Merge branch 'main' into feature/add-sglang-backend

Author: denyszhak

.github/workflows/cli-build.yml

@@ -35,7 +35,7 @@ jobs:
         working-directory: cmd/cli
         run: |
           make release VERSION=${{ github.sha }}
-      - uses: actions/upload-artifact@330a01c490aca151604b8cf639adc76d48f6c5d4
+      - uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f
         with:
           name: dist
           path: |

.golangci.yml

@@ -2,9 +2,6 @@ version: "2"
 
 run:
   timeout: 5m
-  # Skip vendor and generated code
-  skip-dirs:
-    - pkg/go-containerregistry
 
 formatters:
   enable:
@@ -214,6 +211,10 @@ linters:
       - text: '^shadow: declaration of "(ctx|err|ok)" shadows declaration'
         linters:
           - govet
+      - path: main_test\.go # Exclude main_test.go files because of a false positive: https://github.com/docker/model-runner/actions/runs/20296739902/job/58292102948
+        linters:
+          - staticcheck
+        text: "SA5011"
 
 issues:
   # Maximum issues count per one linter. Set to 0 to disable. Default is 50.

cmd/cli/commands/bench.go

@@ -57,7 +57,7 @@ func newBenchCmd() *cobra.Command {
 	)
 
 	cmd := &cobra.Command{
-		Use:   "bench [MODEL]",
+		Use:   "bench MODEL",
 		Short: "Benchmark a model's performance at different concurrency levels",
 		Long: `Benchmark a model's performance showing tokens per second at different concurrency levels.
 

cmd/cli/commands/compose.go

@@ -37,6 +37,7 @@ func newComposeCmd() *cobra.Command {
 func newUpCommand() *cobra.Command {
 	var models []string
 	var ctxSize int64
+	var rawRuntimeFlags string
 	var backend string
 	var draftModel string
 	var numTokens int
@@ -69,6 +70,9 @@ func newUpCommand() *cobra.Command {
 			if ctxSize > 0 {
 				sendInfo(fmt.Sprintf("Setting context size to %d", ctxSize))
 			}
+			if rawRuntimeFlags != "" {
+				sendInfo("Setting raw runtime flags to " + rawRuntimeFlags)
+			}
 
 			// Build speculative config if any speculative flags are set
 			var speculativeConfig *inference.SpeculativeDecodingConfig
@@ -89,10 +93,11 @@ func newUpCommand() *cobra.Command {
 						ContextSize: &size,
 						Speculative: speculativeConfig,
 					},
+					RawRuntimeFlags: rawRuntimeFlags,
 				}); err != nil {
-					configErrFmtString := "failed to configure backend for model %s with context-size %d"
-					_ = sendErrorf(configErrFmtString+": %v", model, ctxSize, err)
-					return fmt.Errorf(configErrFmtString+": %w", model, ctxSize, err)
+					configErrFmtString := "failed to configure backend for model %s with context-size %d  and runtime-flags %s"
+					_ = sendErrorf(configErrFmtString+": %v", model, ctxSize, rawRuntimeFlags, err)
+					return fmt.Errorf(configErrFmtString+": %w", model, ctxSize, rawRuntimeFlags, err)
 				}
 				sendInfo("Successfully configured backend for model " + model)
 			}
@@ -114,6 +119,7 @@ func newUpCommand() *cobra.Command {
 	}
 	c.Flags().StringArrayVar(&models, "model", nil, "model to use")
 	c.Flags().Int64Var(&ctxSize, "context-size", -1, "context size for the model")
+	c.Flags().StringVar(&rawRuntimeFlags, "runtime-flags", "", "raw runtime flags to pass to the inference engine")
 	c.Flags().StringVar(&backend, "backend", llamacpp.Name, "inference backend to use")
 	c.Flags().StringVar(&draftModel, "speculative-draft-model", "", "draft model for speculative decoding")
 	c.Flags().IntVar(&numTokens, "speculative-num-tokens", 0, "number of tokens to predict speculatively")
@@ -177,7 +183,7 @@ func downloadModelsOnlyIfNotFound(desktopClient *desktop.Client, models []string
 			printer := desktop.NewSimplePrinter(func(s string) {
 				_ = sendInfo(s)
 			})
-			_, _, err = desktopClient.Pull(model, false, printer)
+			_, _, err = desktopClient.Pull(model, printer)
 			if err != nil {
 				_ = sendErrorf("Failed to pull model: %v", err)
 				return fmt.Errorf("Failed to pull model: %w\n", err)

cmd/cli/commands/configure.go

@@ -11,15 +11,28 @@ func newConfigureCmd() *cobra.Command {
 	var flags ConfigureFlags
 
 	c := &cobra.Command{
-		Use:    "configure [--context-size=<n>] [--speculative-draft-model=<model>] [--hf_overrides=<json>] [--gpu-memory-utilization=<float>] [--mode=<mode>] [--think] MODEL",
-		Short:  "Configure runtime options for a model",
-		Hidden: true,
+		Use:     "configure [--context-size=<n>] [--speculative-draft-model=<model>] [--hf_overrides=<json>] [--gpu-memory-utilization=<float>] [--mode=<mode>] [--think] MODEL [-- <runtime-flags...>]",
+		Aliases: []string{"config"},
+		Short:   "Manage model runtime configurations",
+		Hidden:  true,
 		Args: func(cmd *cobra.Command, args []string) error {
-			if len(args) != 1 {
-				return fmt.Errorf(
-					"Exactly one model must be specified, got %d: %v\n\n"+
-						"See 'docker model configure --help' for more information",
-					len(args), args)
+			argsBeforeDash := cmd.ArgsLenAtDash()
+			if argsBeforeDash == -1 {
+				// No "--" used, so we need exactly 1 total argument.
+				if len(args) != 1 {
+					return fmt.Errorf(
+						"Exactly one model must be specified, got %d: %v\n\n"+
+							"See 'docker model configure --help' for more information",
+						len(args), args)
+				}
+			} else {
+				// Has "--", so we need exactly 1 argument before it.
+				if argsBeforeDash != 1 {
+					return fmt.Errorf(
+						"Exactly one model must be specified before --, got %d\n\n"+
+							"See 'docker model configure --help' for more information",
+						argsBeforeDash)
+				}
 			}
 			return nil
 		},
@@ -29,11 +42,13 @@ func newConfigureCmd() *cobra.Command {
 			if err != nil {
 				return err
 			}
+			opts.RuntimeFlags = args[1:]
 			return desktopClient.ConfigureBackend(opts)
 		},
 		ValidArgsFunction: completion.ModelNames(getDesktopClient, -1),
 	}
 
 	flags.RegisterFlags(c)
+	c.AddCommand(newConfigureShowCmd())
 	return c
 }

cmd/cli/commands/configure_show.go

@@ -0,0 +1,35 @@
+package commands
+
+import (
+	"encoding/json"
+	"fmt"
+
+	"github.com/docker/model-runner/cmd/cli/commands/completion"
+	"github.com/spf13/cobra"
+)
+
+func newConfigureShowCmd() *cobra.Command {
+	c := &cobra.Command{
+		Use:   "show [MODEL]",
+		Short: "Show model configurations",
+		Args:  cobra.MaximumNArgs(1),
+		RunE: func(cmd *cobra.Command, args []string) error {
+			var modelFilter string
+			if len(args) > 0 {
+				modelFilter = args[0]
+			}
+			configs, err := desktopClient.ShowConfigs(modelFilter)
+			if err != nil {
+				return err
+			}
+			jsonResult, err := json.MarshalIndent(configs, "", "  ")
+			if err != nil {
+				return fmt.Errorf("failed to marshal configs to JSON: %w", err)
+			}
+			cmd.Println(string(jsonResult))
+			return nil
+		},
+		ValidArgsFunction: completion.ModelNames(getDesktopClient, 1),
+	}
+	return c
+}

cmd/cli/commands/configure_test.go

@@ -305,3 +305,90 @@ func TestThinkFlagBehavior(t *testing.T) {
 		})
 	}
 }
+
+func TestRuntimeFlagsValidation(t *testing.T) {
+	tests := []struct {
+		name          string
+		runtimeFlags  []string
+		expectError   bool
+		errorContains string
+	}{
+		{
+			name:         "valid runtime flags without paths",
+			runtimeFlags: []string{"--verbose", "--threads", "4"},
+			expectError:  false,
+		},
+		{
+			name:         "empty runtime flags",
+			runtimeFlags: []string{},
+			expectError:  false,
+		},
+		{
+			name:          "reject absolute path in value",
+			runtimeFlags:  []string{"--log-file", "/var/log/model.log"},
+			expectError:   true,
+			errorContains: "paths are not allowed",
+		},
+		{
+			name:          "reject absolute path in flag=value format",
+			runtimeFlags:  []string{"--output-file=/tmp/output.txt"},
+			expectError:   true,
+			errorContains: "paths are not allowed",
+		},
+		{
+			name:          "reject relative path",
+			runtimeFlags:  []string{"--config", "../config.yaml"},
+			expectError:   true,
+			errorContains: "paths are not allowed",
+		},
+		{
+			name:          "reject URL",
+			runtimeFlags:  []string{"--endpoint", "http://example.com/api"},
+			expectError:   true,
+			errorContains: "paths are not allowed",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			flags := ConfigureFlags{}
+			req, err := flags.BuildConfigureRequest("test-model")
+			if err != nil {
+				t.Fatalf("BuildConfigureRequest failed: %v", err)
+			}
+
+			// Set runtime flags after building request
+			req.RuntimeFlags = tt.runtimeFlags
+
+			// Note: The actual validation happens in scheduler.ConfigureRunner,
+			// but we're testing that the BuildConfigureRequest correctly
+			// preserves the RuntimeFlags for validation downstream.
+			// For a true integration test, we would need to mock the scheduler.
+
+			if tt.expectError {
+				// In this unit test context, we verify the flags are preserved
+				// The actual validation will happen in the scheduler
+				if len(req.RuntimeFlags) == 0 && len(tt.runtimeFlags) > 0 {
+					t.Error("RuntimeFlags should be preserved in the request")
+				}
+			} else {
+				if !equalStringSlices(req.RuntimeFlags, tt.runtimeFlags) {
+					t.Errorf("Expected RuntimeFlags %v, got %v", tt.runtimeFlags, req.RuntimeFlags)
+				}
+			}
+		})
+	}
+}
+
+// equalStringSlices checks if two string slices are equal
+func equalStringSlices(a, b []string) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	for i := range a {
+		if a[i] != b[i] {
+			return false
+		}
+	}
+	return true
+}

cmd/cli/commands/install-runner.go

@@ -138,7 +138,7 @@ func ensureStandaloneRunnerAvailable(ctx context.Context, printer standalone.Sta
 		port = standalone.DefaultControllerPortCloud
 		environment = "cloud"
 	}
-	if err := standalone.CreateControllerContainer(ctx, dockerClient, port, host, environment, false, gpu, "", modelStorageVolume, printer, engineKind, debug, false); err != nil {
+	if err := standalone.CreateControllerContainer(ctx, dockerClient, port, host, environment, false, gpu, "", modelStorageVolume, printer, engineKind, debug, false, ""); err != nil {
 		return nil, fmt.Errorf("unable to initialize standalone model runner container: %w", err)
 	}
 
@@ -172,6 +172,7 @@ type runnerOptions struct {
 	doNotTrack      bool
 	pullImage       bool
 	pruneContainers bool
+	proxyCert       string
 }
 
 // runInstallOrStart is shared logic for install-runner and start-runner commands
@@ -285,7 +286,7 @@ func runInstallOrStart(cmd *cobra.Command, opts runnerOptions, debug bool) error
 		return fmt.Errorf("unable to initialize standalone model storage: %w", err)
 	}
 	// Create the model runner container.
-	if err := standalone.CreateControllerContainer(cmd.Context(), dockerClient, port, opts.host, environment, opts.doNotTrack, gpu, opts.backend, modelStorageVolume, asPrinter(cmd), engineKind, debug, vllmOnWSL); err != nil {
+	if err := standalone.CreateControllerContainer(cmd.Context(), dockerClient, port, opts.host, environment, opts.doNotTrack, gpu, opts.backend, modelStorageVolume, asPrinter(cmd), engineKind, debug, vllmOnWSL, opts.proxyCert); err != nil {
 		return fmt.Errorf("unable to initialize standalone model runner container: %w", err)
 	}
 
@@ -300,6 +301,7 @@ func newInstallRunner() *cobra.Command {
 	var backend string
 	var doNotTrack bool
 	var debug bool
+	var proxyCert string
 	c := &cobra.Command{
 		Use:   "install-runner",
 		Short: "Install Docker Model Runner (Docker Engine only)",
@@ -312,6 +314,7 @@ func newInstallRunner() *cobra.Command {
 				doNotTrack:      doNotTrack,
 				pullImage:       true,
 				pruneContainers: false,
+				proxyCert:       proxyCert,
 			}, debug)
 		},
 		ValidArgsFunction: completion.NoComplete,
@@ -323,6 +326,7 @@ func newInstallRunner() *cobra.Command {
 		Backend:    &backend,
 		DoNotTrack: &doNotTrack,
 		Debug:      &debug,
+		ProxyCert:  &proxyCert,
 	})
 	return c
 }