default attn_implementaion to eager to avoid issues

Signed-off-by: Zhiyu Cheng <[email protected]>

Author: Edwardf0t1

examples/llm_ptq/hf_ptq.py

@@ -748,10 +748,10 @@ def output_decode(generated_ids, input_shape):
     parser.add_argument(
         "--attn_implementation",
         help=(
-            "Specify the attention implementation to use."
+            "Specify the attention implementation to use. "
             "This arg will be passed to the HF model loading if specified."
         ),
-        default=None,
+        default="eager",
         type=str,
     )