Add GitHub Actions CI to build and push Docker image to GHCR

Workflow triggers on push to main/release branches, tags, PRs, and
manual dispatch. Uses Docker Buildx with GHA cache for faster rebuilds.
Tags images with ollama version, git SHA, and branch/tag names.

Co-authored-by: Cursor <cursoragent@cursor.com>

.github/workflows/build-push.yml

@@ -0,0 +1,88 @@
+name: Build and push Docker image
+
+on:
+  push:
+    branches:
+      - main
+      - master
+      - "release/**"
+    tags:
+      - "v*"
+  pull_request:
+    branches:
+      - main
+      - master
+  workflow_dispatch:
+    inputs:
+      ollama_version:
+        description: "Ollama version to build"
+        required: false
+        default: "0.15.6"
+
+env:
+  REGISTRY: ghcr.io
+  IMAGE_NAME: ${{ github.repository }}
+  OLLAMA_VERSION: "0.15.6"
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to GitHub Container Registry
+        if: github.event_name != 'pull_request'
+        uses: docker/login-action@v3
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Resolve Ollama version
+        id: version
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ] && [ -n "${{ inputs.ollama_version }}" ]; then
+            echo "ollama_version=${{ inputs.ollama_version }}" >> "$GITHUB_OUTPUT"
+          else
+            echo "ollama_version=${{ env.OLLAMA_VERSION }}" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Extract metadata (tags, labels)
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
+          tags: |
+            # Tag with ollama version on default branch
+            type=raw,value=ollama-${{ steps.version.outputs.ollama_version }},enable={{is_default_branch}}
+            # Tag "latest" on default branch
+            type=raw,value=latest,enable={{is_default_branch}}
+            # Tag with git tag (v1.0.0 -> 1.0.0)
+            type=semver,pattern={{version}}
+            # Tag with branch name for release branches
+            type=ref,event=branch,enable=${{ startsWith(github.ref, 'refs/heads/release/') }}
+            # Tag with short SHA always
+            type=sha,prefix=
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: Dockerfile
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          build-args: |
+            OLLAMA_VERSION=${{ steps.version.outputs.ollama_version }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

README.md

@@ -6,22 +6,34 @@ A Docker-based setup that pairs [Ollama](https://github.com/ollama/ollama) **v0.
 
 **Why this exists:** Ollama's official release ships only a Vulkan backend for Intel GPUs, leaving significant performance on the table. This repo builds the `ggml-sycl` backend from source with Intel oneAPI, unlocking oneMKL, oneDNN, and Level-Zero direct GPU access.
 
-![screenshot](doc/screenshot.png)
 
 ---
 
 ## Quick start
 
+### Option A: Build from source
+
 ```shell
 git clone https://github.com/mattcurf/ollama-intel-gpu
 cd ollama-intel-gpu
 docker compose up
 ```
 
-Open **http://localhost:3000** — pull a model and start chatting.
-
 The first `docker compose up` builds the SYCL backend from source (~2 min on a modern CPU). Subsequent starts are instant.
 
+### Option B: Use the pre-built image
+
+```shell
+docker run -d \
+  --device /dev/dri:/dev/dri \
+  --shm-size 16G \
+  -p 11434:11434 \
+  -v ollama-data:/root/.ollama \
+  ghcr.io/mattcurf/ollama-intel-gpu:latest
+```
+
+Open **http://localhost:3000** (with WebUI) or use the API directly at `http://localhost:11434`.
+
 > **Multiple GPUs?** Set `ONEAPI_DEVICE_SELECTOR=level_zero:0` in `docker-compose.yml` to pick the right device.
 
 ---