vocab_prefix must be provided

[I3orn2FLY]

I run inference model like that:
python3.7 -m nmt --out_dir=../BestModel/sign2text/ -- inference_input_file=../Data/phoenix2014T.test.sign --inference_output_file=../Data/predictions.de --inference_ref_file=../Data/phoenix2014T.test.de --base_gpu=0

Then i have following error:
nslt/nslt/nmt.py", line 244, in extend_hparams raise ValueError("hparams.vocab_prefix must be provided.") ValueError: hparams.vocab_prefix must be provided.

In arguments i found this line:
parser.add_argument("--vocab_prefix", type=str, default=None, help="""\ Vocab prefix, expect files with src/tgt suffixes.If None, extract from train files.\ """)

Where do i find this vocab_prefix or folder or whatever?

[kyumiouchi]

try this configuration => #13 (comment)
anaconda2, python 2.7, tensorflow-gpu 1.4.0rc1,...

[boyuzz]

I meet the same problem. The comment in #13 is the command for training, not inference.

[EricWangyz]

Under nslt/nslt/ folder, I ran the command below and it worked for me.
python -m nmt --out_dir=<trained_model_dir> --inference_input_file=../Data/phoenix2014T.test.sign --inference_output_file=../Data/predictions_test.de --vocab_prefix=../Data/phoenix2014T.vocab --inference_ref_file=../Data/phoenix2014T.test.de --base_gpu=0