hankcs/HanLP · error · ValueError

You have to specify either decoder_input_ids or decoder_inpu

Error message

You have to specify either decoder_input_ids or decoder_inputs_embeds

What it means

This variant of __call__ (e.g. for multi-label / elementwise accuracy) requires gold_labels to have exactly the same shape as predictions. A mismatch (e.g. class indices of lower rank, or a different batch size) raises ValueError.

Source

Thrown at hanlp/components/amr/amrbart/model_interface/modeling_bart.py:1021

                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
        """
        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
        output_hidden_states = (
            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
        )
        use_cache = use_cache if use_cache is not None else self.config.use_cache
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        # retrieve input_ids and inputs_embeds
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
        elif input_ids is not None:
            input_shape = input_ids.size()
            input_ids = input_ids.view(-1, input_shape[-1])
        elif inputs_embeds is not None:
            input_shape = inputs_embeds.size()[:-1]
        else:
            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")

        # past_key_values_length
        past_key_values_length = past_key_values[0][0].shape[2] if past_key_values is not None else 0

        if inputs_embeds is None:
            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale

        attention_mask = self._prepare_decoder_attention_mask(
            attention_mask, input_shape, inputs_embeds, past_key_values_length
        )

        # expand encoder attention mask
        if encoder_hidden_states is not None and encoder_attention_mask is not None:
            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
            encoder_attention_mask = _expand_mask(encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1])

        # embed positions
        positions = self.embed_positions(input_shape, past_key_values_length)

View on GitHub (pinned to ddb1299bdd)

Solutions

1. Make gold the same shape as predictions (e.g. one-hot / multi-hot encode targets)
2. Verify batch alignment between predictions and targets
3. Use the index-style CategoricalAccuracy call path for single-label data

Example fix

# before
metric(predictions, gold_idx)  # (B,C) vs (B,)
# after
from torch.nn.functional import one_hot
metric(predictions, one_hot(gold_idx, num_classes=predictions.size(-1)))

assert gold_labels.size() == predictions.size(), (gold_labels.shape, predictions.shape)

import torch
def same_shape(a: torch.Tensor, b: torch.Tensor) -> bool:
    return a.size() == b.size()

try:
    metric(predictions, gold, mask)
except ValueError:
    if gold.dim() == predictions.dim() - 1:
        gold = torch.nn.functional.one_hot(gold, predictions.size(-1))
        metric(predictions, gold, mask)
    else:
        raise

When it happens

Trigger: Calling the metric with gold_labels.shape != predictions.shape — e.g. predictions (B, C) scores with gold (B,) indices, or mismatched batch sizes from misaligned batches.

Common situations: Using elementwise/multi-label metrics with single-label data or vice versa; batching bugs where predictions and targets come from different loaders; leftover code assuming index-style gold labels.

Related errors

• The head_mask should be specified for {len(self.layers)} lay
• The `{mask_name}` should be specified for {len(self.layers)}
• Unsupported argument type: {item}
• Attention weights should be of size {(bsz * self.num_heads,
• You cannot specify both input_ids and inputs_embeds at the s

AI-assisted analysis of hankcs/HanLP@ddb1299bdd (2026-08-27). Data as JSON: /api/errors/19211abd47dea194. Report an issue: GitHub.