Update translation.py

translation.py

@@ -12,7 +12,7 @@ def _load_default_model():
 @st.cache_resource
 def load_model(source_lang, target_lang):
     try:
-        if source_lang == target_lang:  # Avoid same language error
+        if source_lang == target_lang:
             return _load_default_model()
         model_name = f"Helsinki-NLP/opus-mt-{source_lang}-{target_lang}"
         tokenizer = MarianTokenizer.from_pretrained(model_name)
@@ -22,7 +22,7 @@ def load_model(source_lang, target_lang):
         st.warning(f"No direct model for {source_lang} to {target_lang}. Using cached en-fr.")
         return _load_default_model()
 
-@st.cache_data
+@st.cache_data(ttl=3600)  # Cache for 1 hour to improve speed
 def translate_cached(text, source_lang, target_lang):
     src_code = {"English": "en", "French": "fr", "Spanish": "es", "German": "de",
                 "Hindi": "hi", "Chinese": "zh", "Arabic": "ar", "Russian": "ru", "Japanese": "ja"}.get(source_lang, "en")
@@ -31,7 +31,7 @@ def translate_cached(text, source_lang, target_lang):
     tokenizer, model = load_model(src_code, tgt_code)
     inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=500)
     with torch.no_grad():
-        translated = model.generate(**inputs, max_length=500)
+        translated = model.generate(**inputs, max_length=500, num_beams=2, early_stopping=True)  # Beam search for speed
     return tokenizer.decode(translated[0], skip_special_tokens=True)
 
 def translate(text, source_lang, target_lang):