gigaword-10k_conf

#load model
vocab_path="ernie_gen_large_160g/vocab.txt"
config_path="ernie_gen_large_160g/ernie_config.json"
init_model="ernie_gen_large_160g/params"

#input
max_src_len=192
max_tgt_len=64
tokenized_input="true"
continuous_position="true"
batch_size=16
in_tokens="false"
tgt_type_id=3

#decode
do_decode="true"
max_dec_len=32
beam_size=5
length_penalty=0.6
use_multi_gpu_test="true"

#train
epoch=30
weight_decay=0.01
label_smooth=0.1
save_and_valid_by_epoch="true"
hidden_dropout_prob=0.1
#lr
warmup_proportion=0.15
lr_scheduler="linear_warmup_decay"
learning_rate=7.5e-6
#noise
random_noise="true"
noise_prob=0.65

#dataset
data_path="./datasets/gigaword/"
train_set="train.10k.tsv"
dev_set="dev.20k.tsv"
test_set="test.tsv"
do_train="true"
do_val="true"
do_test="true"
do_pred="false"

#evaluate
eval_script="sh ./eval/tasks/gigaword/eval.sh"
eval_mertrics="rouge-1,rouge-2,rouge-l"