ginipick commited on
Commit
67016e4
·
verified ·
1 Parent(s): 1c3c162

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +49 -41
app.py CHANGED
@@ -753,8 +753,56 @@ ko_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
753
  ja_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ja-en")
754
  zh_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-zh-en")
755
 
756
- from transformers import MarianMTModel, MarianTokenizer
757
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
758
  def translate_text(text, src_lang, model_name):
759
  try:
760
  tokenizer = MarianTokenizer.from_pretrained(model_name)
@@ -778,46 +826,6 @@ def translate_if_needed(prompt):
778
  return translate_text(prompt, 'zh', 'Helsinki-NLP/opus-mt-zh-en')
779
  return prompt
780
 
781
- @spaces.GPU
782
- @torch.no_grad()
783
- def generate_image(
784
- prompt, width, height, guidance, inference_steps, seed,
785
- do_img2img, init_image, image2image_strength, resize_img,
786
- progress=gr.Progress(track_tqdm=True),
787
- ):
788
- translated_prompt = translate_if_needed(prompt)
789
- if translated_prompt != prompt:
790
- print(f"Translated prompt: {translated_prompt}")
791
- prompt = translated_prompt
792
-
793
-
794
- # 한글, 일본어, 중국어 문자 감지
795
- def contains_korean(text):
796
- return any('\u3131' <= c <= '\u318E' or '\uAC00' <= c <= '\uD7A3' for c in text)
797
-
798
- def contains_japanese(text):
799
- return any('\u3040' <= c <= '\u309F' or '\u30A0' <= c <= '\u30FF' or '\u4E00' <= c <= '\u9FFF' for c in text)
800
-
801
- def contains_chinese(text):
802
- return any('\u4e00' <= c <= '\u9fff' for c in text)
803
-
804
- # 한글, 일본어, 중국어가 있으면 번역
805
- if contains_korean(prompt):
806
- translated_prompt = ko_translator(prompt, max_length=512)[0]['translation_text']
807
- print(f"Translated Korean prompt: {translated_prompt}")
808
- prompt = translated_prompt
809
- elif contains_japanese(prompt):
810
- translated_prompt = ja_translator(prompt, max_length=512)[0]['translation_text']
811
- print(f"Translated Japanese prompt: {translated_prompt}")
812
- prompt = translated_prompt
813
- elif contains_chinese(prompt):
814
- translated_prompt = zh_translator(prompt, max_length=512)[0]['translation_text']
815
- print(f"Translated Chinese prompt: {translated_prompt}")
816
- prompt = translated_prompt
817
-
818
-
819
-
820
-
821
 
822
  if seed == 0:
823
  seed = int(random.random() * 1000000)
 
753
  ja_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ja-en")
754
  zh_translator = pipeline("translation", model="Helsinki-NLP/opus-mt-zh-en")
755
 
 
756
 
757
+ from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
758
+
759
+ def translate_text(text):
760
+ try:
761
+ # M2M100은 다국어 번역을 한 모델로 처리할 수 있습니다
762
+ model_name = "facebook/m2m100_418M"
763
+ tokenizer = M2M100Tokenizer.from_pretrained(model_name)
764
+ model = M2M100ForConditionalGeneration.from_pretrained(model_name).to(device)
765
+
766
+ # 언어 감지
767
+ def detect_language(text):
768
+ if any('\u3131' <= c <= '\u318E' or '\uAC00' <= c <= '\uD7A3' for c in text):
769
+ return 'ko'
770
+ elif any('\u3040' <= c <= '\u309F' or '\u30A0' <= c <= '\u30FF' for c in text):
771
+ return 'ja'
772
+ elif any('\u4e00' <= c <= '\u9fff' for c in text):
773
+ return 'zh'
774
+ return None
775
+
776
+ src_lang = detect_language(text)
777
+ if src_lang is None:
778
+ return text
779
+
780
+ tokenizer.src_lang = src_lang
781
+ encoded = tokenizer(text, return_tensors="pt").to(device)
782
+ generated_tokens = model.generate(
783
+ **encoded,
784
+ forced_bos_token_id=tokenizer.get_lang_id("en"),
785
+ max_length=128
786
+ )
787
+ return tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
788
+ except Exception as e:
789
+ print(f"Translation error: {e}")
790
+ return text
791
+
792
+ @spaces.GPU
793
+ @torch.no_grad()
794
+ def generate_image(
795
+ prompt, width, height, guidance, inference_steps, seed,
796
+ do_img2img, init_image, image2image_strength, resize_img,
797
+ progress=gr.Progress(track_tqdm=True),
798
+ ):
799
+ translated_prompt = translate_text(prompt)
800
+ if translated_prompt != prompt:
801
+ print(f"Translated prompt: {translated_prompt}")
802
+ prompt = translated_prompt
803
+
804
+ if seed == 0:
805
+ seed = int(random.random() * 1000000)
806
  def translate_text(text, src_lang, model_name):
807
  try:
808
  tokenizer = MarianTokenizer.from_pretrained(model_name)
 
826
  return translate_text(prompt, 'zh', 'Helsinki-NLP/opus-mt-zh-en')
827
  return prompt
828
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
829
 
830
  if seed == 0:
831
  seed = int(random.random() * 1000000)