Hello, in here https://github.com/huggingface/transformers/blob/07708793f20ec3a949ccab32cc4fe0c7272dcc4c/src/transformers/modeling_bart.py#L588 I think the comment should be ``` # Convert to Bart output format: (BS, seq_len, model_dim) -> (seq_len, BS, model_dim) ``` Before transpose , shape of x and encoder_hidden_states are both (BS, seq_len, model_dim) to me. Thanks.