# T5.1.1 Efficient small nl16 model. | |
import seqio | |
include 't5x/examples/t5/t5_1_1/base.gin' # imports vocab, optimizer and model. | |
# ------------------- Network specification overrides -------------------------- | |
network.Transformer.config = @network.T5Config() | |
network.T5Config: | |
emb_dim = 512 | |
num_heads = 8 | |
num_encoder_layers = 16 | |
num_decoder_layers = 16 | |
head_dim = 64 | |
mlp_dim = 2048 | |
# ------------------- Model specification overrides -------------------------- | |
VOCABULARY = @seqio.SentencePieceVocabulary() | |
seqio.SentencePieceVocabulary.sentencepiece_model_file = "spiece.model" | |
MODEL = @models.EncoderDecoderModel() | |
models.EncoderDecoderModel: | |
input_vocabulary = %VOCABULARY | |
output_vocabulary = %VOCABULARY |