hankcs/HanLP · error · ValueError

Attention mask should be of size {(bsz, 1, tgt_len, src_len)

Error message

Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}

What it means

TimeDistributed applies a module to every time step by squashing (batch, time, ...) into (batch*time, ...). If no tensor input survived reshaping (some_input is None — all inputs were non-tensors/None), it cannot infer the output batch/time shape and raises RuntimeError.

Source

Thrown at hanlp/components/amr/amrbart/model_interface/modeling_bart.py:238

            past_key_value = (key_states, value_states)

        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
        query_states = self._shape(query_states, tgt_len, bsz).view(*proj_shape)
        key_states = key_states.view(*proj_shape)
        value_states = value_states.view(*proj_shape)

        src_len = key_states.size(1)
        attn_weights = torch.bmm(query_states, key_states.transpose(1, 2))

        if attn_weights.size() != (bsz * self.num_heads, tgt_len, src_len):
            raise ValueError(
                f"Attention weights should be of size {(bsz * self.num_heads, tgt_len, src_len)}, but is"
                f" {attn_weights.size()}"
            )

        if attention_mask is not None:
            if attention_mask.size() != (bsz, 1, tgt_len, src_len):
                raise ValueError(
                    f"Attention mask should be of size {(bsz, 1, tgt_len, src_len)}, but is {attention_mask.size()}"
                )
            attn_weights = attn_weights.view(bsz, self.num_heads, tgt_len, src_len) + attention_mask
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        attn_weights = nn.functional.softmax(attn_weights, dim=-1)

        if layer_head_mask is not None:
            if layer_head_mask.size() != (self.num_heads,):
                raise ValueError(
                    f"Head mask for a single layer should be of size {(self.num_heads,)}, but is"
                    f" {layer_head_mask.size()}"
                )
            attn_weights = layer_head_mask.view(1, -1, 1, 1) * attn_weights.view(bsz, self.num_heads, tgt_len, src_len)
            attn_weights = attn_weights.view(bsz * self.num_heads, tgt_len, src_len)

        if output_attentions:
            # this operation is a bit awkward, but it's required to

View on GitHub (pinned to ddb1299bdd)

Solutions

1. Ensure at least one torch.Tensor input with a batch dimension reaches the TimeDistributed module
2. Filter/convert non-tensor fields to embeddings before the time-distributed layer
3. Guard the batch for empty/None samples before forward

Example fix

# before
layer = TimeDistributed(nn.Linear(10, 5))
out = layer(meta_strings)  # no tensor input
# after
out = layer(embedded_tokens)  # shape (B, T, 10)

import torch
assert any(isinstance(x, torch.Tensor) and x.dim() >= 3 for x in inputs), 'TimeDistributed needs a (B,T,*) tensor input'

def has_tensor_input(args) -> bool:
    return any(isinstance(a, torch.Tensor) for a in args)

try:
    out = layer(*inputs)
except RuntimeError as e:
    if 'time-distribute' in str(e):
        raise ValueError('missing tensor input') from e
    raise

When it happens

Trigger: Calling a TimeDistributed-wrapped module with only non-tensor arguments (strings, lists, None) so that every reshaped input is skipped and some_input stays None.

Common situations: Passing string features or None placeholders (e.g. padded non-numeric fields) through a TimeDistributed layer; refactoring input pipelines so the only tensor argument is dropped; empty batch collation producing None fields.

Related errors

• Head mask for a single layer should be of size {(self.num_he
• error
• output ({}) must be of type bool or str
• Call fit or load before evaluate.
• Unrecognized devices {devices}

AI-assisted analysis of hankcs/HanLP@ddb1299bdd (2026-08-27). Data as JSON: /api/errors/c272c543ffbab120. Report an issue: GitHub.