Update app.py

app.py

@@ -1,25 +1,35 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, pipeline
 
-# Carregar o modelo direto do HuggingFace Hub
-model_id = "microsoft/phi-2"
+# Modelo já quantizado
+model_id = "TheBloke/phi-2-GPTQ"
+
+# Configuração para 4-bit/8-bit quantization
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype="float16",
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4"
+)
+
+# Carregar tokenizer e modelo
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(
     model_id,
     device_map="auto",
-    offload_folder="offload"
+    quantization_config=bnb_config
 )
 
-# Pipeline de geração de texto
+# Pipeline de texto
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-# Função do chat com salvamento de memória
+# Função do chat + salvar memória
 def chat(user_input, history):
     prompt = user_input
     result = pipe(prompt, max_new_tokens=256, temperature=0.7)[0]["generated_text"]
 
     # Salvar memória em arquivo
-    with open("log.txt", "a", encoding="utf-8") as f:
+    with open("memoria.txt", "a", encoding="utf-8") as f:
         f.write(f"User: {user_input}\nAI: {result}\n")
 
     return result