Force numpy 1.x.x

[bangoga]

I noticed that the current requirements of whisper encourages any version of numpy, hence it defaults to higher versions.

The model is compiled with numpy 1.x.x in mind. I would suggest changing the notebook to force it to be numpy 1.x.x or else anyone requesting transcriptions will be faced with errors

So we go from:

  !pip install --root-user-action=ignore git+https://github.com/openai/whisper.git@v20231117 openai==1.9.0 numpy scipy deepl pydub cohere ffmpeg-python torch==2.1.0 tensorflow-probability==0.23.0 typing-extensions==4.9.0

to:

  !pip install --root-user-action=ignore git+https://github.com/openai/whisper.git@v20231117 openai==1.9.0 numpy~=1.2 scipy deepl pydub cohere ffmpeg-python torch==2.1.0 tensorflow-probability==0.23.0 typing-extensions==4.9.0

[mikezio]

Facing same error here - unable to use the tool

[bangoga]

Facing same error here - unable to use the tool

Making the change I suggested should fix your issue, You can edit the code in the Audio Collab for just yourself and it should work

[MLSukhanov]

I used your suggestion, but the errors didn't go away:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. altair 5.5.0 requires typing-extensions>=4.10.0; python_version < "3.14", but you have typing-extensions 4.9.0 which is incompatible. torchaudio 2.6.0+cu124 requires torch==2.6.0, but you have torch 2.1.0 which is incompatible. torchvision 0.21.0+cu124 requires torch==2.6.0, but you have torch 2.1.0 which is incompatible. google-genai 1.4.0 requires typing-extensions<5.0.0dev,>=4.11.0, but you have typing-extensions 4.9.0 which is incompatible. typeguard 4.4.2 requires typing_extensions>=4.10.0, but you have typing-extensions 4.9.0 which is incompatible.

[MLSukhanov]

Facing same error here - unable to use the tool

Making the change I suggested should fix your issue, You can edit the code in the Audio Collab for just yourself and it should work

A new error has been noticed: ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

[Meleres]

Same problem, doesn't work

[bangoga]

Same problem, doesn't work

Can you please copy paste the full error log, I'll look into the dependency issues after work.

If possible try running the original code and then downgrading the numpy to ~=1.2 after, this way you keep all the original dependancies.

[Zennco23]

Did anybody find a solution to get it working again? If yes, could you please share the modified code?

[MLSukhanov]

Any news on this issue?

[WingedSeal]

I used your suggestion, but the errors didn't go away:

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. altair 5.5.0 requires typing-extensions>=4.10.0; python_version < "3.14", but you have typing-extensions 4.9.0 which is incompatible. torchaudio 2.6.0+cu124 requires torch==2.6.0, but you have torch 2.1.0 which is incompatible. torchvision 0.21.0+cu124 requires torch==2.6.0, but you have torch 2.1.0 which is incompatible. google-genai 1.4.0 requires typing-extensions<5.0.0dev,>=4.11.0, but you have typing-extensions 4.9.0 which is incompatible. typeguard 4.4.2 requires typing_extensions>=4.10.0, but you have typing-extensions 4.9.0 which is incompatible.

also getting error after forcing numpy version the solution is probably having to specify every package's version and slowly downgrading each one of the conflicted package until it works or find the version used originally somehow

[Carleslc]

I'll look into this in the following days. Fixing requirements should work, but maybe it is needed to downgrade more than one dependency. Thank you all for reporting.

[mikezio]

If you're encountering issues with this notebook in Google Colab due to package dependency conflicts (especially around torch, torchaudio, torchvision, typing-extensions, and potentially triton), I recently worked through some of these and found a set of installation commands for "[Step 1] ⚙️ Install the required libraries" that resolved the problems for me.

Replace the existing !pip install --root-user-action=ignore git+https://github.com/openai/whisper.git... line with the corrected lines provided below:

# Define the warning suppressor if you keep it, or remove {NO_ROOT_WARNING} from the line below
NO_ROOT_WARNING = '|& grep -v \\\"WARNING: Running pip as the \\'root\\' user\"'

# 1. Upgrade pip
!pip install --no-warn-script-location --user --upgrade pip {NO_ROOT_WARNING}

# 2. Install Whisper and main dependencies with corrected typing-extensions
!pip install --root-user-action=ignore \
    git+https://github.com/openai/whisper.git@v20231117 \
    openai==1.9.0 \
    numpy \
    scipy \
    deepl \
    pydub \
    cohere \
    ffmpeg-python \
    "typing-extensions>=4.12.0,<5.0.0" \
    tensorflow-probability==0.23.0

# 3. Install a compatible PyTorch stack for Colab
!pip install --root-user-action=ignore torch==2.6.0 torchaudio==2.6.0 torchvision==0.21.0 -f https://download.pytorch.org/whl/cu121/torch_stable.html

VERY IMPORTANT: After this cell finishes, YOU MUST RESTART THE COLAB RUNTIME (Runtime > Restart session)

Note on Triton: You will prob see a warning about openai-whisper requiring triton<3.0.0 while triton 3.2.0 (or similar) is installed by torch 2.6.0. After restarting the runtime, Whisper v20231117 worked fine for me with this newer Triton. If you face Triton-specific errors later, you might need to add !pip install --root-user-action=ignore --force-reinstall triton==2.1.0 at the end of the installations and restart again.

Hope this helps!

[MLSukhanov]

If you're encountering issues with this notebook in Google Colab due to package dependency conflicts (especially around torch, torchaudio, torchvision, typing-extensions, and potentially triton), I recently worked through some of these and found a set of installation commands for "[Step 1] ⚙️ Install the required libraries" that resolved the problems for me.

Replace the existing !pip install --root-user-action=ignore git+https://github.com/openai/whisper.git... line with the corrected lines provided below:

# Define the warning suppressor if you keep it, or remove {NO_ROOT_WARNING} from the line below
NO_ROOT_WARNING = '|& grep -v \\\"WARNING: Running pip as the \\'root\\' user\"'

# 1. Upgrade pip
!pip install --no-warn-script-location --user --upgrade pip {NO_ROOT_WARNING}

# 2. Install Whisper and main dependencies with corrected typing-extensions
!pip install --root-user-action=ignore \
    git+https://github.com/openai/whisper.git@v20231117 \
    openai==1.9.0 \
    numpy \
    scipy \
    deepl \
    pydub \
    cohere \
    ffmpeg-python \
    "typing-extensions>=4.12.0,<5.0.0" \
    tensorflow-probability==0.23.0

# 3. Install a compatible PyTorch stack for Colab
!pip install --root-user-action=ignore torch==2.6.0 torchaudio==2.6.0 torchvision==0.21.0 -f https://download.pytorch.org/whl/cu121/torch_stable.html

VERY IMPORTANT: After this cell finishes, YOU MUST RESTART THE COLAB RUNTIME (Runtime > Restart session)

Note on Triton: You will prob see a warning about openai-whisper requiring triton<3.0.0 while triton 3.2.0 (or similar) is installed by torch 2.6.0. After restarting the runtime, Whisper v20231117 worked fine for me with this newer Triton. If you face Triton-specific errors later, you might need to add !pip install --root-user-action=ignore --force-reinstall triton==2.1.0 at the end of the installations and restart again.

Hope this helps!

I will definitely try your suggestion in the near future and let you know the results. This Jupiter Notebook is a very important educational tool.

[mikezio]

tool.

Were you able to get it to work?

[MLSukhanov]

tool.

Were you able to get it to work?

Yes, but the error with Triton still appears in Google Colab.

[WingedSeal]

anyone got a working solution yet?