Update app.py

app.py

@@ -6,13 +6,18 @@ from collections import defaultdict
 
 app = FastAPI(title="Mariza + Qwen3-0.6B CPU Free")
 
-print("Carregando Qwen3-0.6B em fp16 puro na CPU... (2-4 min na primeira vez)")
+print("Carregando Qwen3-0.6B (forçando slow tokenizer pra burlar o bug do tokenizers)")
 
-tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-0.6B", trust_remote_code=True)
+# <<< AS DUAS LINHAS MÁGICAS >>>
+tokenizer = AutoTokenizer.from_pretrained(
+    "Qwen/Qwen3-0.6B",
+    trust_remote_code=True,
+    use_fast=False          # <─ ESSA LINHA SALVA TUDO
+)
 
 model = AutoModelForCausalLM.from_pretrained(
     "Qwen/Qwen3-0.6B",
-    torch_dtype="auto",        # deixa o transformers escolher fp16/float16
+    torch_dtype="auto",
     device_map="cpu",
     trust_remote_code=True,
     low_cpu_mem_usage=True
@@ -23,7 +28,7 @@ MAX_CONTEXT_TOKENS = 3800
 
 @app.get("/")
 async def root():
-    return {"message": "Qwen3-0.6B tá vivo e quente na CPU free, chefe! Sem quantização, sem dor de cabeça 😈"}
+    return {"message": "Qwen3-0.6B tá vivo e foda na CPU free, chefe! (sem bug do tokenizers) 😈"}
 
 @app.post("/chat")
 async def chat(request: Request):
@@ -35,9 +40,8 @@ async def chat(request: Request):
     stream = data.get("stream", False)
 
     if not prompt:
-        return JSONResponse({"error": "prompt vazio, seu safado"})
+        return JSONResponse({"error": "prompt vazio, safado"})
 
-    # Monta histórico
     messages = history_db[user_id]
     full_prompt = ""
     for role, content in messages:
@@ -82,4 +86,4 @@ async def chat(request: Request):
 
         return JSONResponse({"response": resposta})
 
-print("Qwen3-0.6B carregado! Pode mandar o zap que Mariza tá pronta pra responder 24/7 😏")
+print("Qwen3-0.6B carregado e pronto pra dominar o WhatsApp 24/7 de graça, chefe! 🔥")