Flux `uv` command stalls out without errors & `Torch version 2.9.0 has not been tested with coremltools.`

[Bersaelor]

Model name

flux

Command run

Hey, I'm trying to follow the example from the readme exactly, like I did last year for coremltools too.
So I ran the command

uv run coreai.diffusion.export flux2-klein-4b --platform iOS

Skipping import of cpp extensions due to incompatible torch version. Please upgrade to torch >= 2.11.0 (found 2.9.0).
W0627 17:37:59.487000 79452 torch/distributed/elastic/multiprocessing/redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
Torch version 2.9.0 has not been tested with coremltools. You may run into unexpected errors. Torch 2.7.0 is the most recent version that has been tested.
17:38:11 [INFO] coreai_models.diffusion.pipeline: Loading black-forest-labs/FLUX.2-klein-4B (type=flux2, dtype=torch.float16)...
Unable to import `torchao` Tensor objects. This may affect loading checkpoints serialized with `torchao``

Last year I remember similar issues and having to manually hardset my pytorch version to an older version to get around the Torch version 2.9.0 has not been tested with coremltools but I also see that you set "torch==2.9.0" in python /pyproject.toml so maybe thats not an issue I need to worry about?

macOS / iOS target

macOS 26.5.1

Xcode version

Xcode 27 Beta 2

Python / uv version

uv 0.11.25, Python 3.12.7

Full error output

% uv run coreai.diffusion.export flux2-klein-4b --platform iOS
Skipping import of cpp extensions due to incompatible torch version. Please upgrade to torch >= 2.11.0 (found 2.9.0).
W0627 17:37:59.487000 79452 torch/distributed/elastic/multiprocessing/redirects.py:29] NOTE: Redirects are currently not supported in Windows or MacOs.
Torch version 2.9.0 has not been tested with coremltools. You may run into unexpected errors. Torch 2.7.0 is the most recent version that has been tested.
17:38:11 [INFO] coreai_models.diffusion.pipeline: Loading black-forest-labs/FLUX.2-klein-4B (type=flux2, dtype=torch.float16)...
Unable to import `torchao` Tensor objects. This may affect loading checkpoints serialized with `torchao`
Fetching 17 files:   0%|                                                                                                                | 0/17 [00:00<?, ?it/s]

Anything else?

Nothing seems to happen, my internet is working fine but it just sits there at 0%.

[Bersaelor]

I'm wondering whether this is a toolchain issue rather then network, when I abort the call manually, it seems to display some progress:

// ...
Unable to import `torchao` Tensor objects. This may affect loading checkpoints serialized with `torchao`
Fetching 17 files:   0%|                                                                                                                | 0/17 [00:00<?, ?it/s]^x
Fetching 17 files:  18%|██████████████████▏                                                                                    | 3/17 [06:12<28:57, 124.07s/it]
^Z

maybe some networking dependency gets stuck in a weird state instead of sending an error?
(I did let it run for several hours, so it's definitely stuck and not just secretly running with failed text updates)

EDIT:

I was able to execute
uv run coreai.diffusion.export stabilityai/stable-diffusion-3.5-medium
successfully even though it threw all the same warnings about torch versions 2.9 and 2.11.. But it still downloaded and exported the aimodel. Just downloading Flux always gets stuck in the same way.

[kevchengcodes]

Hello! Those warnings can be ignored.

I was unable to reproduce this problem. Maybe clearing the hugging face cache would help?

rm -rf ~/.cache/huggingface/models--black-forest-labs--FLUX.2-klein-4B

Then retry: uv run coreai.diffusion.export flux2-klein-4b --platform iOS --overwrite

I will close this issue but please feel free to file a new issue if it persists for you.

[Bersaelor]

Thank you, clearing the cache worked!

I had also googled it and asked gemini and claude, but they just asked me to check my network or restart my mac.