Update app.py

app.py

@@ -9,6 +9,9 @@ from llama_cpp_agent.chat_history.messages import Roles
 import gradio as gr
 from huggingface_hub import hf_hub_download
 
+llm = None
+llm_model = None
+
 hf_hub_download(
     repo_id="bartowski/Meta-Llama-3-70B-Instruct-GGUF",
     filename="Meta-Llama-3-70B-Instruct-Q3_K_M.gguf",
@@ -81,15 +84,21 @@ def respond(
     top_k,
     repeat_penalty,
 ):
+    global llm
+    global llm_model
+    
     chat_template = get_messages_formatter_type(model)
-
-    llm = Llama(
-        model_path=f"models/{model}",
-        flash_attn=True,
-        n_gpu_layers=81,
-        n_batch=1024,
-        n_ctx=8192,
-    )
+    
+    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
+    
     provider = LlamaCppPythonProvider(llm)
 
     agent = LlamaCppAgent(