Having an issue with installing on M1 Mac

[asimpletune]

After the pip install and running whisper --help I get:

Traceback (most recent call last):
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/utils/import_utils.py", line 1031, in _get_module
    return importlib.import_module("." + module_name, self.__name__)
  File "/opt/homebrew/Cellar/[email protected]/3.9.14/Frameworks/Python.framework/Versions/3.9/lib/python3.9/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 972, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1030, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1007, in _find_and_load
  File "<frozen importlib._bootstrap>", line 986, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 680, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 850, in exec_module
  File "<frozen importlib._bootstrap>", line 228, in _call_with_frames_removed
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/models/__init__.py", line 19, in <module>
    from . import (
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/models/mt5/__init__.py", line 33, in <module>
    from ..t5.tokenization_t5 import T5Tokenizer
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/models/t5/tokenization_t5.py", line 26, in <module>
    from ...tokenization_utils import PreTrainedTokenizer
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/tokenization_utils.py", line 26, in <module>
    from .tokenization_utils_base import (
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/tokenization_utils_base.py", line 70, in <module>
    from tokenizers import AddedToken
  File "/opt/homebrew/lib/python3.9/site-packages/tokenizers/__init__.py", line 79, in <module>
    from .tokenizers import (
ImportError: dlopen(/opt/homebrew/lib/python3.9/site-packages/tokenizers/tokenizers.cpython-39-darwin.so, 0x0002): tried: '/opt/homebrew/lib/python3.9/site-packages/tokenizers/tokenizers.cpython-39-darwin.so' (mach-o file, but is an incompatible architecture (have (x86_64), need (arm64e)))

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/opt/homebrew/bin/whisper", line 5, in <module>
    from whisper.transcribe import cli
  File "/opt/homebrew/lib/python3.9/site-packages/whisper/__init__.py", line 12, in <module>
    from .decoding import DecodingOptions, DecodingResult, decode, detect_language
  File "/opt/homebrew/lib/python3.9/site-packages/whisper/decoding.py", line 11, in <module>
    from .tokenizer import Tokenizer, get_tokenizer
  File "/opt/homebrew/lib/python3.9/site-packages/whisper/tokenizer.py", line 8, in <module>
    from transformers import GPT2TokenizerFast
  File "<frozen importlib._bootstrap>", line 1055, in _handle_fromlist
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/utils/import_utils.py", line 1021, in __getattr__
    module = self._get_module(self._class_to_module[name])
  File "/opt/homebrew/lib/python3.9/site-packages/transformers/utils/import_utils.py", line 1033, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import transformers.models.gpt2 because of the following error (look up to see its traceback):
dlopen(/opt/homebrew/lib/python3.9/site-packages/tokenizers/tokenizers.cpython-39-darwin.so, 0x0002): tried: '/opt/homebrew/lib/python3.9/site-packages/tokenizers/tokenizers.cpython-39-darwin.so' (mach-o file, but is an incompatible architecture (have (x86_64), need (arm64e)))

Not exactly sure but I wanted to post this and get the discussion started before I start digging deeper.

(btw, I have an M1 MB Air, running MacOS 12.6)

[jongwook]

This seems to be caused by mixing up ARM and x86_64 binaries, specifically for the tokenizers module. There could be many ways to fix this, but installing Anaconda Python and using that instead of Homebrew's could be an easy solution.

[asimpletune]

Thanks for the reply. I figure the issue was with the binaries, but I'm not sure how to direct whisper to use which dependencies. I guess I can try just uninstalling the tokenizer module completely and using anaconda to download the ARM specific one.

[dsiroker]

On a fresh install of macOS Ventura on a Apple M1 Ultra here are the steps I had to take to get it working:

1. pip install virtualenv
2. virtualenv env/
3. source env/bin/activate
4. brew install rust
5. pip install git+https://github.com/openai/whisper.git

[asimpletune]

Interesting. I tried those same steps and had the same issue. I wonder if the big difference is I’m on Monterey vs Ventura.

[dogcalas]

Installing rust solve my problem!!. Thanks.
I am on M1 Pro with Monterrey.

[dipperLyra]

Installing Rust did it. Thanks a million for this contribution!

[astrowonk]

If one doesn't want to install rust and compile, conda-forge has a M1 apple/arm binary of tokenizers per my post below.

[astrowonk]

I'd advise installing tokenizers not from pip but from conda-forge. They have an ARM mac binary. I had a similar crash (and I even tried to install rust compiler, but pip wasn't finding it) so it was simpler to just (since I run python from miniforge anyway) do mamba install tokenizers before installing whisper. (or conda install tokenizers)

What python environment are you using on your Mac @asimpletune? I'm on Monterrey as well.

[mbaykara]

pip install whisper-mps works like a charm