Update utils/translator.py

utils/translator.py

@@ -1,14 +1,15 @@
 # utils/translator.py
 
 import os
-from transformers import MarianMTModel, MarianTokenizer
+import torch
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
 DEEPL_API_KEY = os.getenv("DEEPL_API_KEY")
 
-# Fallback model (HF)
-model_name = "Helsinki-NLP/opus-mt-en-ROMANCE"
-tokenizer = MarianTokenizer.from_pretrained(model_name)
-model = MarianMTModel.from_pretrained(model_name)
+# ✅ Better fallback model (Brazilian Portuguese)
+model_name = "unicamp-dl/translation-en-pt-t5"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 
 def translate_text(text):
     if not text.strip():
@@ -22,13 +23,14 @@ def translate_text(text):
             data={
                 "auth_key": DEEPL_API_KEY,
                 "text": text,
-                "target_lang": "PT"
+                "target_lang": "PT-BR"  # 🟢 Brazil-specific
             },
         )
         return response.json()["translations"][0]["text"]
 
     except Exception:
-        # Use Hugging Face fallback model
-        inputs = tokenizer.encode(text, return_tensors="pt", truncation=True)
-        outputs = model.generate(inputs, max_length=512, num_beams=4)
+        # 🔁 Use HF fallback model
+        inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
+        with torch.no_grad():
+            outputs = model.generate(**inputs, max_length=512, num_beams=4)
         return tokenizer.decode(outputs[0], skip_special_tokens=True)