Use `attn_implementation` instead of `_attn_implementation`

README.md

@@ -84,7 +84,7 @@ processor = AutoProcessor.from_pretrained(model_path)
 model = AutoModelForImageTextToText.from_pretrained(
     model_path,
     torch_dtype=torch.bfloat16,
-    _attn_implementation="flash_attention_2"
+    attn_implementation="flash_attention_2"
 ).to("cuda")
 ```