Update app.py

app.py

@@ -558,24 +558,45 @@ def summarize(
     text, max_len=20, min_len=10
 ):
     log(f'CALL summarize')
-    inputs = tokenizer.encode("summarize: " + text, return_tensors="pt", max_length=float('inf'), truncation=False)
-    if get_tensor_length(inputs) < 3:
-        print("Summarization Error: Text is too short, 3 words minimum!")
+    
+    words = text.split()
+    
+    if get_tensor_length(words) < 5:
+        print("Summarization Error: Text is too short, 5 words minimum!")
         return text
-    i = 1
-    while get_tensor_length(inputs) > max_len:
-        print(f'DBG summarize 1 {i}')
-        outputs = model.generate(
-            torch.tensor(list(inputs[0][:512])),
-            length_penalty=2.0,
-            num_beams=min(4,get_tensor_length(inputs) - 1),
-            early_stopping=True,
-            max_length=max( get_tensor_length(inputs) // 4 , max_len ),
-            min_length=min_len
-        )
-        inputs = torch.tensor([[*list(outputs[0]), *list(inputs[0][512:])]])
-        i = i + 1
-    summary = tokenizer.decode(inputs[0], skip_special_tokens=True)
+        
+    prefix = "summarize: "
+    ret = ""
+
+    for index in math.ceil( len(words) / 512 ):
+        
+        chunk = " ".join(words[ index*512:(index+1)*512 ])
+        inputs = tokenizer.encode( prefix + chunk, return_tensors="pt", max_length=float('inf'), truncation=False)
+        
+        while get_tensor_length(inputs) > max_len:
+            
+            inputs = model.generate(
+                inputs,
+                length_penalty=2.0,
+                num_beams=4,
+                early_stopping=True,
+                max_length=max( get_tensor_length(inputs) // 4 , max_len ),
+                min_length=min_len
+            )
+
+        toks = tokenizer.decode(inputs[0], skip_special_tokens=True)
+        ret = ret + ("" if ret == "" else " ") + toks
+
+    inputs = tokenizer.encode( prefix + ret, return_tensors="pt", max_length=float('inf'), truncation=False)
+    gen = model.generate(
+        inputs,
+        length_penalty=1.0,
+        num_beams=4,
+        early_stopping=True,
+        max_length=max_len,
+        min_length=min_len
+    )
+    summary = tokenizer.decode(gen[0], skip_special_tokens=True)
     log(f'RET summarize with summary as {summary}')
     return summary
 
@@ -633,6 +654,64 @@ def all_pipes(pos,neg,artist,song):
 
     return imgs
 
+language_codes = {
+    "af": "Afrikaans",
+    "ar": "Arabic",
+    "bg": "Bulgarian",
+    "bn": "Bengali",
+    "ca": "Catalan",
+    "cs": "Czech",
+    "cy": "Welsh",
+    "da": "Danish",
+    "de": "German",
+    "el": "Greek",
+    "en": "English",
+    "es": "Spanish",
+    "et": "Estonian",
+    "fa": "Persian (Farsi)",
+    "fi": "Finnish",
+    "fr": "French",
+    "gu": "Gujarati",
+    "he": "Hebrew",
+    "hi": "Hindi",
+    "hr": "Croatian",
+    "hu": "Hungarian",
+    "id": "Indonesian",
+    "it": "Italian",
+    "ja": "Japanese",
+    "kn": "Kannada",
+    "ko": "Korean",
+    "lt": "Lithuanian",
+    "lv": "Latvian",
+    "mk": "Macedonian",
+    "ml": "Malayalam",
+    "mr": "Marathi",
+    "ne": "Nepali",
+    "nl": "Dutch",
+    "no": "Norwegian",
+    "pa": "Punjabi",
+    "pl": "Polish",
+    "pt": "Portuguese",
+    "ro": "Romanian",
+    "ru": "Russian",
+    "sk": "Slovak",
+    "sl": "Slovenian",
+    "so": "Somali",
+    "sq": "Albanian",
+    "sv": "Swedish",
+    "sw": "Swahili",
+    "ta": "Tamil",
+    "te": "Telugu",
+    "th": "Thai",
+    "tl": "Tagalog (Filipino)",
+    "tr": "Turkish",
+    "uk": "Ukrainian",
+    "ur": "Urdu",
+    "vi": "Vietnamese",
+    "zh-cn": "Chinese (Simplified)",
+    "zh-tw": "Chinese (Traditional)",
+}
+
 def translate(txt,to_lang="en",from_lang=False):
     log(f'CALL translate')
     if not from_lang:
@@ -640,12 +719,13 @@ def translate(txt,to_lang="en",from_lang=False):
     if(from_lang == to_lang):
         log(f'RET translate with txt as {txt}')
         return txt
-    inputs = tokenizer.encode(f"translate {from_lang} to {to_lang}: " + txt, return_tensors="pt", max_length=float('inf'), truncation=False)
-    chunks_length = math.ceil(get_tensor_length(inputs) / 512)
+    prefix = f"translate {language_codes[from_lang]} to {language_codes[to_lang]}: "
+    words = txt.split()
     ret = ""
-    for index in range(chunks_length):
-        chunk = torch.tensor([list(inputs[0][ index*512:index*512+512 ])])
-        gen = model.generate(chunk)
+    for index in math.ceil( len(words) / 512 ):
+        chunk = " ".join(words[ index*512:(index+1)*512 ])
+        inputs = tokenizer.encode(prefix+chunk, return_tensors="pt", max_length=float('inf'), truncation=False)
+        gen = model.generate(chunk,input)
         toks = tokenizer.decode(gen[0], skip_special_tokens=True)
         ret = ret + ("" if ret == "" else " ") + toks
     log(f'RET translate with ret as {ret}')