oahziur
commented
6 years ago @lipiny
Using inference_list=0,1,5, will force the inference only translate index 0, 1, 5, and skipping anything else in you inference file. You can also create a smaller inference file to achieve the same thing.
Did you view the image in tensorboard?
ONE I found the usage the usage in the previous issue, like --inference_list=0,1,5 But when I set flags like this, in inference.py line51: assert nmt_outputs.shape[0] == 1 shows an error, is it normal?
TWO the size of attention image generated by _decode_inference_indices is so small, I can't figure out anything on it.