bugfix: Float64 error for mps devices on set_timesteps

[ZachNagengast]

What type of PR is this? (check all applicable)

• Refactor
• Feature
• Bug Fix
• Optimization
• Documentation Update
• Community Node Submission

Have you discussed this change with the InvokeAI team?

• Yes
• No, because: minor fix, let me know your thoughts

Have you updated all relevant documentation?

• Yes
• No

Description

Related Tickets & Documents

• Related Issue # [bug]: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead. #4017
• Closes #

QA Instructions, Screenshots, Recordings

Added/updated tests?

• Yes
• No : Requires mps device

[optional] Are there any post deployment tasks we need to perform?

Please test on an MPS (M1/M2) device.

Relevant code causing the error in #4017

https://github.com/huggingface/diffusers/blob/01b6ec21faf2dce3373238b12eb450030ab1f318/src/diffusers/schedulers/scheduling_euler_discrete.py#L263C3-L268C75

        self.sigmas = torch.from_numpy(sigmas).to(device=device)
        if str(device).startswith("mps"):
            # mps does not support float64
            self.timesteps = torch.from_numpy(timesteps).to(device, dtype=torch.float32)
        else:
            self.timesteps = torch.from_numpy(timesteps).to(device=device)

[Millu]

Just tested this on mine - 2020 M1 w/ 16GB of RAM and it is working as expected.

HUGE shoutout to @psychedelicious for the help in setting up the local dev environment

[psychedelicious]

Would like approval from @StAlKeR7779 for this one.

[lstein]

On a Linux CUDA system, running with this PR gives me this for "banana sushi", SDXL-base-1.0, and 1024x1024 using the Euler scheduler:

Same settings on main give me this:

[lstein]

With @StAlKeR7779 's help, I committed a small change that fixes the issue on Linux systems. This should be tested again on MPS systems.

[ZachNagengast]

I'm not sure why the commit isn't showing up here, but it just needs a quick black . It's showing up now

ZachNagengast@2164674

[Millu]

@ZachNagengast are you in the Discord?