Количество параметров подсчета в ГРУ

Question

У меня модель ГРУ выглядит следующим образом.

      class CharGenModel(tf.keras.Model):

    def __init__(self, vocab_size, num_timesteps, embedding_dim, **kwargs):
        super(CharGenModel, self).__init__(**kwargs)
        self.embedding_layer = tf.keras.layers.Embedding(vocab_size, embedding_dim)
        self.rnn_layer = tf.keras.layers.GRU(
            num_timesteps,
            recurrent_initializer="glorot_uniform",
            recurrent_activation="sigmoid",
            stateful=True,
            return_sequences=True
        )
        self.dense_layer = tf.keras.layers.Dense(vocab_size)

    def call(self, x):
        print(x.shape)
        x = self.embedding_layer(x)
        print(x.shape)
        x = self.rnn_layer(x)
        print(x.shape)
        x = self.dense_layer(x)
        print(x.shape)
        return x

vocab_size = 92
embedding_dim = 256
seq_length = 100
batch_size = 64
model = CharGenModel(vocab_size, seq_length, embedding_dim)
model.build(input_shape=(batch_size, seq_length))
model.summary()

model.summary() произвел количество обучаемых параметров следующим образом.

      (64, 100)
(64, 100, 256)
(64, 100, 100)
(64, 100, 92)
Model: "char_gen_model_4"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 embedding_4 (Embedding)     multiple                  23552     
                                                                 
 gru_4 (GRU)                 multiple                  107400    
                                                                 
 dense_4 (Dense)             multiple                  9292      
                                                                 
=================================================================
Total params: 140,244
Trainable params: 140,244
Non-trainable params: 0

Я смущен двумя вещами.

В соответствии с определением слоя внедрения

      tf.keras.layers.Embedding(
    input_dim,
    output_dim,
    embeddings_initializer='uniform',
    embeddings_regularizer=None,
    activity_regularizer=None,
    embeddings_constraint=None,
    mask_zero=False,
    input_length=None,
    **kwargs
)
input_dim for embedding layer for my application is 64x100.
(1)But why embedding layer trainable parameters are 92x256=23552. Why not 100x256?
(2)Number of parameters counting for GRU is 
num_params = number of FFNNs × [number of hidden units x (number of hidden units+number of inputs) + number of bias]
number of FFNNs(Number of feedforward networks) in GRU is 3
number of hidden units is 100
number of inputs is 256
number of bias is 100
so num_params = 3 x [100x(100+256)+100] = 107100
But model summary output is 107400

Где я ошибся в расчетах?