Move passthrough model name to constant and clarify error message.

Signed-off-by: Jacob Howard <jacob.howard@docker.com>

Author: xenoscopic

pkg/inference/scheduling/loader.go

@@ -25,6 +25,9 @@ const (
 	// defaultRunnerIdleTimeout is the default maximum amount of time that a
 	// runner can sit idle (i.e. without any requests) before being terminated.
 	defaultRunnerIdleTimeout = 5 * time.Minute
+	// passthroughModelName is the model name used for runners associated with
+	// passthrough backends.
+	passthroughModelName = "passthrough"
 )
 
 var (

pkg/inference/scheduling/scheduler.go

@@ -321,7 +321,7 @@ func (s *Scheduler) handleOpenAIInference(w http.ResponseWriter, r *http.Request
 	// backends, we assume they have external model management.
 	if backendMode == inference.BackendModeCompletion || backendMode == inference.BackendModeEmbedding {
 		if request.Model == "" {
-			http.Error(w, "invalid request, model is required", http.StatusBadRequest)
+			http.Error(w, `invalid request - "model" is required`, http.StatusBadRequest)
 			return
 		}
 		model, err := s.modelManager.GetModel(request.Model)
@@ -343,7 +343,7 @@ func (s *Scheduler) handleOpenAIInference(w http.ResponseWriter, r *http.Request
 	// runners (one for each model used by a passthrough backend).
 	runnerModel := request.Model
 	if backendMode == inference.BackendModePassthrough {
-		runnerModel = "passthrough"
+		runnerModel = passthroughModelName
 	}
 
 	// Request a runner to execute the request and defer its release.