Add ExtraArgs support for Llama, Mlx, and Vllm server options

pkg/backends/builder.go

@@ -93,3 +93,22 @@ func BuildDockerCommand(backendConfig *config.BackendSettings, instanceArgs []st
 
 	return "docker", dockerArgs, nil
 }
+
+// convertExtraArgsToFlags converts map[string]string to command flags
+// Empty values become boolean flags: {"flag": ""} → ["--flag"]
+// Non-empty values: {"flag": "value"} → ["--flag", "value"]
+func convertExtraArgsToFlags(extraArgs map[string]string) []string {
+	var args []string
+
+	for key, value := range extraArgs {
+		if value == "" {
+			// Boolean flag
+			args = append(args, "--"+key)
+		} else {
+			// Value flag
+			args = append(args, "--"+key, value)
+		}
+	}
+
+	return args
+}

pkg/backends/llama.go

@@ -187,6 +187,10 @@ type LlamaServerOptions struct {
 	FIMQwen7BDefault      bool `json:"fim_qwen_7b_default,omitempty"`
 	FIMQwen7BSpec         bool `json:"fim_qwen_7b_spec,omitempty"`
 	FIMQwen14BSpec        bool `json:"fim_qwen_14b_spec,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/llama.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
 }
 
 // UnmarshalJSON implements custom JSON unmarshaling to support multiple field names
@@ -354,6 +358,18 @@ func (o *LlamaServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
@@ -361,7 +377,12 @@ func (o *LlamaServerOptions) Validate() error {
 func (o *LlamaServerOptions) BuildCommandArgs() []string {
 	// Llama uses multiple flags for arrays by default (not comma-separated)
 	// Use package-level llamaMultiValuedFlags variable
-	return BuildCommandArgs(o, llamaMultiValuedFlags)
+	args := BuildCommandArgs(o, llamaMultiValuedFlags)
+
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
+	return args
 }
 
 func (o *LlamaServerOptions) BuildDockerArgs() []string {

pkg/backends/mlx.go

@@ -29,6 +29,10 @@ type MlxServerOptions struct {
 	TopK      int     `json:"top_k,omitempty"`
 	MinP      float64 `json:"min_p,omitempty"`
 	MaxTokens int     `json:"max_tokens,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/mlx.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
 }
 
 func (o *MlxServerOptions) GetPort() int {
@@ -57,13 +61,30 @@ func (o *MlxServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
 // BuildCommandArgs converts to command line arguments
 func (o *MlxServerOptions) BuildCommandArgs() []string {
 	multipleFlags := map[string]struct{}{} // MLX doesn't currently have []string fields
-	return BuildCommandArgs(o, multipleFlags)
+	args := BuildCommandArgs(o, multipleFlags)
+
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
+	return args
 }
 
 func (o *MlxServerOptions) BuildDockerArgs() []string {

pkg/backends/vllm.go

@@ -142,6 +142,10 @@ type VllmServerOptions struct {
 	OverridePoolingConfig     string `json:"override_pooling_config,omitempty"`
 	OverrideNeuronConfig      string `json:"override_neuron_config,omitempty"`
 	OverrideKVCacheALIGNSize  int    `json:"override_kv_cache_align_size,omitempty"`
+
+	// ExtraArgs are additional command line arguments.
+	// Example: {"verbose": "", "log-file": "/logs/vllm.log"}
+	ExtraArgs map[string]string `json:"extra_args,omitempty"`
 }
 
 func (o *VllmServerOptions) GetPort() int {
@@ -171,6 +175,18 @@ func (o *VllmServerOptions) Validate() error {
 		return validation.ValidationError(fmt.Errorf("invalid port range: %d", o.Port))
 	}
 
+	// Validate extra_args keys and values
+	for key, value := range o.ExtraArgs {
+		if err := validation.ValidateStringForInjection(key); err != nil {
+			return validation.ValidationError(fmt.Errorf("extra_args key %q: %w", key, err))
+		}
+		if value != "" {
+			if err := validation.ValidateStringForInjection(value); err != nil {
+				return validation.ValidationError(fmt.Errorf("extra_args value for %q: %w", key, err))
+			}
+		}
+	}
+
 	return nil
 }
 
@@ -193,6 +209,9 @@ func (o *VllmServerOptions) BuildCommandArgs() []string {
 	flagArgs := BuildCommandArgs(&optionsCopy, vllmMultiValuedFlags)
 	args = append(args, flagArgs...)
 
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
 	return args
 }
 
@@ -203,6 +222,9 @@ func (o *VllmServerOptions) BuildDockerArgs() []string {
 	flagArgs := BuildCommandArgs(o, vllmMultiValuedFlags)
 	args = append(args, flagArgs...)
 
+	// Append extra args at the end
+	args = append(args, convertExtraArgsToFlags(o.ExtraArgs)...)
+
 	return args
 }