feat: Add support for custom llama.cpp base image and forked llama-swap repositories

[ryan-steed-usa]

• Introduce BASE_LLAMACPP_IMAGE env var to customize llama.cpp base image
• Introduce LS_REPO env var to customize llama-swap source
• Use GITHUB_REPOSITORY env var to automatically detect forked repos
• Update container tagging to use dynamic repo paths
• Pass build args for BASE_IMAGE and LS_REPO to Containerfile
• Enable flexible release downloads from forked repositories

[coderabbitai]

Walkthrough

Replaces hardcoded base image and repository references with configurable ARGs/variables in the Docker build script and Containerfile; adds build-arg propagation, non-root user setup (UID, GID, USER_HOME, HOME), HEALTHCHECK, and ENTRYPOINT to the container image.

Changes

Cohort / File(s)	Summary
Build script parameterization docker/build-container.sh	Introduced environment-driven variables (BASE_IMAGE via BASE_LLAMACPP_IMAGE, LS_REPO via GITHUB_REPOSITORY); fetches latest release from https://api.github.com/repos/${LS_REPO}/releases/latest; updated CONTAINER_TAG/CONTAINER_LATEST to use ${LS_REPO}; added --build-arg LS_REPO=${LS_REPO} and --build-arg BASE_IMAGE=${BASE_IMAGE} to build invocation.
Containerfile base/image indirection docker/llama-swap.Containerfile	Added ARG BASE_IMAGE and switched FROM to FROM ${BASE_IMAGE}:${BASE_TAG}; added ARG LS_REPO and used it for download URLs and tar extraction paths.
Non-root runtime setup docker/llama-swap.Containerfile	Added public ARGs UID, GID, USER_HOME and ENV HOME; created group/user, set up directories and ownership, copied config, removed extracted tarball, and switched to non-root user.
Runtime behavior & health docker/llama-swap.Containerfile	Added HEALTHCHECK directive and ENTRYPOINT to run llama-swap with /app/config.yaml; retained COPY of example config to /app/config.yaml.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

• Verify replacement of hardcoded mostlygeek/llama-swap occurrences with ${LS_REPO} across build and Containerfile.
• Validate default ARG values and that --build-arg propagation matches build-time expectations.
• Check user creation, directory ownership, and cleanup order (tarball extraction/removal) in the Containerfile.

Possibly related PRs

• feat: improve llama.cpp base image tag for cpu #391 — Modifies docker/build-container.sh image/tag resolution and container tagging logic (related to CONTAINER_TAG/CONTAINER_LATEST changes).
• feat: build both root and non-root container images #412 — Changes build-arg handling and dynamic tagging in docker/build-container.sh (overlaps with added build-args).
• refactor: update Containerfile to support non-root user execution and improve security #368 — Adds UID/GID/USER_HOME handling and non-root user adjustments in the Containerfile (directly related to user/runtime setup).

Pre-merge checks and finishing touches

✅ Passed checks (3 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main changes: adding support for custom base images and forked repositories, which are the primary features introduced in this PR.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

docker/build-container.sh (1)

54-54: Consider passing BASE_IMAGE for complete flexibility.

The Containerfile defines ARG BASE_IMAGE with a default, but the build script doesn't pass it as a build-arg. While this works due to the default value, adding --build-arg BASE_IMAGE=${BASE_IMAGE:-ghcr.io/ggml-org/llama.cpp} would provide full control over the base image via the script, completing the flexibility improvements from this PR.

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 3567b7d and 9f38155.

📒 Files selected for processing (4)

• .github/workflows/release.yml (2 hunks)
• .goreleaser.yaml (2 hunks)
• docker/build-container.sh (2 hunks)
• docker/llama-swap.Containerfile (2 hunks)

🧰 Additional context used

🧠 Learnings (3)

📓 Common learnings

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 391
File: docker/build-container.sh:29-32
Timestamp: 2025-11-08T16:56:35.618Z
Learning: In the mostlygeek/llama-swap repository, llama.cpp container tags from ghcr.io/ggml-org/llama.cpp use build identifiers (e.g., `b6981`) rather than semantic versions. The tag format is `server-<something>-<build-id>` for cpu and `server-<arch>-<something>-<build-id>` for other architectures, where extracting the third field with `awk -F '-' '{print $3}'` correctly retrieves the build identifier.

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 368
File: docker/llama-swap.Containerfile:14-20
Timestamp: 2025-10-27T16:04:23.381Z
Learning: In docker/llama-swap.Containerfile, the user/group creation logic intentionally supports adding a non-root user (non-zero UID) to the root group (GID=0) as a valid configuration. This allows flexible UID/GID customization where users can specify only a UID or both UID and GID.

📚 Learning: 2025-11-08T16:56:35.618Z

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 391
File: docker/build-container.sh:29-32
Timestamp: 2025-11-08T16:56:35.618Z
Learning: In the mostlygeek/llama-swap repository, llama.cpp container tags from ghcr.io/ggml-org/llama.cpp use build identifiers (e.g., `b6981`) rather than semantic versions. The tag format is `server-<something>-<build-id>` for cpu and `server-<arch>-<something>-<build-id>` for other architectures, where extracting the third field with `awk -F '-' '{print $3}'` correctly retrieves the build identifier.

Applied to files:

• docker/build-container.sh
• docker/llama-swap.Containerfile

📚 Learning: 2025-10-27T16:04:23.381Z

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 368
File: docker/llama-swap.Containerfile:14-20
Timestamp: 2025-10-27T16:04:23.381Z
Learning: In docker/llama-swap.Containerfile, the user/group creation logic intentionally supports adding a non-root user (non-zero UID) to the root group (GID=0) as a valid configuration. This allows flexible UID/GID customization where users can specify only a UID or both UID and GID.

Applied to files:

• docker/build-container.sh
• docker/llama-swap.Containerfile

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)

• GitHub Check: run-tests
• GitHub Check: run-tests

🔇 Additional comments (5)

.github/workflows/release.yml (1)

21-22: LGTM! GO111MODULE configuration aligns with goreleaser.

The environment variable setup correctly uses the repository variable with a safe 'auto' fallback, coordinating with the .goreleaser.yaml configuration to ensure consistent Go module handling during releases.

docker/build-container.sh (1)

23-24: LGTM! Repository parameterization implemented correctly.

The LS_REPO variable is well-designed, sourcing from GITHUB_REPOSITORY with a sensible default that maintains backward compatibility.

docker/llama-swap.Containerfile (2)

1-3: LGTM! Base image parameterization enhances flexibility.

The BASE_IMAGE and BASE_TAG arguments with appropriate defaults enable custom base images while maintaining backward compatibility.

---

7-7: LGTM! Repository parameterization supports forked releases.

The LS_REPO argument correctly enables downloading llama-swap binaries from custom repositories, aligning with the PR's goal of supporting forked builds.

.goreleaser.yaml (1)

3-4: LGTM! GO111MODULE template configured correctly.

The goreleaser template syntax properly checks for the environment variable and provides a sensible 'auto' default, coordinating with the workflow configuration to enable flexible Go module mode handling for custom builds.

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

docker/llama-swap.Containerfile (1)

31-35: Optional: quote the full curl URL for readability and robustness.

Current command works, but quoting the entire URL is a bit clearer and future-proof against any odd characters:

-    curl -LO https://github.com/"${LS_REPO}"/releases/download/v"${LS_VER}"/llama-swap_"${LS_VER}"_linux_amd64.tar.gz && \
+    curl -LO "https://github.com/${LS_REPO}/releases/download/v${LS_VER}/llama-swap_${LS_VER}_linux_amd64.tar.gz" && \

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 9f38155 and be71f3f.

📒 Files selected for processing (2)

• docker/build-container.sh (3 hunks)
• docker/llama-swap.Containerfile (2 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

• docker/build-container.sh

🧰 Additional context used

🧠 Learnings (3)

📓 Common learnings

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 391
File: docker/build-container.sh:29-32
Timestamp: 2025-11-08T16:56:35.626Z
Learning: In the mostlygeek/llama-swap repository, llama.cpp container tags from ghcr.io/ggml-org/llama.cpp use build identifiers (e.g., `b6981`) rather than semantic versions. The tag format is `server-<something>-<build-id>` for cpu and `server-<arch>-<something>-<build-id>` for other architectures, where extracting the third field with `awk -F '-' '{print $3}'` correctly retrieves the build identifier.

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 368
File: docker/llama-swap.Containerfile:14-20
Timestamp: 2025-10-27T16:04:23.381Z
Learning: In docker/llama-swap.Containerfile, the user/group creation logic intentionally supports adding a non-root user (non-zero UID) to the root group (GID=0) as a valid configuration. This allows flexible UID/GID customization where users can specify only a UID or both UID and GID.

📚 Learning: 2025-10-27T16:04:23.381Z

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 368
File: docker/llama-swap.Containerfile:14-20
Timestamp: 2025-10-27T16:04:23.381Z
Learning: In docker/llama-swap.Containerfile, the user/group creation logic intentionally supports adding a non-root user (non-zero UID) to the root group (GID=0) as a valid configuration. This allows flexible UID/GID customization where users can specify only a UID or both UID and GID.

Applied to files:

• docker/llama-swap.Containerfile

📚 Learning: 2025-11-08T16:56:35.626Z

Learnt from: ryan-steed-usa
Repo: mostlygeek/llama-swap PR: 391
File: docker/build-container.sh:29-32
Timestamp: 2025-11-08T16:56:35.626Z
Learning: In the mostlygeek/llama-swap repository, llama.cpp container tags from ghcr.io/ggml-org/llama.cpp use build identifiers (e.g., `b6981`) rather than semantic versions. The tag format is `server-<something>-<build-id>` for cpu and `server-<arch>-<something>-<build-id>` for other architectures, where extracting the third field with `awk -F '-' '{print $3}'` correctly retrieves the build identifier.

Applied to files:

• docker/llama-swap.Containerfile

🔇 Additional comments (1)

docker/llama-swap.Containerfile (1)

1-7: Configurable base image and repo args look correct and backward-compatible.

Using ARG BASE_IMAGE/BASE_TAG in FROM and ARG LS_REPO=mostlygeek/llama-swap preserves prior defaults while allowing forks and alternate base images via build-args. No issues spotted here.

[mostlygeek]

LGTM