Publish ARM64 Build for DGX Spark

[naftulikay]

Checklist

• I searched existing Issues, and did not find a similar enhancement request
• I read the README or the relevant section
• This request is about the Whisper Docker image/config/API, not only faster-whisper itself

Describe the enhancement request
I'm testing out a DGX Spark, which is an Ubuntu ARM64 host with an NVIDIA GB10 GPU and unified RAM between the ARM64 CPU and GPU. I'm attempting to run this image in a Docker Compose environment, and this is not possible since the cuda image tag does not include a build for linux/arm64.

If possible, it would be extremely helpful to publish a linux/arm64 build for the cuda tag.

Problem or use case
Running Whisper on a DGX Spark or DGX Station or any other similar ARM64 platform with an NVIDIA GPU.

Area

• Docker image or build
• Runtime configuration or env vars
• API behavior
• Helper script or management command
• Docker Compose / self-hosted-ai-stack integration
• Documentation

Current workaround
I'm looking into other images to run Whisper.

Additional context
N/A, please ask if you need more info.

[hwdsl2]

@naftulikay Hello! Thanks for the request and for opening PR #10.

After checking the current dependencies, ARM64 CUDA is not yet supported by this image. The AArch64 CTranslate2 package installed with faster-whisper does not appear to include CUDA support, and upstream currently requires compiling CTranslate2 from source to enable CUDA on AArch64.

PR #10 provides a useful starting point for the ARM64 build workflow, but we should not publish ARM64 under the :cuda tag until we have a maintainable CTranslate2 build and a successful GPU transcription test on DGX Spark.

If you would like to investigate or test a supported CTranslate2 build path on your DGX Spark, the results would be very helpful. Further implementation details are being discussed in PR #10.