[dangarfield]

Any solutions for this? M4 mac here