Is it possible to support dynamic length input like:
l_input_ids = keras.layers.Input(shape=(None,), dtype='int32') l_token_type_ids = keras.layers.Input(shape=(None,), dtype='int32')
Padding every single example makes the inference quite slower, especially when the length of inputs vary a lot.
sataliulan
commented
4 years ago
Instead of fixed length input:
max_seq_len = 128 l_input_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32') l_token_type_ids = keras.layers.Input(shape=(max_seq_len,), dtype='int32')Is it possible to support dynamic length input like:
l_input_ids = keras.layers.Input(shape=(None,), dtype='int32') l_token_type_ids = keras.layers.Input(shape=(None,), dtype='int32')Padding every single example makes the inference quite slower, especially when the length of inputs vary a lot.