Update vlm.py

vlm.py

@@ -127,7 +127,11 @@ def build_messages(input_dict: dict, history: list[tuple]):
 #
 @spaces.GPU
 @torch.inference_mode()
-def stream_response(messages: list[dict]):
+def stream_response(
+        messages: list[dict],
+        max_new_tokens: int=1_024,
+        temperature: float=0.15
+    ):
     """Stream the model's response to the chat interface.
     
     Args:
@@ -148,7 +152,9 @@ def stream_response(messages: list[dict]):
     generation_args = dict(
         inputs,
         streamer=streamer,
-        max_new_tokens=2_048,
+        max_new_tokens=max_new_tokens,
+        temperature=temperature,
+        top_p=0.9,
         do_sample=True
     )