The vocabulary size of the `token_type_ids` passed when calling [`RealmEmbedder`], [`RealmScorer`],
            [`RealmKnowledgeAugEncoder`], or [`RealmReader`].
        initializer_range (`float`, *optional*, defaults to 0.02):
            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
        layer_norm_eps (`float`, *optional*, defaults to 1e-12):
            The epsilon used by the layer normalization layers.
        span_hidden_size (`int`, *optional*, defaults to 256):
            Dimension of the reader's spans.
        max_span_width (`int`, *optional*, defaults to 10):
            Max span width of the reader.
        reader_layer_norm_eps (`float`, *optional*, defaults to 1e-3):
            The epsilon used by the reader's layer normalization layers.
        reader_beam_size (`int`, *optional*, defaults to 5):
            Beam size of the reader.
        reader_seq_len (`int`, *optional*, defaults to 288+32):
            Maximum sequence length of the reader.
        num_block_records (`int`, *optional*, defaults to 13353718):
            Number of block records.
        searcher_beam_size (`int`, *optional*, defaults to 5000):
            Beam size of the searcher. Note that when eval mode is enabled, *searcher_beam_size* will be the same as
            *reader_beam_size*.

    Example:

    ```python
    >>> from transformers import RealmConfig, RealmEmbedder

    >>> # Initializing a REALM realm-cc-news-pretrained-* style configuration
    >>> configuration = RealmConfig()

    >>> # Initializing a model (with random weights) from the google/realm-cc-news-pretrained-embedder style configuration
    >>> model = RealmEmbedder(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```Ú