Update app.py

app.py

@@ -559,24 +559,21 @@ def get_tensor_length(tensor):
         ret = ret * num
     return ret
 
-def _summarize(text, max_words=20):
+def _summarize(text):
     log(f'CALL _summarize')
     prefix = "summarize: "
     toks = tokenizer.encode( prefix + text, return_tensors="pt", truncation=False)
     gen = model.generate(
         toks,
-        length_penalty=2.0,
-        num_beams=4,
+        length_penalty=3.0,
+        num_beams=math.ceil(len(text.split()) / 6),
         early_stopping=True,
-        max_length=max_words
+        max_length=math.ceil(len(text.split()) / 4)
     )
     ret = tokenizer.decode(gen[0], skip_special_tokens=True)
     log(f'RET _summarize with ret as {ret}')
     return ret
 
-def _summ_step(length):
-    return length // 1.5
-
 def summarize(text, max_words=20):
     log(f'CALL summarize')
     
@@ -603,9 +600,7 @@ def summarize(text, max_words=20):
             words_length = len(text.split())
     
     while words_length > max_words:
-        step = _summ_step(words_length)
-        mx = words_length - step
-        text = _summarize(text, mx)
+        text = _summarize(text)
         words_length = len(text.split())
         
     log(f'RET summarize with text as {text}')
@@ -1187,7 +1182,7 @@ class GoogleTranslator(BaseTranslator):
         @param text: desired text to translate
         @return: str: translated text
         """
-        if is_input_valid(text, max_chars=5000):
+        if is_input_valid(text, max_chars=1000):
             text = text.strip()
             if self._same_source_target() or is_empty(text):
                 return text
@@ -1274,7 +1269,7 @@ def translate(txt,to_lang="en",from_lang="auto"):
         translation = translator.translate(txt)
     translation = translation.strip()
     log(f'RET translate with translation as {translation}')
-    return translation
+    return translation.lower()
 
 @spaces.GPU(duration=120)
 def handle_generation(artist,song,genre,lyrics):
@@ -1285,11 +1280,14 @@ def handle_generation(artist,song,genre,lyrics):
     pos_song = re.sub(r"([ \t]){1,}", " ", song).lower().strip()
     pos_song = ' '.join(word[0].upper() + word[1:] for word in pos_song.split())
 
+    lyrics = re.sub(r"(, | ,)",". ",lyrics)
     lyrics_lines = re.split(r"([\n]){1,}", lyrics)
     
     for line_index in range(len(lyrics_lines)):
         if re.sub(f'[{punctuation}]$', '', lyrics_lines[line_index]) == lyrics_lines[line_index]:
-            lyrics_lines[line_index] = lyrics_lines[line_index] + "."
+            lyrics_lines[line_index] = lyrics_lines[line_index].strip() + "."
+        else:
+            lyrics_lines[line_index] = lyrics_lines[line_index].strip()
     lyrics = " ".join(lyrics_lines)
     
     pos_genre = re.sub(f'[{punctuation}]', '', re.sub(r"([ \t]){1,}", " ", genre)).lower().strip()