"Not enough space error" in decode but installation without any errors

[marekrydlewski]

After successful installation (I don't get any error messages, everything looks good) I got error when using CTCBeamDecoder, it throws exceptions even when running included tests in test.py.

Error message is always the same: "not enough space" which is quite odd.

> Traceback (most recent call last):
>   File "test.py", line 82, in test_beam_search_decoder_3
>     model_path=lm_path)
>   File "/home/marek/anaconda3/envs/espnet/lib/python3.7/site-packages/ctcdecode/__init__.py", line 18, in __init__
>     self._num_labels)
> RuntimeError: Not enough space

If I pass lm_path to CTCBeamDecoder constructor it fails on line:

 if model_path:
     self._scorer = ctc_decode.paddle_get_scorer(alpha, beta, model_path.encode(), 
                                                                    self._labels,  self._num_labels)

if not, it fails on call to ctc_decode.paddle_beam_decode

I use pytorch in version 1.1 I've tried multiple installations but it didn't help. I use RTX 2080, cuda 10.1 but I don't know if it's related.

Any ideas?

[SeanNaren]

What LM are you using?

[marekrydlewski]

For included test.py in this repo I use included arpa file. But for my usage (deepspeech.pytorch) I use LM binary generated from KenLM)

[baicaitongee]

i have the same error , @marekrydlewski .Have you solved the problem?

[Dikay1]

I meet the same error , @marekrydlewski @baicaitongee .Have you solved the problem?