bug fixes

gemmademo/_chat.py

@@ -21,8 +21,7 @@ class GradioChat:
         self.current_model_name = "gemma-3b"
         self.current_task_name = "Question Answering"
 
-        # Load model lazily on first use instead of at initialization
-        self.model = None
+        self.model = self._load_model("gemma-3b")
         self.prompt_manager = self._load_task(self.current_task_name)
 
         # Cache.

gemmademo/_model.py

@@ -51,11 +51,11 @@ class LlamaCppGemmaModel:
         self.messages = []
 
         # Model response generation attributes
-        self.max_tokens = (512,)
-        self.temperature = (0.7,)
-        self.top_p = (0.95,)
-        self.top_k = (40,)
-        self.repeat_penalty = (1.1,)
+        self.max_tokens = 512
+        self.temperature = 0.7
+        self.top_p = 0.95
+        self.top_k = 40
+        self.repeat_penalty = 1.1
 
     def load_model(self, n_ctx: int = 2048, n_gpu_layers: int = 0, system_prompt=""):
         """