hankcs/HanLP · error · ValueError

Unrecognized type for {embed}

Error message

Unrecognized type for {embed}

What it means

CharCNN's embedding constructor only accepts embed as an int (vocab size + embedding dim); any other type is rejected. The int is used to build nn.Embedding(num_embeddings=vocab_size, embedding_dim=embed) which is then wrapped in TimeDistributed for character-level encoding. Passing an nn.Module or str bypasses that path and raises.

Source

Thrown at hanlp/layers/embeddings/char_cnn.py:67

                ngrams of size 2 to 5 with some number of filters.
            conv_layer_activation: `Activation`, optional (default=`torch.nn.ReLU`)
                Activation to use after the convolution layers.
            output_dim: After doing convolutions and pooling, we'll project the collected features into a vector of
                this size.  If this value is `None`, we will just return the result of the max pooling,
                giving an output of shape `len(ngram_filter_sizes) * num_filters`.
            vocab_size: The size of character vocab.

        Returns:
            A tensor of shape `(batch_size, output_dim)`.
        """
        super().__init__()
        EmbeddingDim.__init__(self)
        # the embedding layer
        if isinstance(embed, int):
            embed = nn.Embedding(num_embeddings=vocab_size,
                                 embedding_dim=embed)
        else:
            raise ValueError(f'Unrecognized type for {embed}')
        self.field = field
        self.embed = TimeDistributed(embed)
        self.encoder = TimeDistributed(
            CnnEncoder(embed.embedding_dim, num_filters, ngram_filter_sizes, conv_layer_activation, output_dim))
        self.embedding_dim = output_dim or num_filters * len(ngram_filter_sizes)

    def forward(self, batch: dict, **kwargs):
        tokens: torch.Tensor = batch[f'{self.field}_char_id']
        mask = tokens.ge(0)
        x = self.embed(tokens)
        return self.encoder(x, mask)

    def get_output_dim(self) -> int:
        return self.embedding_dim


class CharCNNEmbedding(Embedding, AutoConfigurable):
    def __init__(self,

View on GitHub (pinned to ddb1299bdd)

Solutions

1. Pass embed as an int, e.g. CharCNN(vocab, embed=50, ...)
2. If config-driven, coerce: embed=int(embed) before constructing
3. If you need a custom module, extend the class instead of passing a module today

Example fix

# before
embed = CharCNNEmbedding(vocab, embed='50')  # str -> error
# after
embed = CharCNNEmbedding(vocab, embed=50)

embed = int(embed) if isinstance(embed, (str, float)) else embed
assert isinstance(embed, int)

def is_valid_embed_arg(embed) -> bool:
    return isinstance(embed, int) and not isinstance(embed, bool)

When it happens

Trigger: Calling CharCNN(vocab_size, embed='100') or embed=nn.Embedding(...) or a config string parsed as non-int; also passing a float dimension.

Common situations: Loading a config from YAML/JSON where embed comes out as a string; reusing a pattern from char_rnn.py which also accepts nn.Module; typo'd config key yielding None.

Related errors

• Unrecognized type for {embed}
• Unsupported parameter type: {embed}
• error
• output ({}) must be of type bool or str
• Call fit or load before evaluate.

AI-assisted analysis of hankcs/HanLP@ddb1299bdd (2026-08-27). Data as JSON: /api/errors/09d5faa5ea0b2ce8. Report an issue: GitHub.