add server version endpoint and client version display

.github/workflows/release.yml

@@ -308,6 +308,7 @@ jobs:
           platforms: linux/amd64, linux/arm64
           build-args: |
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -323,6 +324,7 @@ jobs:
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
             "LLAMA_SERVER_VARIANT=cuda"
             "BASE_IMAGE=nvidia/cuda:12.9.0-runtime-ubuntu24.04"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -341,6 +343,7 @@ jobs:
             "VLLM_VERSION=${{ env.VLLM_VERSION }}"
             "VLLM_CUDA_VERSION=cu130"
             "VLLM_PYTHON_TAG=cp38-abi3"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -357,6 +360,7 @@ jobs:
             "LLAMA_SERVER_VARIANT=cuda"
             "BASE_IMAGE=nvidia/cuda:12.9.0-runtime-ubuntu24.04"
             "SGLANG_VERSION=${{ env.SGLANG_VERSION }}"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -370,6 +374,7 @@ jobs:
           platforms: linux/amd64, linux/arm64
           build-args: |
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -385,6 +390,7 @@ jobs:
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
             "LLAMA_SERVER_VARIANT=rocm"
             "BASE_IMAGE=rocm/dev-ubuntu-22.04"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -401,6 +407,7 @@ jobs:
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
             "LLAMA_SERVER_VARIANT=musa"
             "BASE_IMAGE=mthreads/musa:rc4.3.0-runtime-ubuntu22.04-amd64"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -417,6 +424,7 @@ jobs:
             "LLAMA_SERVER_VERSION=${{ env.LLAMA_SERVER_VERSION }}"
             "LLAMA_SERVER_VARIANT=cann"
             "BASE_IMAGE=ascendai/cann:8.2.rc2-910b-ubuntu22.04-py3.11"
+            "VERSION=${{ env.RELEASE_TAG }}"
           push: true
           sbom: true
           provenance: mode=max
@@ -677,26 +685,19 @@ jobs:
           SUMMARY
 
   # ---------------------------------------------------------------------------
-  # Verify Docker CE installation — run test-docker-ce-installation.sh
-  # to confirm the CLI version is available via get.docker.com
+  # Verify Docker CE installation and server version — install Docker CE,
+  # start the released model-runner image, and use `docker model version`
+  # to confirm both client and server versions match the release tag.
   # ---------------------------------------------------------------------------
   verify-docker-ce:
     needs: [prepare, release-cli-docker-ce, build]
     runs-on: ubuntu-latest
     timeout-minutes: 15
-    strategy:
-      fail-fast: false
-      matrix:
-        base_image:
-          - ubuntu:24.04
-          - debian:12
     steps:
       - name: Checkout code
         uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd
 
-      - name: Verify Docker CE installation
-        env:
-          BASE_IMAGE: ${{ matrix.base_image }}
+      - name: Verify client and server versions
         run: |
           chmod +x scripts/test-docker-ce-installation.sh
           ./scripts/test-docker-ce-installation.sh "${{ needs.prepare.outputs.release_tag }}"

Dockerfile

@@ -9,8 +9,12 @@ ARG LLAMA_BINARY_PATH=/com.docker.llama-server.native.linux.${LLAMA_SERVER_VARIA
 # use 22.04 for gpu variants to match ROCm/CUDA base images
 ARG BASE_IMAGE=ubuntu:26.04
 
+ARG VERSION=dev
+
 FROM docker.io/library/golang:${GO_VERSION}-bookworm AS builder
 
+ARG VERSION
+
 # Install git for go mod download if needed
 RUN apt-get update && apt-get install -y --no-install-recommends git && rm -rf /var/lib/apt/lists/*
 
@@ -30,13 +34,14 @@ COPY --link . .
 # Build the Go binary (static build)
 RUN --mount=type=cache,target=/go/pkg/mod \
     --mount=type=cache,target=/root/.cache/go-build \
-    CGO_ENABLED=1 GOOS=linux go build -ldflags="-s -w" -o model-runner .
+    CGO_ENABLED=1 GOOS=linux go build -ldflags="-s -w -X main.Version=${VERSION}" -o model-runner .
 
 # Build the Go binary for SGLang (without vLLM)
 FROM builder AS builder-sglang
+ARG VERSION
 RUN --mount=type=cache,target=/go/pkg/mod \
     --mount=type=cache,target=/root/.cache/go-build \
-    CGO_ENABLED=1 GOOS=linux go build -tags=novllm -ldflags="-s -w" -o model-runner .
+    CGO_ENABLED=1 GOOS=linux go build -tags=novllm -ldflags="-s -w -X main.Version=${VERSION}" -o model-runner .
 
 # --- Get llama.cpp binary ---
 FROM docker/docker-model-backend-llamacpp:${LLAMA_SERVER_VERSION}-${LLAMA_SERVER_VARIANT} AS llama-server

Makefile

@@ -32,7 +32,7 @@ BUILD_DMR ?= 1
 
 # Build the Go application
 build:
-	CGO_ENABLED=1 go build -ldflags="-s -w" -o $(APP_NAME) .
+	CGO_ENABLED=1 go build -ldflags="-s -w -X main.Version=$(shell git describe --tags --always --dirty --match 'v*')" -o $(APP_NAME) .
 
 build-cli:
 	$(MAKE) -C cmd/cli

cmd/cli/commands/version.go

@@ -1,6 +1,8 @@
 package commands
 
 import (
+	"runtime"
+
 	"github.com/docker/model-runner/cmd/cli/commands/completion"
 	"github.com/docker/model-runner/cmd/cli/desktop"
 	"github.com/spf13/cobra"
@@ -11,8 +13,24 @@ func newVersionCmd() *cobra.Command {
 		Use:   "version",
 		Short: "Show the Docker Model Runner version",
 		Run: func(cmd *cobra.Command, args []string) {
-			cmd.Printf("Docker Model Runner version %s\n", desktop.Version)
-			cmd.Printf("Docker Engine Kind: %s\n", modelRunner.EngineKind())
+			cmd.Println("Client:")
+			cmd.Printf(" Version:    %s\n", desktop.Version)
+			cmd.Printf(" OS/Arch:    %s/%s\n", runtime.GOOS, runtime.GOARCH)
+
+			cmd.Println()
+			cmd.Println("Server:")
+			serverVersion := "(not reachable)"
+			if desktopClient != nil {
+				if sv, err := desktopClient.ServerVersion(); err == nil {
+					serverVersion = sv.Version
+				}
+			}
+			cmd.Printf(" Version:    %s\n", serverVersion)
+			if modelRunner != nil {
+				cmd.Printf(" Engine:     %s\n", modelRunner.EngineKind())
+			} else {
+				cmd.Println(" Engine:     (not reachable)")
+			}
 		},
 		ValidArgsFunction: completion.NoComplete,
 	}

cmd/cli/desktop/desktop.go

@@ -713,6 +713,34 @@ func (c *Client) Remove(modelArgs []string, force bool) (string, error) {
 	return modelRemoved, nil
 }
 
+type ServerVersionResponse struct {
+	Version string `json:"version"`
+}
+
+func (c *Client) ServerVersion() (ServerVersionResponse, error) {
+	resp, err := c.doRequest(http.MethodGet, "/version", nil)
+	if err != nil {
+		return ServerVersionResponse{}, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return ServerVersionResponse{}, fmt.Errorf("failed to get server version: %s", resp.Status)
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return ServerVersionResponse{}, fmt.Errorf("failed to read response body: %w", err)
+	}
+
+	var version ServerVersionResponse
+	if err := json.Unmarshal(body, &version); err != nil {
+		return ServerVersionResponse{}, fmt.Errorf("failed to unmarshal response body: %w", err)
+	}
+
+	return version, nil
+}
+
 // BackendStatus to be imported from docker/model-runner when https://github.com/docker/model-runner/pull/42 is merged.
 type BackendStatus struct {
 	BackendName string               `json:"backend_name"`

main.go

@@ -3,6 +3,7 @@ package main
 import (
 	"context"
 	"crypto/tls"
+	"encoding/json"
 	"net"
 	"net/http"
 	"os"
@@ -264,6 +265,14 @@ func main() {
 	anthropicHandler := anthropic.NewHandler(log, schedulerHTTP, nil, modelManager)
 	router.Handle(anthropic.APIPrefix+"/", anthropicHandler)
 
+	// Register /version endpoint
+	router.HandleFunc("/version", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		if err := json.NewEncoder(w).Encode(map[string]string{"version": Version}); err != nil {
+			log.Warnf("failed to write version response: %v", err)
+		}
+	})
+
 	// Register root handler LAST - it will only catch exact "/" requests that don't match other patterns
 	router.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
 		// Only respond to exact root path

pkg/inference/models/handler_test.go

@@ -135,8 +135,6 @@ func TestPullModel(t *testing.T) {
 }
 
 func TestHandleGetModel(t *testing.T) {
-	tempDir := t.TempDir()
-
 	// Create a test registry
 	server := httptest.NewServer(testregistry.New())
 	defer server.Close()
@@ -207,7 +205,9 @@ func TestHandleGetModel(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			log := logrus.NewEntry(logrus.StandardLogger())
+			tempDir := t.TempDir()
+			logger := logrus.New()
+			log := logrus.NewEntry(logger)
 			manager := NewManager(log.WithFields(logrus.Fields{"component": "model-manager"}), ClientConfig{
 				StoreRootPath: tempDir,
 				Logger:        log.WithFields(logrus.Fields{"component": "model-manager"}),
@@ -263,14 +263,6 @@ func TestHandleGetModel(t *testing.T) {
 					t.Errorf("Failed to decode response body: %v", err)
 				}
 			}
-
-			// Clean tempDir after each test
-			if err := os.RemoveAll(tempDir); err != nil {
-				t.Fatalf("Failed to clean temp directory: %v", err)
-			}
-			if err := os.MkdirAll(tempDir, 0755); err != nil {
-				t.Fatalf("Failed to recreate temp directory: %v", err)
-			}
 		})
 	}
 }

scripts/test-docker-ce-in-container.sh

@@ -14,10 +14,30 @@ echo "Testing docker model version..."
 version_output=$(docker model version 2>&1 || true)
 echo "Output: $version_output"
 
-if echo "$version_output" | grep -q "version $EXPECTED_VERSION"; then
-  echo "✓ Success: Found expected version $EXPECTED_VERSION"
-  exit 0
+# Extract client version from the "Client:" section (first "Version:" after "Client:")
+client_version=$(echo "$version_output" | awk '/^Client:/{found=1} found && /Version:/{print $2; exit}')
+
+# Extract server version from the "Server:" section (first "Version:" after "Server:")
+server_version=$(echo "$version_output" | awk '/^Server:/{found=1} found && /Version:/{print $2; exit}')
+
+errors=0
+
+if [ "$client_version" = "$EXPECTED_VERSION" ]; then
+  echo "✓ Client version matches expected $EXPECTED_VERSION"
 else
-  echo "✗ Error: Expected version $EXPECTED_VERSION not found in output"
+  echo "✗ Error: Expected client version $EXPECTED_VERSION, got '$client_version'"
+  errors=$((errors + 1))
+fi
+
+if [ "$server_version" = "$EXPECTED_VERSION" ]; then
+  echo "✓ Server version matches expected $EXPECTED_VERSION"
+else
+  echo "✗ Error: Expected server version $EXPECTED_VERSION, got '$server_version'"
+  errors=$((errors + 1))
+fi
+
+if [ "$errors" -gt 0 ]; then
   exit 1
 fi
+
+echo "✓ All version checks passed!"

scripts/test-docker-ce-installation.sh

@@ -25,24 +25,51 @@ main() {
     fi
   fi
 
-  echo "Testing Docker CE installation with expected CLI version: $cli_version"
+  echo "Testing Docker CE installation with expected version: $cli_version"
 
-  if [ -z "${BASE_IMAGE:-}" ]; then
-    echo "Error: BASE_IMAGE is not set" >&2
-    exit 1
-  fi
+  local base_image="${BASE_IMAGE:-ubuntu:24.04}"
+  echo "Using base image: $base_image"
+
+  local server_image="docker/model-runner:$cli_version"
+  echo "Using server image: $server_image"
+
+  # Start the model-runner server container
+  echo "Starting model-runner server..."
+  docker run -d --name dmr-version-test -p 12434:12434 "$server_image"
 
-  echo "Using base image: $BASE_IMAGE"
+  # Ensure cleanup on exit
+  cleanup() {
+    echo "Stopping model-runner server..."
+    docker stop dmr-version-test 2>/dev/null || true
+    docker rm dmr-version-test 2>/dev/null || true
+  }
+  trap cleanup EXIT
 
-  echo "Starting container and installing Docker CE..."
+  # Wait for server to be ready
+  echo "Waiting for server to be ready..."
+  for i in $(seq 1 30); do
+    if curl -sf http://localhost:12434/version > /dev/null 2>&1; then
+      echo "Server is ready"
+      break
+    fi
+    if [ "$i" -eq 30 ]; then
+      echo "Error: Server did not become ready in time" >&2
+      docker logs dmr-version-test
+      exit 1
+    fi
+    sleep 1
+  done
 
   local script_dir
   script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 
+  echo "Starting test container..."
   docker run --rm \
+    --network host \
     -e "EXPECTED_VERSION=$cli_version" \
+    -e "MODEL_RUNNER_HOST=http://localhost:12434" \
     -v "$script_dir/test-docker-ce-in-container.sh:/test.sh:ro" \
-    "$BASE_IMAGE" \
+    "$base_image" \
     /test.sh
 
   echo "✓ Docker CE installation test passed!"

version.go

@@ -0,0 +1,3 @@
+package main
+
+var Version = "dev"