Enable MPS backend for bitsandbytes quantization

Is your feature request related to a problem? Please describe.

Bitsandbytes now has basic support for the Apple MPS backend, as I can tell by bitsandbytes-foundation/bitsandbytes#1818 and
bitsandbytes-foundation/bitsandbytes#1875.

The issue is that diffusers does not allow me to use the quantization on Apple hardware, because of the error No GPU found. A GPU is needed for quantization. from here

diffusers/src/diffusers/quantizers/bitsandbytes/bnb_quantizer.py

Lines 64 to 65 in f2be8bd

    
           if not (torch.cuda.is_available() or torch.xpu.is_available()): 
        
               raise RuntimeError("No GPU found. A GPU is needed for quantization.")

.

Describe the solution you'd like.

I modified the above lines to

if not (torch.cuda.is_available() or torch.xpu.is_available() or torch.mps.is_available()):
    raise RuntimeError("No GPU found. A GPU is needed for quantization.")

and tested the change with the quantized version of FLUX.2-dev as described in https://github.com/black-forest-labs/flux2/blob/main/docs/flux2_dev_hf.md#4-bit-transformer-and-4-bit-text-encoder-20g-of-vram, and all worked fine.

I am wondering if that solution could be adopted in the repository. I can open a PR if that is the case.

Describe alternatives you've considered.

None.

Additional context.
Add any other context or screenshots about the feature request here.

Hi, I've just had a go a getting ideogram working and it required the change flagged above and a change to update_device_map

    def update_device_map(self, device_map):
        if device_map is None:
            if torch.xpu.is_available():
                current_device = f"xpu:{torch.xpu.current_device()}"
            elif torch.mps.is_available():
                current_device = "mps"
            else:

I'd really like to see the mps restriction lifted without me having to change the code

@sayakpaul

cc: @sayakpaul WDYT?

Sure why not.

@LucasSte

@LucasSte are you still planning to PR it? I'd be happy to take this

@LucasSte

@LucasSte are you still planning to PR it? I'd be happy to take this

Now that someone has approved the change, I'll open a PR.

The PR is here: #13915

Navigation Menu

Repository navigation

Enable MPS backend for bitsandbytes quantization #13361

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	if not (torch.cuda.is_available() or torch.xpu.is_available()):
	raise RuntimeError("No GPU found. A GPU is needed for quantization.")

Enable MPS backend for bitsandbytes quantization #13361

Description

Activity

Vargol commented on Jun 9, 2026

asomoza commented on Jun 9, 2026

sayakpaul commented on Jun 10, 2026

akshan-main commented on Jun 10, 2026

LucasSte commented on Jun 10, 2026

LucasSte commented on Jun 10, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions