def Task(cls):
p = base_config.SetupTransformerParams(
name='wmt14_en_de_transformer_base',
vocab_size=cls.VOCAB_SIZE,
model_dim=512,
hidden_dim=2048,
num_heads=8,
num_layers=6,
residual_dropout_prob=0.1,
input_dropout_prob=0.1,
learning_rate=3.0,
warmup_steps=40000)
p.eval.samples_per_summary = 7500
return p
def Task(cls):
p = base_config.SetupTransformerParams(
name='wmt14_en_de_transformer_base',
vocab_size=cls.VOCAB_SIZE,
model_dim=256,
hidden_dim=512,
num_heads=2,
num_layers=2,
residual_dropout_prob=0.2,
input_dropout_prob=0.2,
learning_rate=1.0,
warmup_steps=1000)
p.eval.samples_per_summary = 7500
p.train.save_interval_seconds = 60
p.train.max_steps = 12000
return p
