whisper doesn't work on colab

[sherbertzz]

Hi, i've been using whisper for few days now, it worked perfectly. Since yesterday it doesnt work, im using google colab, installing with this

!pip install git+https://github.com/openai/whisper.git
!sudo apt update && sudo apt install ffmpeg

After installation is complete im using this command
!whisper "fff.mp3" --model medium --language pl --word_timestamps True --max_line_width 15 --max_words_per_line 1

It used to work perfectly but now error pops up.

/usr/local/lib/python3.10/dist-packages/whisper/transcribe.py:474: UserWarning: --max_words_per_line has no effect with --max_line_width
  warnings.warn("--max_words_per_line has no effect with --max_line_width")
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/dist-packages/whisper/transcribe.py", line 478, in cli
    result = transcribe(model, audio_path, temperature=temperature, **args)
  File "/usr/local/lib/python3.10/dist-packages/whisper/transcribe.py", line 323, in transcribe
    add_word_timestamps(
  File "/usr/local/lib/python3.10/dist-packages/whisper/timing.py", line 298, in add_word_timestamps
    alignment = find_alignment(model, tokenizer, text_tokens, mel, num_frames, **kwargs)
  File "/usr/local/lib/python3.10/dist-packages/whisper/timing.py", line 210, in find_alignment
    weights = median_filter(weights, medfilt_width)
  File "/usr/local/lib/python3.10/dist-packages/whisper/timing.py", line 40, in median_filter
    result = median_filter_cuda(x, filter_width)
  File "/usr/local/lib/python3.10/dist-packages/whisper/triton_ops.py", line 107, in median_filter_cuda
    kernel[(grid,)](y, x, x.stride(-2), y.stride(-2), BLOCK_SIZE=BLOCK_SIZE)
  File "<string>", line 63, in kernel
  File "/usr/local/lib/python3.10/dist-packages/triton/compiler/compiler.py", line 425, in compile
    so_path = make_stub(name, signature, constants)
  File "/usr/local/lib/python3.10/dist-packages/triton/compiler/make_launcher.py", line 39, in make_stub
    so = _build(name, src_path, tmpdir)
  File "/usr/local/lib/python3.10/dist-packages/triton/common/build.py", line 61, in _build
    cuda_lib_dirs = libcuda_dirs()
  File "/usr/local/lib/python3.10/dist-packages/triton/common/build.py", line 30, in libcuda_dirs
    assert any(os.path.exists(os.path.join(path, 'libcuda.so')) for path in dirs), msg
AssertionError: libcuda.so cannot found!

Skipping fff.mp3 due to AssertionError: libcuda.so cannot found!

The only command that works is
!whisper "fff.mp3" --model medium --language pl
But I need one word per line so its not helpful for me.

Anyone know what is the problem?
Thanks

[signalprime]

This just happened to me on Colab Pro, using a saved notebook that worked last month and the file was found here:

/usr/local/cuda-11.8/compat/libcuda.so
/usr/local/cuda-11.8/targets/x86_64-linux/lib/stubs/libcuda.so
/usr/lib64-nvidia/libcuda.so

[Woodahs]

It's most likely related to the neweset release. For the time being I'm using the command !pip install git+https://github.com/openai/whisper.git@v20231106 instead of !pip install git+https://github.com/openai/whisper.git for installing the whisper

[DollaJakab]

Same here! Reverted to the previous version you recommended and works like before. Probably just an incompatibility issues, although I don't know on whos end.

[glangford]

This sequence of installs works on colab with the whisper release out today.

!pip install --no-deps "triton==2.0.0"
!pip install --no-deps "tiktoken==0.3.3"
!pip install --no-deps "openai-whisper==20231117"

[7k50]

Thank you, I might try that.
But what is the relation between libcuda.so and these packages? In other words, when installing them, what solution do they provide, or which problem do they fix, exactly?

[glangford]

@7k50 This link describes is the underlying issue with the Colab environment; I am installing triton 2.0 rather than using Colab's preinstalled 2.1 version.

triton-lang/triton#2507

[dustonmouse]

It works for me on Google Colab
!echo /usr/lib64-nvidia/ >/etc/ld.so.conf.d/libcuda.conf; ldconfig

[russkiimike]

It works for me on Google Colab !echo /usr/lib64-nvidia/ >/etc/ld.so.conf.d/libcuda.conf; ldconfig

This worked for me as well on colab

[saadqaisar]

I am using CPU and downgraded to v20231106. But still my issue persists. I dont have GPU. I am giving a simple command whisper filename.mp3 . Please assist as it is highly frustrating.
C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\transcribe.py:115: UserWarning: FP16 is not supported on CPU; using FP32 instead
warnings.warn("FP16 is not supported on CPU; using FP32 instead")
Traceback (most recent call last):
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\transcribe.py", line 478, in cli
result = transcribe(model, audio_path, temperature=temperature, **args)
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\transcribe.py", line 122, in transcribe
mel = log_mel_spectrogram(audio, model.dims.n_mels, padding=N_SAMPLES)
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\audio.py", line 140, in log_mel_spectrogram
audio = load_audio(audio)
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\site-packages\whisper\audio.py", line 58, in load_audio
out = run(cmd, capture_output=True, check=True).stdout
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 503, in run
with Popen(*popenargs, **kwargs) as process:
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 971, in init
self._execute_child(args, executable, preexec_fn, close_fds,
File "C:\Users\ProBook\AppData\Local\Programs\Python\Python310\lib\subprocess.py", line 1440, in _execute_child
hp, ht, pid, tid = _winapi.CreateProcess(executable, args,
FileNotFoundError: [WinError 2] The system cannot find the file specified
Skipping JonesMedical.mp3 due to FileNotFoundError: [WinError 2] The system cannot find the file specified

F:\Downloads\Transcribe>openai-whisper JonesMedical.mp3
'openai-whisper' is not recognized as an internal or external command,
operable program or batch file.

[saadqaisar]

you need to use anaconda on windows. Installed Anaconda, gave this command to install ffmpeg running anaconda as an Administrator in command prompt, and runs like a charm:

conda install -c conda-forge ffmpeg