updated

app.py

@@ -28,8 +28,8 @@ callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])
 
 def load_model(
     directory: str = ".",
-    model_name: str = "natsumura-storytelling-rp-1.0-llama-3.1-8B.Q4_K_M.gguf",
-    model_url: str = "https://huggingface.co/tohur/natsumura-storytelling-rp-1.0-llama-3.1-8b-GGUF/resolve/main/natsumura-storytelling-rp-1.0-llama-3.1-8B.Q4_K_M.gguf"
+    model_name: str = "natsumura-storytelling-rp-1.0-llama-3.1-8B.Q3_K_M.gguf",
+    model_url: str = "https://huggingface.co/tohur/natsumura-storytelling-rp-1.0-llama-3.1-8b-GGUF/resolve/main/natsumura-storytelling-rp-1.0-llama-3.1-8B.Q3_K_M.gguf"
 ):
     final_model_path = os.path.join(directory, model_name)
     
@@ -60,6 +60,7 @@ def load_model(
         temperature=0.3,
         max_tokens=2000,
         n_ctx=1024,
+        top_p=1,
         # n_threads=8,
         echo=False
     )