Update converse.py

converse.py

@@ -23,6 +23,8 @@ def discussion(
     top_k=50,
     temperature=0.7,
     full_text=False,
+    length_penalty=0.8,
+    no_repeat_ngram_size=2,
     num_return_sequences=1,
     device=-1,
     verbose=False,
@@ -73,6 +75,8 @@ def discussion(
         top_k=top_k,
         temperature=temperature,
         full_text=full_text,
+        no_repeat_ngram_size=no_repeat_ngram_size,
+        length_penalty=length_penalty,
         num_return_sequences=num_return_sequences,
         device=device,
         verbose=verbose,
@@ -106,13 +110,15 @@ def gen_response(
     responder: str,
     timeout=45,
     min_length=4,
-    max_length=64,
+    max_length=48,
     top_p=0.95,
-    top_k=50,
-    temperature=0.7,
+    top_k=20,
+    temperature=0.5,
     full_text=False,
     num_return_sequences=1,
-    length_penalty=0.6,
+    length_penalty:float=0.8,
+    repetition_penalty:float=3.5,
+    no_repeat_ngram_size=2,
     device=-1,
     verbose=False,
     **kwargs,
@@ -158,8 +164,8 @@ def gen_response(
         num_return_sequences=num_return_sequences,
         max_time=timeout,
         return_full_text=full_text,
-        no_repeat_ngram_size=2,
-        repetition_penalty=3.5,
+        no_repeat_ngram_size=no_repeat_ngram_size,
+        repetition_penalty=repetition_penalty,
         length_penalty=length_penalty,
         clean_up_tokenization_spaces=True,
         remove_invalid_values=True,
@@ -168,7 +174,6 @@ def gen_response(
     rt = round(time.perf_counter() - st, 2)
     if verbose:
         print(f"took {rt} sec to respond")
-
     if verbose:
         print("\n[DEBUG] generated:\n")
         pp.pprint(response)  # for debugging
@@ -186,7 +191,6 @@ def gen_response(
     )
     if verbose:
         print(f"DEBUG: {bot_dialogue} was original response pre-SC")
-
     return bot_dialogue  #
 
 
@@ -251,5 +255,4 @@ def consolidate_texts(
         print("\nthe full response is:\n")
         print("\n".join(fn_resp))
         print("--" * 10)
-
     return fn_resp