From 4dfa43a056ecc07ee6e12a2d3445ae87bf9aa977 Mon Sep 17 00:00:00 2001
From: Iulian Meghea <iulian@plumelo.com>
Date: Tue, 26 May 2026 09:30:47 +0000
Subject: [PATCH] feat: add llamacpp Docker image

Based on ghcr.io/ggml-org/llama.cpp:server-cuda with dstack runner
prerequisites (openssh-server, sudo, dstack user). Includes MTP and
Jinja autoparser support from the official llama.cpp image.
---
 .github/workflows/build.yml | 43 +++++++++++++++++++++++++++++++++++++
 .github/workflows/ci.yml    | 23 +++++++++++++++++++-
 README.md                   | 17 ++++++++++++++-
 images/llamacpp/Dockerfile  | 14 ++++++++++++
 4 files changed, 95 insertions(+), 2 deletions(-)
 create mode 100644 images/llamacpp/Dockerfile

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index c3193d4..8f73d2f 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -89,6 +89,49 @@ jobs:
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
 
+      - name: Generate artifact attestation
+        uses: actions/attest-build-provenance@v4
+        with:
+          subject-name: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          subject-digest: ${{ steps.push.outputs.digest }}
+          push-to-registry: true
+
+  build-llamacpp:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+      attestations: write
+      id-token: write
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Log in to the Container registry
+        uses: docker/login-action@v4
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Extract metadata (tags, labels) for Docker
+        id: meta
+        uses: docker/metadata-action@v6
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            type=raw,value=llamacpp-latest
+            type=sha,prefix=llamacpp-
+
+      - name: Build and push Docker image
+        id: push
+        uses: docker/build-push-action@v7
+        with:
+          context: ./images/llamacpp
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+
       - name: Generate artifact attestation
         uses: actions/attest-build-provenance@v4
         with:
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index ca4bbc7..097f68c 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -45,4 +45,25 @@ jobs:
         run: docker run --rm test-vllm-nightly id dstack
 
       - name: Verify dstack user has sudo
-        run: docker run --rm test-vllm-nightly sudo -n true
\ No newline at end of file
+        run: docker run --rm test-vllm-nightly sudo -n true
+
+  check-llamacpp:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Build llamacpp image
+        run: docker build -t test-llamacpp ./images/llamacpp
+
+      - name: Verify openssh-server installed
+        run: docker run --rm test-llamacpp which sshd
+
+      - name: Verify sudo installed
+        run: docker run --rm test-llamacpp which sudo
+
+      - name: Verify dstack user exists
+        run: docker run --rm test-llamacpp id dstack
+
+      - name: Verify dstack user has sudo
+        run: docker run --rm test-llamacpp sudo -n true
\ No newline at end of file
diff --git a/README.md b/README.md
index d028569..b2d9205 100644
--- a/README.md
+++ b/README.md
@@ -18,6 +18,13 @@ Same as vllm but based on `vllm/vllm-openai:nightly` — bleeding edge vLLM buil
 - **Tag**: `ghcr.io/plumelo/dstack-images:vllm-nightly-latest`
 - **What's added**: openssh-server, sudo, dstack user (uid/gid 1000), `/dstack/run`
 
+### llamacpp
+
+Based on `ghcr.io/ggml-org/llama.cpp:server-cuda` with pre-installed dstack runner prerequisites. Includes MTP (multi-token prediction) support and the Jinja-based autoparser for automatic tool call detection.
+
+- **Tag**: `ghcr.io/plumelo/dstack-images:llamacpp-latest`
+- **What's added**: openssh-server, sudo, dstack user (uid/gid 1000), `/dstack/run`
+
 ## Usage
 
 ```yaml
@@ -36,6 +43,14 @@ image: ghcr.io/plumelo/dstack-images:vllm-nightly-latest
 commands:
   - |
     vllm serve $MODEL_ID --host 0.0.0.0 --port 8000 ...
+
+# llama.cpp
+type: service
+name: my-llamacpp
+image: ghcr.io/plumelo/dstack-images:llamacpp-latest
+commands:
+  - |
+    /app/llama-server --hf-repo $MODEL_REPO --hf-file $MODEL_FILE --port 8000 ...
 ```
 
 ## Cold start improvement
@@ -47,5 +62,5 @@ commands:
 
 ## CI/CD
 
-- **Push to main**: builds and pushes both image tags + SHA tags
+- **Push to main**: builds and pushes all image tags + SHA tags
 - **PR**: builds and runs verification checks per image
diff --git a/images/llamacpp/Dockerfile b/images/llamacpp/Dockerfile
new file mode 100644
index 0000000..57724f8
--- /dev/null
+++ b/images/llamacpp/Dockerfile
@@ -0,0 +1,14 @@
+FROM ghcr.io/ggml-org/llama.cpp:server-cuda
+
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+      openssh-server sudo \
+    && sed -i "s/.*PasswordAuthentication.*/PasswordAuthentication no/g" /etc/ssh/sshd_config \
+    && mkdir /run/sshd \
+    && rm -f /etc/ssh/ssh_host_* \
+    && groupadd -g 1000 dstack \
+    && useradd -u 1000 -g 1000 -G sudo -s /bin/bash -m dstack \
+    && echo 'dstack ALL=(ALL) NOPASSWD: ALL' > /etc/sudoers.d/dstack \
+    && mkdir -p /dstack/run && chmod a+rwx /dstack/run \
+    && rm -rf /var/lib/apt/lists/*
+
+EXPOSE 8000
\ No newline at end of file