DL4MT
DL4MT copied to clipboard
Dual Learning for Machine Translation
Hello! Here is a question. When I used the command "./run_train_dual.py sample_config.py", next I want to test or translate. What should I do? I can't know where is the module...
I get this warning interspersed with the normal logs (eg. `training the a -> b -> a loop.`, `training the b -> a -> b loop.` and `training on bitext`)...
Update unit tests to depend only on Rico's wmt16 models and small data files (put them in test_data dir).
Need a clean way to run the script - config file or command line args
When you get validation working, use this to save the best model. Currently only saves after each epoch, but you don't know which one is the best to save as...
Make sure that if the run dies make sure that it exits cleanly. Delete tmp files. Anything else?
So it looks like a bunch of validation jobs happen altogether towards the end. And it doesn't look like anything changes. Any idea why this happens? INFO:root:epoch=4, MT a->b valid_err=53.3...