Update app.py

app.py

@@ -18,7 +18,22 @@ hf_hub_download(
     local_dir = "./models"
 )
 
+def initialize_model(model):
+    global llm, llm_model
+    if llm is None or llm_model != model:
+        llm = Llama(
+            model_path=f"models/{model}",
+            flash_attn=True,
+            n_gpu_layers=81,
+            n_batch=1024,
+            n_ctx=8192,
+        )
+        llm_model = model
+    return llm
 
+# Initialize the model with the default model
+default_model = "Reflection-Llama-3.1-70B-Q3_K_M.gguf"
+initialize_model(default_model)
 
 def get_messages_formatter_type(model_name):
     if "Llama" in model_name:
@@ -39,20 +54,12 @@ def respond(
     top_k,
     repeat_penalty,
 ):
-    global llm
-    global llm_model
+    global llm, llm_model
     
     chat_template = get_messages_formatter_type(model)
     
-    if llm is None or llm_model != model:
-        llm = Llama(
-            model_path=f"models/{model}",
-            flash_attn=True,
-            n_gpu_layers=81,
-            n_batch=1024,
-            n_ctx=8192,
-        )
-        llm_model = model
+    if llm_model != model:
+        llm = initialize_model(model)
     
     provider = LlamaCppPythonProvider(llm)