SDXL does not generate when using DDIM with directML backend

[sleppyrobot]

It appears that DDIM does not work with SDXL and direct ML.

I was trying to use someone else's optimized workflow but could not.

Not really a big deal, works with other samplers, just wanted to test out this method.

Output will be ignored 100%|██████████████████████████████████████████████████████████████████████████████████| 24/24 [00:51<00:00, 2.16s/it] 100%|████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:38<00:00, 4.80s/it] Prompt executed in 103.00 seconds got prompt Failed to validate prompt for output 160:

• VAEDecodeTiled 166:
  • Required input is missing: samples Output will be ignored DDIM Sampler: 0%| | 0/24 [00:00<?, ?it/s]

[aa-parky]

Windows 11
RX 7900XTX
Python --version 3.10.6
pip 23.2.1
win-pyenv 3.1.1

I have noticed this too, DDIM just hangs.

I'm guessing this comes down to DirectML and Cuda differences, also not a big deal for me too. May be of interest for the devs.

The following samplers fail for me with errors:

• DPM_Fast
• DPM_Adaptive
• DPMPP_SDE_GPU
• DPMPP_2M_SDE_GPU

The following work, with some artifacts and some warnings in the console:

• DPMPP_SDE

Happy to post the errors and warning I get if of value.