[bug]: PixelWave Flux.1-dev 03 NF4 quant stopped working (regression from 5.3.x?)

[dsent]

Is there an existing issue for this problem?

• I have searched the existing issues

Operating system

Windows

GPU vendor

Nvidia (CUDA)

GPU model

RTX 4090

GPU VRAM

24GB

Version number

5.4.2

Browser

Edge 131.0.2903.51

Python dependencies

{
"accelerate": "1.0.1",
"compel": "2.0.2",
"cuda": "12.4",
"diffusers": "0.31.0",
"numpy": "1.26.4",
"opencv": "4.9.0.80",
"onnx": "1.16.1",
"pillow": "11.0.0",
"python": "3.11.9",
"torch": "2.4.1+cu124",
"torchvision": "0.19.1+cu124",
"transformers": "4.41.1",
"xformers": null
}

What happened

Image generation fails with Error(s) in loading state_dict for Flux.
The output: Queue Item.json

What you expected to happen

Image generation succeeds 🙂

How to reproduce the problem

Start the app, add NF4 quant of PixelWave FLUX.1-dev 03, select it in the generation interface, enter a prompt, click "Invoke".

Additional context

I've used this model/quant with Invoke successfully, but after updating to 5.4.2, the workflow is broken.
FP8 and GGUF quants work, but I like them less (and they take more memory).

Discord username

No response

[dsent]

Not really sure about this, but it also seems that FP8 and GGUF quants became an order of magnitude slower. FP8 generation of a single image takes 250s. It wasn't that slow before.

[dsent]

Not sure if the root cause is the same, but basically all Flux-based models (except Flux itself, installed from Invoke Starter Models) stopped working. They give various errors (mostly out-of-memory) and fail to generate.
The same models worked well with 5.3, and still work without problems with e.g. Forge.