Cannot resolve a new error after updating native client

DeepSpeech
#1

For quick and portable tests, I use CPU native client, which works ok.
For the real many-days training, I switch to GPU, this used to work until today.

Now I get this error: : TypeError: ctc_beam_search_decoder_with_lm

$ bin/run.sh

  • [ ! -f DeepSpeech.py ]
  • [ ! -d ]
  • COMPUTE_DATA_DIR=data/arabic
  • [ ! -f data/arabic/arabic70p_train.csv ]
  • [ -d ]
  • python -c from xdg import BaseDirectory as xdg; print(xdg.save_data_path(“deeps
    peech/arabic”))
  • checkpoint_dir=/home/teldeeb/.local/share/deepspeech/arabic
  • python -u DeepSpeech.py --train_files data/arabic/arabic70p_train.csv --dev_files
    data/arabic/arabic70p_dev.csv --test_files data/arabic/arabic70p_test.csv --alphabet
    _config_path data/arabic/arabic-alphabets.txt --lm_binary_path data/arabic/lm.binary
    –lm_trie_path data/arabic/arabic.trie --export_dir data/arabic/exported.model --tr
    ain_batch_size 4 --dev_batch_size 4 --test_batch_size 4 --noearly_stop --epoch 10
    0 --display_step 4 --validation_step 1 --dropout_rate 0.30 --n_hidden 1024 --defa
    ult_stddev 0.046875 --learning_rate 0.0001 --checkpoint_dir data/arabic/…/checkpo
    ints --checkpoint_secs 1800 --summary_secs 180
    Traceback (most recent call last):
    File “DeepSpeech.py”, line 1870, in
    tf.app.run()
    File “/home/teldeeb/.conda/envs/fastai/lib/python3.6/site-packages/tensorflow/p
    ython/platform/app.py”, line 126, in run
    sys.exit(main(argv))
    File “DeepSpeech.py”, line 1827, in main
    train()
    File “DeepSpeech.py”, line 1500, in train
    results_tuple, gradients, mean_edit_distance, loss = get_tower_results(model
    feeder, optimizer)
    File “DeepSpeech.py”, line 635, in get_tower_results
    calculate_mean_edit_distance_and_loss(model_feeder, i, dropout_rates)
    File “DeepSpeech.py”, line 528, in calculate_mean_edit_distance_and_loss
    decoded, _ = decode_with_lm(logits, batch_seq_len, merge_repeated=False, beam
    _width=FLAGS.beam_width)
    File “DeepSpeech.py”, line 487, in decode_with_lm
    top_paths=top_paths, merge_repeated=merge_repeated))
    TypeError: ctc_beam_search_decoder_with_lm() got an unexpected keyword argument ’
    word_count_weight’

I cannot figure out what had I changed that caused this error.
Please support.

#2

re generate lm binary for your architecture …

use taskcluster --arch flag

i assume u mixed the binary lm for a gpu arch on your cpu laptop

hope it fixes it
Saeed

#3

Looks like you are using a mix of the old 0.1.1 code and the current master. For example, word_count_weight is no longer supported in master.

What I would do is to make sure I’m using code at commit c579b74 or before as code after that commit is not backwards compatible.

#4

@kdavis, that’s the exact problem. So I either:
1- revert to an older git, as you have mentioned

2- get a new native client. Is this an option? How?

Regards,

#5

I’d say both are options.

  1. This is easier if you are using our pre-trained models, which as far as I remember you are not using as you are creating your own models from scratch.

  2. This is also an option, but the code is less mature as it’s for an upcoming release which hasn’t had all the bugs worked out yet.

#6

Mr Tarek
alsalamo alaikum

Please can you contact me
nezar_dasan@yahoo.de