adjust params

app.py

@@ -78,9 +78,9 @@ def ask_gpt(
     chat_pipe,
     speaker="person alpha",
     responder="person beta",
-    max_len=128,
+    max_len=96,
     top_p=0.95,
-    top_k=50,
+    top_k=25,
     temperature=0.6,
 ):
     """

converse.py

@@ -112,6 +112,7 @@ def gen_response(
     temperature=0.7,
     full_text=False,
     num_return_sequences=1,
+    length_penalty=0.6,
     device=-1,
     verbose=False,
     **kwargs,
@@ -155,10 +156,12 @@ def gen_response(
         num_return_sequences=num_return_sequences,
         max_time=timeout,
         return_full_text=full_text,
-        no_repeat_ngram_size=3,
-        length_penalty=0.3,
-        repetition_penalty=3.4,
+        no_repeat_ngram_size=2,
+        repetition_penalty=3.5,
+        length_penalty=length_penalty,
         clean_up_tokenization_spaces=True,
+        skip_special_tokens=True,
+        remove_invalid_values=True,
         **kwargs,
     )  # the likely better beam-less method
     rt = round(time.perf_counter() - st, 2)