OpenNMT Forum

Training doesn't start with start_from for domain training

opennmt-py
(Alp) #1

Hi,

I am having a strange problem trying to do domain adaptation for an MT model in OpenNMT-py. I have previously trained a model on a large dataset. Now I want to resume training on a different smaller dataset. I first preprocessed the new dataset and obtained pt files for that. Then to resume training I make the following command:

python train.py -data -save_model -coverage_attn -word_vec_size 512 -layers 3 -rnn_size 512 -rnn_type LSTM -encoder_type brnn -batch_size 64 -dropout 0.2 -input_feed 1 -global_attention mlp -optim adam -learning_rate 0.0001 -gpu_ranks 0 -train_steps 10000 -report_every 1 -save_checkpoint_steps 10 -valid_steps 50 -train_from

It loads the files and the old model but then just stops there. This is the err output that I get

I tried taking out the -train_from part and it trains without any problem.

I’d appreciate your help.

0 Likes

(Guillaume Klein) #2
0 Likes

(Guillaume Klein) closed #3
0 Likes