Tensor.item() cannot be called on meta tensors

[heyitsgalaxycreeper]

i have been trying for the last few days to run this on my gaming laptop (which is admittedly quite shitty since it only has a 4GB VRAM GTX 1650), and i've managed to get the webui and ffmpeg working at least. however, when i try to actually get the beatmap to generate ("inference") it gives me this error message:

Exception in thread Thread-22 (_start_server):
Traceback (most recent call last):
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\threading.py", line 1016, in _bootstrap_inner
    self.run()
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\threading.py", line 953, in run
    self._target(*self._args, **self._kwargs)
  File "C:\Users\chris\Mapperatorinator\osuT5\osuT5\inference\server.py", line 406, in _start_server
    model = model_loader()
  File "C:\Users\chris\Mapperatorinator\osuT5\osuT5\utils\model_utils.py", line 183, in model_loader
    model = Mapperatorinator.from_pretrained(
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\transformers\modeling_utils.py", line 4094, in from_pretrained
    model = cls(config, *model_args, **model_kwargs)
  File "C:\Users\chris\Mapperatorinator\osuT5\osuT5\model\modeling_mapperatorinator.py", line 75, in __init__
    self.spectrogram = MelSpectrogram(
  File "C:\Users\chris\Mapperatorinator\osuT5\osuT5\model\spectrogram.py", line 40, in __init__
    self.transform = torchaudio.transforms.MelSpectrogram(
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\torchaudio\transforms\_transforms.py", line 608, in __init__
    self.mel_scale = MelScale(
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\torchaudio\transforms\_transforms.py", line 400, in __init__
    fb = F.melscale_fbanks(n_stft, self.f_min, self.f_max, self.n_mels, self.sample_rate, self.norm, self.mel_scale)
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\torchaudio\functional\functional.py", line 580, in melscale_fbanks
    if (fb.max(dim=0).values == 0.0).any():
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\utils\_device.py", line 109, in __torch_function__
    return func(*args, **kwargs)
  File "C:\Users\chris\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\_meta_registrations.py", line 7779, in meta_local_scalar_dense
    raise RuntimeError("Tensor.item() cannot be called on meta tensors")
RuntimeError: Tensor.item() cannot be called on meta tensors

i've tried uninstalling and reinstalling transformers, checking torch and torchaudio (both versions match) and installing CUDA, but none of these have done anything. what should i do?

[OliBomby]

Are all your dependencies the correct version as in requirements.txt? I've never seen this error before.

[heyitsgalaxycreeper]

assuming 0.8.2 is the correct version of slider, then yes. but i am missing audioop-lts since its python version is specified as 3.13, while i have 3.10 as the readme recommends. though i don't think that's related

[heyitsgalaxycreeper]

fuck i didnt mean to close it

[OliBomby]

Maybe the pytorch version is outdated? Try using Pytorch 2.10.0 with CUDA 13.2. If that's not it I have no clue what's going on.

[heyitsgalaxycreeper]

i have both. i wonder if it's just that my computer is too low-end? i have an ASUS TUF F15 with a 4GB VRAM Nvidia GTX 1650, 8GB of total RAM and an Intel Core i5-10300H CPU.

[heyitsgalaxycreeper]

i don't mean to bump this or to pester, but i've recently uninstalled all of the versions of python on my computer and installed only 3.10 before trying to reinstall mapperatorinator again. i had to modify requirements.txt a bit since just doing "pip install -r requirements.txt" yielded more then a couple of errors, and it still doesn't work.

it would be nice to have some updates on this (if you have any) since i do want to run this locally, assuming that's possible

[OliBomby]

What errors are you getting when installing requirements.txt? There shouldnt be errors there.

[heyitsgalaxycreeper]

installing with a powershell command (Get-Content requirements.txt | ForEach-Object { pip install $_ }) (to avoid it stopping at errors) reveals the following:

"ERROR: No matching distribution found for accelerate==1.12.0
ERROR: No matching distribution found for transformers==4.57.3
ERROR: No matching distribution found for tensorboard==2.20.0
ERROR: No matching distribution found for matplotlib==3.10.8
ERROR: No matching distribution found for pandas==2.3.3
ERROR: No matching distribution found for pyarrow==22.0.0
ERROR: No matching distribution found for lightning==2.6.0
ERROR: No matching distribution found for flask==3.1.2
ERROR: No matching distribution found for peft==0.18.1
ERROR: No matching distribution found for datasets==4.8.3
ERROR: No matching distribution found for torchcodec==0.10.0"

on python 3.10

[OliBomby]

These packages and versions do exist on PyPI. Is this using the wrong or outdated pip? or wrong package index?

what is your output of these commands?

python --version
python -m pip --version
pip --version
where python
where pip

[heyitsgalaxycreeper]

well running the version helped me figure the problem, which was that i was running python 3.8.5 (i had miniconda on my computer) instead of python 3.10. after fixing that everything installs fine and i'm able to generate maps now.
thank you for the help!

[OliBomby]

Awesome