> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/openai/whisper/llms.txt
> Use this file to discover all available pages before exploring further.

# Installation

> Set up Whisper on your system

## Prerequisites

Before installing Whisper, ensure your system meets these requirements:

* **Python**: Version 3.8 or higher (3.8-3.13 supported)
* **PyTorch**: Recent versions (developed with PyTorch 1.10.1+)
* **FFmpeg**: Required for audio processing

## Installation Steps

<Steps>
  <Step title="Install Whisper via pip">
    The easiest way to install Whisper is via pip. This will install the latest stable release:

    ```bash theme={null}
    pip install -U openai-whisper
    ```

    <Tip>
      The `-U` flag ensures you get the latest version if Whisper is already installed.
    </Tip>

    ### Alternative: Install from GitHub

    To get the latest development version directly from the repository:

    <CodeGroup>
      ```bash Fresh Install theme={null}
      pip install git+https://github.com/openai/whisper.git
      ```

      ```bash Update Existing theme={null}
      pip install --upgrade --no-deps --force-reinstall git+https://github.com/openai/whisper.git
      ```
    </CodeGroup>
  </Step>

  <Step title="Install FFmpeg">
    FFmpeg is required for audio file processing. Install it using your system's package manager:

    <CodeGroup>
      ```bash Ubuntu/Debian theme={null}
      sudo apt update && sudo apt install ffmpeg
      ```

      ```bash Arch Linux theme={null}
      sudo pacman -S ffmpeg
      ```

      ```bash macOS (Homebrew) theme={null}
      brew install ffmpeg
      ```

      ```bash Windows (Chocolatey) theme={null}
      choco install ffmpeg
      ```

      ```bash Windows (Scoop) theme={null}
      scoop install ffmpeg
      ```
    </CodeGroup>

    <Note>
      Make sure FFmpeg is accessible from your command line by running `ffmpeg -version` after installation.
    </Note>
  </Step>

  <Step title="Verify Installation">
    Verify that Whisper is installed correctly:

    ```bash theme={null}
    whisper --help
    ```

    You can also verify the installation in Python:

    ```python theme={null}
    import whisper
    print(whisper.available_models())
    ```

    This should display a list of available model names without errors.
  </Step>
</Steps>

## Dependencies

Whisper automatically installs the following Python dependencies:

* **more-itertools**: Utilities for working with iterables
* **numba**: JIT compiler for numerical functions
* **numpy**: Numerical computing library
* **tiktoken**: Fast tokenizer implementation by OpenAI
* **torch**: PyTorch deep learning framework
* **tqdm**: Progress bar library
* **triton**: GPU programming (Linux x86\_64 only)

## Troubleshooting

<AccordionGroup>
  <Accordion title="Rust installation errors">
    If you see installation errors related to `tiktoken`, you may need to install Rust:

    1. Visit the [Rust installation page](https://www.rust-lang.org/learn/get-started)
    2. Follow the installation instructions for your platform
    3. Configure your PATH:
       ```bash theme={null}
       export PATH="$HOME/.cargo/bin:$PATH"
       ```
    4. Retry the Whisper installation
  </Accordion>

  <Accordion title="Missing setuptools_rust">
    If installation fails with `No module named 'setuptools_rust'`, install it explicitly:

    ```bash theme={null}
    pip install setuptools-rust
    ```

    Then retry installing Whisper.
  </Accordion>

  <Accordion title="FFmpeg not found">
    If you get an error that FFmpeg is not found:

    1. Verify FFmpeg is installed: `ffmpeg -version`
    2. Ensure FFmpeg is in your system PATH
    3. On Windows, you may need to restart your terminal or add FFmpeg to PATH manually
  </Accordion>

  <Accordion title="CUDA/GPU issues">
    Whisper will automatically use CUDA if available. To verify GPU support:

    ```python theme={null}
    import torch
    print(f"CUDA available: {torch.cuda.is_available()}")
    print(f"CUDA device: {torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'N/A'}")
    ```

    If CUDA is not detected but you have an NVIDIA GPU, reinstall PyTorch with CUDA support following the [PyTorch installation guide](https://pytorch.org/get-started/locally/).
  </Accordion>

  <Accordion title="SHA256 checksum errors">
    If you encounter SHA256 checksum errors when downloading models:

    1. Delete the cached model files in `~/.cache/whisper`
    2. Retry loading the model
    3. Check your internet connection for stability

    The models are downloaded on first use and cached locally for future use.
  </Accordion>
</AccordionGroup>

## System Requirements by Model

Ensure your system has sufficient VRAM for your chosen model:

| Model  | VRAM Required | Best For                              |
| ------ | ------------- | ------------------------------------- |
| tiny   | \~1 GB        | Fast transcription, limited resources |
| base   | \~1 GB        | Balanced speed and accuracy           |
| small  | \~2 GB        | Better accuracy, moderate speed       |
| medium | \~5 GB        | High accuracy, translation tasks      |
| large  | \~10 GB       | Maximum accuracy                      |
| turbo  | \~6 GB        | Fast with high accuracy (recommended) |

<Warning>
  CPU inference is supported but will be significantly slower than GPU inference, especially for larger models.
</Warning>

## Next Steps

<Card title="Quickstart Guide" icon="rocket" href="/quickstart">
  Start using Whisper with practical examples for both CLI and Python API
</Card>