search

google / seq2seq

A general-purpose encoder-decoder framework for Tensorflow
https://google.github.io/seq2seq/
Apache License 2.0
5.61k stars 1.3k forks source link

Issues in nmt_medium and nmt_large config files #282

Open kadir-gunel opened 7 years ago

kadir-gunel commented 7 years ago

I try to execute prepared config files for nmt but seems that can not run all but nmt_small only. When I try to run them I get NotFoundError : Key model/att_seq2seq/OptimizeLoss/model/att_seq2seq/decode/attention_decoder/decoder/extended_multi_rnn_cell/cell_3/lstm_cell/weights/Adam_1 not found in checkpoint

Tried to locate where model/att_seq2seq/.../ is but could not manage to find it. Probably it refers to the tensorflow lib., right?

Since I can run nmt_small, I assume that everything is properly installed on my pc(btw it passes the tests). Any help ?

geert-heyman commented 7 years ago

My guess is that you are running the nmt_medium/nmt_large models with the same _outputdir you used for training the nmt_small model. Therefore, seq2seq is trying to initialize your nmt_medium/nmt_large model with a checkpoint created from the nmt_small model, which won't have parameters for _cell3.

Try something like python -m bin.train --ouput_dir nmt_medium_model ... python -m bin.train --ouput_dir nmt_large_model ...

dadashkarimi commented 7 years ago

Maybe you made mistake in writing this part: ` --input_pipeline_train " class: ParallelTextInputPipeline params: source_files: