Update app.py

app.py

@@ -51,13 +51,20 @@ def predict(message, history):
     generate_kwargs = dict(
         model_inputs,
         streamer=streamer,
+        # max_new_tokens=300,
+        # do_sample=True,
+        # top_p=0.95,
+        # top_k=50,
+        # temperature=0.3,
+        # repetition_penalty=10.,
+        # num_beams=1,
         max_new_tokens=300,
         do_sample=True,
-        top_p=0.95,
-        top_k=50,
-        temperature=0.3,
-        repetition_penalty=10.,
+        top_k=5,
         num_beams=1,
+        use_cache=False,
+        temperature=0.4,
+        repetition_penalty=10.
         stopping_criteria=StoppingCriteriaList([stop])
     )
     t = Thread(target=model.generate, kwargs=generate_kwargs)