[sc-147336] new help option to see available vllm parameters

Author: habfast

cmd/aiservices/deployment/deployment_create.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"os"
+	"sort"
 	"strings"
 
 	exocmd "github.com/exoscale/cli/cmd"
@@ -18,14 +19,15 @@ type DeploymentCreateCmd struct {
 
 	_ bool `cli-cmd:"create"`
 
-	Name     string `cli-arg:"#" cli-usage:"NAME"`
+	Name     string `cli-arg:"?" cli-usage:"NAME"`
 	GPUType  string `cli-flag:"gpu-type" cli-usage:"GPU type family (e.g., gpua5000, gpu3080ti)"`
 	GPUCount int64  `cli-flag:"gpu-count" cli-usage:"Number of GPUs (1-8)"`
 	Replicas int64  `cli-flag:"replicas" cli-usage:"Number of replicas (>=1)"`
 
 	ModelID                   string      `cli-flag:"model-id" cli-usage:"Model ID (UUID)"`
 	ModelName                 string      `cli-flag:"model-name" cli-usage:"Model name (as created)"`
 	InferenceEngineParameters string      `cli-flag:"inference-engine-params" cli-usage:"Space-separated inference engine server CLI arguments (e.g., \"--gpu-memory-usage=0.8 --max-tokens=4096\")"`
+	InferenceEngineHelp       bool        `cli-flag:"inference-engine-parameter-help" cli-usage:"Show inference engine parameters help"`
 	Zone                      v3.ZoneName `cli-short:"z" cli-usage:"zone"`
 }
 
@@ -38,13 +40,88 @@ func (c *DeploymentCreateCmd) CmdPreRun(cmd *cobra.Command, args []string) error
 	exocmd.CmdSetZoneFlagFromDefault(cmd)
 	return exocmd.CliCommandDefaultPreRun(c, cmd, args)
 }
+func (c *DeploymentCreateCmd) showInferenceEngineParameterHelp(ctx context.Context, client *v3.Client) error {
+	resp, err := client.GetInferenceEngineHelp(ctx)
+	if err != nil {
+		return err
+	}
+
+	sections := make(map[string][]v3.InferenceEngineParameterEntry)
+	var sectionNames []string
+	for _, p := range resp.Parameters {
+		if _, ok := sections[p.Section]; !ok {
+			sectionNames = append(sectionNames, p.Section)
+		}
+		sections[p.Section] = append(sections[p.Section], p)
+	}
+	sort.Strings(sectionNames)
+
+	for i, section := range sectionNames {
+		if i > 0 {
+			fmt.Println()
+		}
+		fmt.Printf("%s:\n", section)
+		for _, p := range sections[section] {
+			flags := strings.Join(p.Flags, ", ")
+			if p.Type != "boolean" && p.Type != "enum" {
+				flags += " " + strings.ToUpper(strings.ReplaceAll(p.Name, "-", "_"))
+			}
+
+			fmt.Printf("  %s\n", flags)
+
+			desc := p.Description
+			if p.Default != "" {
+				// The description in example sometimes already includes default, but the example output
+				// shows (default: ...) at the end.
+				if !strings.Contains(desc, fmt.Sprintf("(default: %s)", p.Default)) {
+					desc += fmt.Sprintf(" (default: %s)", p.Default)
+				}
+			}
+
+			// Simple wrapping
+			words := strings.Fields(desc)
+			if len(words) > 0 {
+				line := "                        "
+				for _, word := range words {
+					if len(line)+len(word) > 160 {
+						fmt.Println(line)
+						line = "                        " + word
+					} else {
+						if line == "                        " {
+							line += word
+						} else {
+							line += " " + word
+						}
+					}
+				}
+				fmt.Println(line)
+			}
+		}
+	}
+
+	return nil
+}
+
 func (c *DeploymentCreateCmd) CmdRun(_ *cobra.Command, _ []string) error {
 	ctx := exocmd.GContext
+
+	if c.InferenceEngineHelp {
+		client, err := exocmd.SwitchClientZoneV3(ctx, globalstate.EgoscaleV3Client, c.Zone)
+		if err != nil {
+			return err
+		}
+		return c.showInferenceEngineParameterHelp(ctx, client)
+	}
+
 	client, err := exocmd.SwitchClientZoneV3(ctx, globalstate.EgoscaleV3Client, c.Zone)
 	if err != nil {
 		return err
 	}
 
+	if c.Name == "" {
+		return fmt.Errorf("NAME is required")
+	}
+
 	if c.GPUType == "" || c.GPUCount == 0 {
 		return fmt.Errorf("--gpu-type and --gpu-count are required")
 	}

cmd/aiservices/deployment/deployment_create_test.go

@@ -123,3 +123,62 @@ func TestDeploymentCreateWithInferenceEngineParameters(t *testing.T) {
 		}
 	}
 }
+
+func TestDeploymentCreateInferenceEngineHelp(t *testing.T) {
+	mux := http.NewServeMux()
+	mux.HandleFunc("/ai/deployment/inference-engine-help", func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodGet {
+			w.WriteHeader(http.StatusMethodNotAllowed)
+			return
+		}
+		resp := v3.GetInferenceEngineHelpResponse{
+			Parameters: []v3.InferenceEngineParameterEntry{
+				{
+					Name:        "config-format",
+					Flags:       []string{"--config-format"},
+					Type:        "enum",
+					Default:     "auto",
+					Section:     "ModelConfig",
+					Description: "The format of the model config to load.",
+				},
+				{
+					Name:        "max-model-len",
+					Flags:       []string{"--max-model-len"},
+					Type:        "integer",
+					Default:     "None",
+					Section:     "ModelConfig",
+					Description: "Model context length.",
+				},
+			},
+		}
+		writeJSON(t, w, http.StatusOK, resp)
+	})
+	srv := httptest.NewServer(mux)
+	defer srv.Close()
+
+	exocmd.GContext = context.Background()
+	globalstate.Quiet = true
+	creds := credentials.NewStaticCredentials("key", "secret")
+	client, err := v3.NewClient(creds)
+	if err != nil {
+		t.Fatalf("new client: %v", err)
+	}
+	globalstate.EgoscaleV3Client = client.WithEndpoint(v3.Endpoint(srv.URL))
+
+	c := &DeploymentCreateCmd{
+		CliCommandSettings:  exocmd.DefaultCLICmdSettings(),
+		InferenceEngineHelp: true,
+	}
+	if err := c.CmdRun(nil, nil); err != nil {
+		t.Fatalf("deployment create help without name: %v", err)
+	}
+
+	c = &DeploymentCreateCmd{
+		CliCommandSettings:  exocmd.DefaultCLICmdSettings(),
+		Name:                "test-deploy",
+		InferenceEngineHelp: true,
+	}
+	if err := c.CmdRun(nil, nil); err != nil {
+		t.Fatalf("deployment create help with name: %v", err)
+	}
+}