modeling_deformable_detr.py DeformableDetrMultiheadAttention.foward function report error for "hidden_states_original" if position_embeddings is None

[susanbao]

System Info

src/transformers/models/deformable_detr/modeling_deformable_detr.py

class DeformableDetrMultiheadAttention(nn.Module)

def forward(
    self,
    hidden_states: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    position_embeddings: Optional[torch.Tensor] = None,
    output_attentions: bool = False,
) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
    """Input shape: Batch x Time x Channel"""

    batch_size, target_len, embed_dim = hidden_states.size()
    # add position embeddings to the hidden states before projecting to queries and keys
    if position_embeddings is not None:
        hidden_states_original = hidden_states
        hidden_states = self.with_pos_embed(hidden_states, position_embeddings)

    # get queries, keys and values
    query_states = self.q_proj(hidden_states) * self.scaling
    key_states = self._shape(self.k_proj(hidden_states), -1, batch_size)
    value_states = self._shape(self.v_proj(hidden_states_original), -1, batch_size)
     ....

If position_embeddings is None, the "hidden_states_original" won't be created. Then use it on "value_states = self._shape(self.v_proj(hidden_states_original), -1, batch_size)" reports error "UnboundLocalError: local variable 'hidden_states_original' referenced before assignment".

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Set the input position_embeddings of DeformableDetrMultiheadAttention.forward be None

Expected behavior

Report error "UnboundLocalError: local variable 'hidden_states_original' referenced before assignment"

[Rocketknight1]

cc @NielsRogge for DeformableDetr!