:param x: The attention output. It should be [batch_size*seq_length, dim] :param intermediate_size: the hidden projection. By default this is the input_dim * 4. in the original GPT we would return layer_norm(x_norm + h1) rather than layer_norm(x + h1) :return:
(x_flat, intermediate_size, initializer_range=0.02, hidden_dropout_prob=0.1)
source not stored for this graph (policy: none)
no test coverage detected