OpenNMT Forum

The min ppl is about 4.5; should I continune the training?

opennmt-py
#1

Hi, I trained with this script:

python train.py -data data/my -save_model my-model -gpu_ranks 0 -layers 6 -rnn_size 512 -word_vec_size 512 -transformer_ff 2048 -heads 8 -encoder_type transformer -decoder_type transformer -position_encoding -train_steps 200000 -max_generator_batches 2 -dropout 0.1 -batch_size 4096 -batch_type tokens -normalization tokens -accum_count 2 -optim adam -adam_beta2 0.998 -decay_method noam -warmup_steps 8000 -learning_rate 2 -max_grad_norm 0 -param_init 0 -param_init_glorot -label_smoothing 0.1 -valid_steps 10000 -save_checkpoint_steps 10000 -log_file mylog.train -keep_checkpoint 10

and I found the ppl is now about 4.5 within 200000 steps. the dataset contains about 10000000 couples.
should I continue this training?

(Guillaume Klein) #2

Hi,

You should translate some test files and judge if the quality is good enough.