Revert "Enable Ollama engine by default" (#13481)

This reverts commit 56f754f46b87749581f73ef3625314bb0e51bfed.

envconfig/config.go

@@ -199,7 +199,7 @@ var (
 	// MultiUserCache optimizes prompt caching for multi-user scenarios
 	MultiUserCache = Bool("OLLAMA_MULTIUSER_CACHE")
 	// Enable the new Ollama engine
-	NewEngine = BoolWithDefault("OLLAMA_NEW_ENGINE")
+	NewEngine = Bool("OLLAMA_NEW_ENGINE")
 	// ContextLength sets the default context length
 	ContextLength = Uint("OLLAMA_CONTEXT_LENGTH", 4096)
 	// Auth enables authentication between the Ollama client and server
@@ -291,7 +291,7 @@ func AsMap() map[string]EnvVar {
 		"OLLAMA_SCHED_SPREAD":      {"OLLAMA_SCHED_SPREAD", SchedSpread(), "Always schedule model across all GPUs"},
 		"OLLAMA_MULTIUSER_CACHE":   {"OLLAMA_MULTIUSER_CACHE", MultiUserCache(), "Optimize prompt caching for multi-user scenarios"},
 		"OLLAMA_CONTEXT_LENGTH":    {"OLLAMA_CONTEXT_LENGTH", ContextLength(), "Context length to use unless otherwise specified (default: 4096)"},
-		"OLLAMA_NEW_ENGINE":        {"OLLAMA_NEW_ENGINE", NewEngine(true), "Enable the new Ollama engine"},
+		"OLLAMA_NEW_ENGINE":        {"OLLAMA_NEW_ENGINE", NewEngine(), "Enable the new Ollama engine"},
 		"OLLAMA_REMOTES":           {"OLLAMA_REMOTES", Remotes(), "Allowed hosts for remote models (default \"ollama.com\")"},
 
 		// Informational

llm/server.go

@@ -143,7 +143,7 @@ func NewLlamaServer(systemInfo ml.SystemInfo, gpus []ml.DeviceInfo, modelPath st
 	var llamaModel *llama.Model
 	var textProcessor model.TextProcessor
 	var err error
-	if envconfig.NewEngine(true) || f.KV().OllamaEngineRequired() {
+	if envconfig.NewEngine() || f.KV().OllamaEngineRequired() {
 		if len(projectors) == 0 {
 			textProcessor, err = model.NewTextProcessor(modelPath)
 		} else {