reduce logging verbosity

I think you forgot to transfer over the logging change too from https://huggingface.co/LnL-AI/dbrx-base-converted/discussions/1

modeling_dbrx.py

@@ -412,7 +412,7 @@ class DbrxFlashAttention2(DbrxAttention):
         **kwargs: Any,
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor],
                Optional[Tuple[torch.Tensor]]]:
-        logger.info(
+        logger.debug(
             'Implicitly setting `output_attentions` to False as it is not supported in Flash Attention.'
         )
         output_attentions = False