Bugfix for max_new_tokens

app.py

@@ -258,7 +258,11 @@ class QwenModel(EvalModel):
             skip_special_tokens=True,
         )
 
-        generate_kwargs = dict(**model_inputs)
+        generate_kwargs = dict(
+            **model_inputs,
+            streamer=streamer,
+            max_new_tokens=self.max_output_tokens,
+        )
         t = Thread(target=self.model.generate, kwargs=generate_kwargs)
         t.start()