fix app legacy

app.py

@@ -1,8 +1,7 @@
 from fastapi import FastAPI
-from transformers import pipeline
+from transformers import AutoTokenizer, T5ForConditionalGeneration
 import json
 import os
-from huggingface_hub import HfApi
 
 app = FastAPI()
 
@@ -13,27 +12,32 @@ try:
 except FileNotFoundError:
     examples = []
 
-# Função para verificar e carregar o modelo
+# Função para carregar o modelo e tokenizer
 def load_model():
     try:
-        return pipeline(
-            "text2text-generation",
-            model="unicamp-dl/ptt5-base-portuguese-vocab",
-            tokenizer="unicamp-dl/ptt5-base-portuguese-vocab",
-            device_map="auto" if os.getenv("HF_TOKEN") else None,
-            model_kwargs={"load_in_8bit": True if os.getenv("HF_TOKEN") else False},
-            trust_remote_code=True  # Permite carregar código remoto, se necessário
+        tokenizer = AutoTokenizer.from_pretrained(
+            "unicamp-dl/ptt5-base-portuguese-vocab",
+            legacy=False,  # Usa o novo comportamento do tokenizer
+            clean_up_tokenization_spaces=True  # Define explicitamente para evitar warnings futuros
         )
+        model = T5ForConditionalGeneration.from_pretrained(
+            "unicamp-dl/ptt5-base-portuguese-vocab",
+            device_map="auto" if os.getenv("HF_TOKEN") else None
+        )
+        return {"tokenizer": tokenizer, "model": model}
     except Exception as e:
         print(f"Erro ao carregar o modelo: {e}")
         return None
 
-# Inicializa o modelo
-ptt5 = load_model()
+# Inicializa o modelo e tokenizer
+model_data = load_model()
 
 def generate_question_from_prompt(theme, difficulty, example_question=None):
-    if not ptt5:
-        return {"question": "Erro: Modelo não carregado.", "options": [], "answer": "", "explanation": "Por favor, verifique os logs."}
+    if not model_data or not model_data["tokenizer"] or not model_data["model"]:
+        return {"question": "Erro: Modelo ou tokenizer não carregado.", "options": [], "answer": "", "explanation": "Por favor, verifique os logs."}
+
+    tokenizer = model_data["tokenizer"]
+    model = model_data["model"]
 
     if example_question:
         example_text = (
@@ -58,7 +62,9 @@ def generate_question_from_prompt(theme, difficulty, example_question=None):
             "Explicação: [texto].'"
         )
     try:
-        response = ptt5(prompt, max_new_tokens=512, temperature=0.7, top_p=0.9)[0]['generated_text']
+        inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=512)
+        outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, top_p=0.9)
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
         # Parseia a resposta para extrair os componentes
         parts = response.split("Alternativas:")
         if len(parts) > 1:

requirements.txt

@@ -3,4 +3,4 @@ uvicorn==0.23.2
 transformers==4.45.0
 accelerate==0.21.0
 huggingface_hub
-sentencepiece  # Adicionada para suportar o tokenizer do PTT5
+sentencepiece