|
|
@ -11,7 +11,7 @@ data:
|
|
|
|
max_output_input_ratio: .inf
|
|
|
|
max_output_input_ratio: .inf
|
|
|
|
|
|
|
|
|
|
|
|
collator:
|
|
|
|
collator:
|
|
|
|
batch_size: 32 # one gpu
|
|
|
|
batch_size: 64 # one gpu
|
|
|
|
mean_std_filepath: data/mean_std.json
|
|
|
|
mean_std_filepath: data/mean_std.json
|
|
|
|
unit_type: char
|
|
|
|
unit_type: char
|
|
|
|
vocab_filepath: data/vocab.txt
|
|
|
|
vocab_filepath: data/vocab.txt
|
|
|
@ -41,12 +41,12 @@ model:
|
|
|
|
rnn_direction: forward # [forward, bidirect]
|
|
|
|
rnn_direction: forward # [forward, bidirect]
|
|
|
|
num_fc_layers: 1
|
|
|
|
num_fc_layers: 1
|
|
|
|
fc_layers_size_list: 512,
|
|
|
|
fc_layers_size_list: 512,
|
|
|
|
use_gru: True
|
|
|
|
use_gru: False
|
|
|
|
|
|
|
|
|
|
|
|
training:
|
|
|
|
training:
|
|
|
|
n_epoch: 50
|
|
|
|
n_epoch: 50
|
|
|
|
lr: 2e-3
|
|
|
|
lr: 2e-3
|
|
|
|
lr_decay: 0.83 # 0.83
|
|
|
|
lr_decay: 0.91 # 0.83
|
|
|
|
weight_decay: 1e-06
|
|
|
|
weight_decay: 1e-06
|
|
|
|
global_grad_clip: 3.0
|
|
|
|
global_grad_clip: 3.0
|
|
|
|
log_interval: 100
|
|
|
|
log_interval: 100
|
|
|
|