Update app.py

app.py

@@ -102,7 +102,7 @@ def model_inference(input_dict, history):
         ).to("cuda")
         # Set up streaming generation.
         streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-        generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
+        generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=2048)
         thread = Thread(target=model.generate, kwargs=generation_kwargs)
         thread.start()
         buffer = ""
@@ -144,7 +144,7 @@ def model_inference(input_dict, history):
         padding=True,
     ).to("cuda")
     streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
-    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
+    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=2048)
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()
     buffer = ""