diff --git a/examples/funcall.py b/examples/funcall.py
index 6f99e78..c82b5b8 100755
--- a/examples/funcall.py
+++ b/examples/funcall.py
@@ -94,6 +94,7 @@ leprompt += r'''
 
 #model_name = "llama-2-7b-chat-codeCherryPop.Q5_K_M.gguf"
 model_name = "mistral-7b-instruct-v0.1.Q2_K.gguf"
+# model_name = "dolphin-2_6-phi-2.Q5_K_M.gguf"
 # model_name = "codellama-7b-instruct.Q4_K_M.gguf"
 # model_name = "codellama-34b-instruct.Q4_K_M.gguf"
 # model_name = "Wizard-Vicuna-7B-Uncensored.Q2_K.gguf"
diff --git a/main.py b/main.py
index f141ba6..719805c 100755
--- a/main.py
+++ b/main.py
@@ -383,16 +383,16 @@ if have_r2pipe:
 if have_rlang:
 	def r2ai_rlang_plugin(unused_but_required_argument):
 		def _call(s):
-			if s == "r2ai":
-				print(help_message)
-				return True
-			elif s.startswith("r2ai"):
-				usertext = s[4:].strip()
-				try:
-					runline(usertext)
-				except Exception as e:
-					print(e)
-					traceback.print_exc()
+			if s.startswith("r2ai"):
+				if len(s) == 4:
+					builtins.print(help_message)
+				else:
+					usertext = s[4:].strip()
+					try:
+						runline(usertext)
+					except Exception as e:
+						builtins.print(e)
+						traceback.print_exc()
 				return True
 			return False
 
@@ -423,4 +423,4 @@ elif have_r2pipe:
 	r2ai_repl()
 else:
 	print("r2ai plugin not initialized, you need to install rlang-python")
-sys.stderr.close()
+# sys.stderr.close()
diff --git a/r2ai/interpreter.py b/r2ai/interpreter.py
index 8941993..d220992 100644
--- a/r2ai/interpreter.py
+++ b/r2ai/interpreter.py
@@ -562,7 +562,7 @@ class Interpreter:
         debug_mode = False # maybe true when debuglevel=2 ?
         self.llama_instance = new_get_hf_llm(self.model, debug_mode, ctxwindow)
         if self.llama_instance == None:
-          print("Cannot find the model")
+          builtins.print("Cannot find the model")
           return
       except:
         traceback.print_exc()
diff --git a/r2ai/models.py b/r2ai/models.py
index 81d2fd9..327e9a9 100644
--- a/r2ai/models.py
+++ b/r2ai/models.py
@@ -95,7 +95,7 @@ def get_hf_llm(repo_id, debug_mode, context_window):
             repo_id = get_default_model()
         if usermodels is not None and repo_id in usermodels:
             model_path = usermodels[repo_id]
-#            print("[r2ai] Using " + r2ai_model_json+": " + model_path)
+#            print(f"[r2ai] Using {r2ai_model_json} {model_path}")
             return llama_cpp.Llama(model_path=model_path, n_gpu_layers=n_gpu_layers, verbose=debug_mode, n_ctx=context_window)
     except:
         traceback.print_exc()
@@ -278,7 +278,6 @@ def get_hf_llm(repo_id, debug_mode, context_window):
             else:
                 install_llama("OpenBLAS")
           
-            from llama_cpp import Llama
             print('', Markdown("Finished downloading `Code-Llama` interface."), '')
 
             # Check if on macOS
@@ -305,9 +304,7 @@ def get_hf_llm(repo_id, debug_mode, context_window):
         json.dump(usermodels, fd)
         fd.close()
         print("Saved")
-    llama_2 = Llama(model_path=model_path, n_gpu_layers=n_gpu_layers, verbose=debug_mode, n_ctx=context_window)
-#    print("[r2ai] Using model: " + model_path)
-    return llama_2
+    return llama_cpp.Llama(model_path=model_path, n_gpu_layers=n_gpu_layers, verbose=debug_mode, n_ctx=context_window)
 
 def set_default_model(repo_id):
     usermodels = {"default": repo_id}
@@ -438,24 +435,14 @@ def new_get_hf_llm(repo_id, debug_mode, context_window):
         return repo_id
     if not os.path.exists(repo_id):
         return get_hf_llm(repo_id, debug_mode, context_window)
-    # print("LOADING FILE: " + repo_id)
+    # print(f"LOADING FILE: {repo_id}")
     n_gpu_layers = -1 # = 0 to use cpu
-#    n_gpu_layers = 0
-    # Third stage: GPU confirm
-#if confirm_action("Use GPU? (Large models might crash on GPU, but will run more quickly)"):
-##      n_gpu_layers = -1
-#    else:
-#      n_gpu_layers = 0
-
-    # Get user data directory
     user_data_dir = appdirs.user_data_dir("Open Interpreter")
     default_path = os.path.join(user_data_dir, "models")
 
     # Ensure the directory exists
     os.makedirs(default_path, exist_ok=True)
     model_path = repo_id
-    # This is helpful for folks looking to delete corrupted ones and such
-#print(Markdown(f"Model found at `{model_path}`"))
   
     try:
         from llama_cpp import Llama
@@ -517,7 +504,6 @@ def new_get_hf_llm(repo_id, debug_mode, context_window):
             else:
                 install_llama("OpenBLAS")
           
-            from llama_cpp import Llama
             print('', Markdown("Finished downloading `Code-Llama` interface."), '')
 
             # Tell them if their architecture won't work well
@@ -542,4 +528,4 @@ def new_get_hf_llm(repo_id, debug_mode, context_window):
     # Initialize and return Code-Llama
     if not os.path.isfile(model_path):
         print("Model is not a file")
-    return Llama(model_path=model_path, n_gpu_layers=n_gpu_layers, verbose=debug_mode, n_ctx=context_window)
+    return llama_cpp.Llama(model_path=model_path, n_gpu_layers=n_gpu_layers, verbose=debug_mode, n_ctx=context_window)