Update app.py

app.py

@@ -26,7 +26,7 @@ print("🔄 Loading model...")
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_ID,
     token=HF_TOKEN,
-    torch_dtype=torch.float32,
+    torch_dtype=torch.float32,  # pode trocar pra bfloat16 se tiver GPU
     device_map="auto"
 )
 model.eval()
@@ -51,7 +51,7 @@ def chat(user_input, system_prompt, temperature, top_p, max_tokens):
         return_tensors="pt"
     ).to(model.device)
 
-        with torch.no_grad():
+    with torch.no_grad():
         output = model.generate(
             **inputs,
             max_new_tokens=int(max_tokens),
@@ -63,14 +63,14 @@ def chat(user_input, system_prompt, temperature, top_p, max_tokens):
             pad_token_id=tokenizer.eos_token_id
         )
 
-generated_tokens = output[0][inputs["input_ids"].shape[-1]:]
+    generated_tokens = output[0][inputs["input_ids"].shape[-1]:]
 
-decoded = tokenizer.decode(
-    generated_tokens,
-    skip_special_tokens=True
-)
+    decoded = tokenizer.decode(
+        generated_tokens,
+        skip_special_tokens=True
+    )
 
-return decoded.strip()
+    return decoded.strip()
 
 # =========================
 # GRADIO UI