Version problems

[michaelotto]

First of all, thanks a lot for this. Is it planned to incorporate the functionality of qwen-image-edit in the future?

Then, the readme doesn't mention that it looks like you have to do git clone https://github.com/ivanfioravanti/qwen-image-mps in the "external" directory.

Then, I couldn't get the rest to work. I have a Mac M1 Pro with 32 GB. I tried with both python 3.9 and 3.12

First, it compains like this:

A module that was compiled using NumPy 1.x cannot be run in
NumPy 2.0.2 as it may crash. To support both 1.x and 2.x
versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.

If you are a user of the module, the easiest solution will be to
downgrade to 'numpy<2' or try to upgrade the affected module.
We expect that some modules will need time to support NumPy 2.

Okay, that's easy to solve, just install numpy<2.0

Next problem when running qwen.py test:

Error: Failed to load model: Failed to import diffusers.pipelines.qwenimage.pipeline_qwenimage because of the following error (look up to see its traceback):
Failed to import diffusers.models.autoencoders.autoencoder_kl_qwenimage because of the following error (look up to see its traceback):
module 'torch.library' has no attribute 'custom_op'
Try: pip install --upgrade diffusers transformers

So it looks like it needs torch>=2.4, but the requirement says 2.0.0. But there does not seem to be torch 2.4.0 for Python 3.12 or 3.9.

So is the requirements.txt wrong? Is my Python version wrong? Any clues to which version of Python can actually run this?

[michaelotto]

Ok, I was able to resolve most of this myself:

As my system evolved from an x86 Mac, it did have x86 brew installed, which installed the x86 versions of Python - and these only support PyTorch up to 2.2

So make sure you have Arm64-Python installed if you run into the same problem. Python 3.11 worked for me.

Then again, the "torch>=2.0.0" requirement is definitely wrong, this needs >=2.4.0. Also, the remark about the external repository needed is still true.

[michaelotto]

It works now, but it takes about 15 minutes per image on my M1 Pro 32 GB and the process uses about 62 GB of RAM. Am I doing something wrong here?