Add Docker GitHub Workflow

[xashr]

Summary

#30 added Dockerfiles for CPU and CUDA including examples. This PR adds a full GitHub workflow for building, caching and publishing multi-architecture Docker images to GHCR. Inspired by llama.cpp, adapted to audio.cpp.

This allows the user to simply reference a public docker image without having to build it locally (which is still supported).

Example (if merged): docker pull ghcr.io/0xShug0/audio.cpp:full-cpu

See the fork workflow and fork image tags in action.

Workflow

Trigger

Workflow gets triggered daily at 3:21 UTC or manually. The workflow checks for new commits. If there is no new commit since the last build, the workflow jobs get cancelled to avoid wasting resources.

Can be adjusted as needed, e.g. build weekly to reduce actions usage.

Stages

1. check-commits — Compares HEAD to last-docker-build tag. Skips the entire run if no new commits since the last daily build. workflow_dispatch with force_build=true overrides the skip.

2. metadata — Computes build_date and date_tag once so all jobs use the same timestamp.

3. build — Runs in parallel across architectures (amd64, arm64) and variants (cpu, cuda12, cuda13). Each job:

   • Builds a single-arch image targeting the full stage
   • Pushes by digest to ghcr.io
   • Uploads the digest as an artifact for the merge stage

4. merge — Downloads per-arch digests and uses docker buildx imagetools create to assemble multi-arch manifests with OCI annotations. Produces:

   • full-cpu — mutable "latest" tag
   • full-cpu-YYYYMMDD-HHHHHHH — immutable pinned tag
   • full-cuda12 — mutable "latest" tag
   • full-cuda12-YYYYMMDD-HHHHHHH — immutable pinned tag
   • full-cuda13 — mutable "latest" tag
   • full-cuda13-YYYYMMDD-HHHHHHH — immutable pinned tag

5. update-tag — Force-pushes the last-docker-build git tag to HEAD, enabling the skip-detection in step 1.

Tested variants

Tested:
amd64: cpu, cuda12, cuda13 on Intel CPU + RTX 5090
arm64: cpu on Macbook Pro M5

Not tested (lacking hardware):
arm64: cuda12, cuda13

What is not included in this PR

This PR provides the workflow. If the PR gets accepted and the workflow works reliably, I can extend the documentation and examples.

Open issues

• ENGINE_ENABLE_NATIVE_CPU=OFF: For local Docker builds this can be set to ON for improved performance. For published Docker builds it needs to be OFF to ensure portability, resulting in reduced performance. llama.cpp offers the option to build with GGML_CPU_ALL_VARIANTS=ON which builds multiple separate CPU backend shared libraries, each optimized for a different CPU generation and loaded dynamically. It would be great to have that option for audio.cpp as well. Otherwise I could try to enable native cpu for local and disable it for published images for now.
  => Will create separate issue. For now: OFF for published builds, ON for local builds
• arm64 build warning
  => Removed "free disk space" step, currently not needed

[0xShug0]

@xashr Thanks! One runtime concern: The image currently copies binaries only. Current audio.cpp needs package specs unless the build uses embedded specs, so the Docker build should either enable AUDIOCPP_DEPLOYMENT_BUILD=ON or include model_specs/ in the runtime image.

Comments from ChatGPT (not sure if they are relevent):
Add permissions: contents: read
move the commet outside the build-args block: ENGINE_ENABLE_NATIVE_CPU=OFF # disable for cross-arch portability

[xashr]

@0xShug0 : Valuable feedback! Look's like I was testing with the only model that does not require a model spec (pocket tts)... Fixed that. I will add another example which requires a model with spec in the follow up PR (docs).

Applied the other two suggestions as well. (permissions setting not needed if repo default, but makes it more explicit)

[0xShug0]

@xashr Thank you for this great improvement! Merged.