Update model.py

model.py

@@ -81,7 +81,7 @@ def get_input_token_length(message: str, chat_history: list[tuple[str, str]], sy
 def run(message: str,
         chat_history: list[tuple[str, str]],
         system_prompt: str,
-        max_new_tokens: int = 256,
+        max_new_tokens: int = 1024,
         temperature: float = 0.8,
         top_p: float = 0.95,
         top_k: int = 50) -> Iterator[str]:
@@ -107,5 +107,11 @@ def run(message: str,
 
     outputs = []
     for text in streamer:
-        outputs.append(text)
-        yield ''.join(outputs)
+        if "instruction:" in text:
+            # Append only the part of text before "instruction:" and stop streaming
+            outputs.append(text.split("instruction:")[0])
+            break
+        else:
+            outputs.append(text)
+        
+        yield ''.join(outputs)