Skip to content

Use SCALE with Docker#

SCALE offers docker images with a similar tag layout to those offered by NVIDIA.

These serve two main purposes:

  • Deploying the compiler in your CI, to benefit from its improved diagnostics etc.
  • Use as a base image for applications that depend on the SCALE runtime library.

The images are available here:

For example:

# Downloads the latest version of SCALE from Docker Hub that:
# - Imitates CUDA 13.0.2
# - Includes the full SCALE developer toolkit
# - Is based on Ubuntu 24.04
docker pull docker.io/spectralcompute/scale:13.0.2-devel-ubuntu24.04

# Downloads SCALE 1.5.1 from Quay.io that:
# - Imitates CUDA 12.1.0
# - Includes the SCALE runtime, but not the compiler nor other development tools
# - Is based on Ubuntu 22.04
docker pull quay.io/spectral-compute/scale:12.1.0-runtime-ubuntu22.04-1.5.1

Using SCALE with Docker#

To use the container, you need to accept the SCALE License. It can be done by setting an environment variable SCALE_LICENSE_ACCEPT=1 in the container. If you are using docker run, this is what starting bash in the container would look like.

docker run -it -e SCALE_LICENSE_ACCEPT=1 docker.io/spectralcompute/scale:latest

Example: whisper.cpp#

Let's see how you can build and run whisper.cpp using the SCALE Docker image.

This example will use the docker.io/spectralcompute/scale:latest image for simplicity, and will build the most recent version of whisper.cpp at the time of writing: 21411d8. You can find the results of automated testing of SCALE against whisper.cpp on GitHub: spectral-compute/scale-validation.

# 1. Clone the whisper.cpp repository.
git clone https://github.com/ggml-org/whisper.cpp

# 2. Start the container, mount the whisper.cpp repository inside.
#    `--device` flags allow accessing the GPU from the container.
#    See `docker run --help` to learn more about other flags.
docker run -it \
    --mount type=bind,src=$(pwd)/whisper.cpp,dst=/root/whisper.cpp \
    --env SCALE_LICENSE_ACCEPT=1 \
    --device /dev/dri \
    --device /dev/kfd \
    docker.io/spectralcompute/scale:latest

# 3. Inside of the container, activate scaleenv.
#    Replace `gfx1201` with your GPU architecture.
source /opt/scale/bin/scaleenv gfx1201

# 4. Configure the whisper.cpp build tree.
cd /root/whisper.cpp
cmake \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_CUDA_ARCHITECTURES="86" \
    -DGGML_CCACHE=OFF \
    -DGGML_CUDA=ON \
    -DGGML_CUDA_NO_PEER_COPY=ON \
    -B"build" \
    .

# 5. Build whisper.cpp.
cmake \
    --build "build" \
    -j $(nproc)

# 6. Download the base model for whisper.cpp.
sh ./models/download-ggml-model.sh base.en

# 7. Transcribe an example audio file.
./build/bin/whisper-cli -m ./models/ggml-base.en.bin -f ./samples/jfk.wav

You should then see whisper.cpp logs and the transcription result:

And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.