When running app.py and generating on Mac, an error occurs

[harmlessman]

When I run app.py and try to generate on a Mac, I encounter the following error.

My environment setup steps were:

git clone https://github.com/nari-labs/dia.git
cd dia
python -m venv .venv
source .venv/bin/activate
pip install -e .
python app.py

After following the steps above and running the code, the following error occurs:

To create a public link, set `share=True` in `launch()`.
loc("mps_matmul"("(mpsFileLoc): /AppleInternal/Library/BuildRoots/d187755d-b9a3-11ef-83e5-aabfac210453/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm":43:0)): error: incompatible dimensions
loc("mps_matmul"("(mpsFileLoc): /AppleInternal/Library/BuildRoots/d187755d-b9a3-11ef-83e5-aabfac210453/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm":43:0)): error: invalid shape
LLVM ERROR: Failed to infer result type(s).
zsh: abort      python3 app.py
(.venv) (base) harmlessman@harmlessmanui-MacBookPro dia % /opt/anaconda3/lib/python3.12/multiprocessing/resource_tracker.py:254: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

If anyone knows the cause of this issue, I would appreciate your help.

[coolcoastcat]

same issue on Mac Mini M4

[vit0-9]

can confirm, M1 Max processor same error

error: invalid dimensions
error: invalid shape

[kevincobain2000]

Have same error on M4 48GB Mac Pro

[harmlessman]

I try this solution, but didn't work.. (on mac m4 pro)
#124 (comment)

[TeoConan]

Same issue on M1 air (not rich as you are guys)

[felipeff]

It took me a while to figure out, but the Python venv setup for the project installs Pytorch 2.6.0.
I had to run:
pip3 install torch torchaudio torch-stoi --upgrade
to upgrade to the latest version 2.7.0.
After that, I was able to run the code

[kevincobain2000]

Thanks @felipeff
Did that work with Gradio too?
I believe your torch and the once on the comment below are similar type of solution.
For me on #79 (comment) has worked but just on CLI, not on Gradio.

[felipeff]

@kevincobain2000 both the cil and gradio work for me after upgrading torch

[donkodimov]

@felipeff did you change any settings after the upgrade to torch 2.7.0?
I get an error:

Error during inference: probability tensor contains either `inf`, `nan` or element < 0
Traceback (most recent call last):

        text_input,
    ...<6 lines>...
        audio_prompt=prompt_path_for_generate,
    )

[felipeff]

@donkodimov nope, didn't change anything else, sorry

[Goactivemedia]

I am on Mac M4 MacStudio.
I have it installed using Pinokio. The interface comes up but I cannot get it to work. I just get an error anytime I try to do anything.

[xpatronum]

torchaudio==2.6.0
Upon generating error occurs:

Inference failed: Output channels > 65536 not supported at the MPS device.

❗️Turns out that during F.conv1d is not supported yet on MPS. See issue

[robbiemu]

wondering if this works for anyone else dealing with this issue?

    compute_dtype = torch.float16 if self.device == "cuda" else torch.float32

    self.model = Dia.from_pretrained(
        model_checkpoint, compute_dtype=compute_dtype, device=self.device
    )

I am working on a for of dia where I changed libraries in a few places and and getting ready to replace the encoder to see how it sounds because this one is unmaintained. So Im not sure if it works in the general case, but it worked for me (I also changed how this variable is used in model.py, getting rid of .to_dtype() calls on it.. not sure why that is okay in the source for anyone)

edit: FWIW, in dia I changed the two to_dtype()'s as indicated and the imports:

    "gradio~=5.42.0",
    "huggingface-hub~=0.34.4",
    "pydantic~=2.11.7",
    "safetensors~=0.6.2",
    "soundfile~=0.13.1",
    "torch~=2.8.0",
    "torchaudio~=2.8.0",