pase
pase copied to clipboard
santi-pdp
TypeError: ('Unrecognized rnn type: ', 'qrnn')
I am following the example set in the README, and I am getting the following error. Is this familiar to anyone?
In [1]: from pase.models.frontend import wf_builder
In [2]: pase = wf_builder('pase_models/cfg/frontend/PASE+.cfg').eval()
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-2-8accba11a31c> in <module>
----> 1 pase = wf_builder('pase_models/cfg/frontend/PASE+.cfg').eval()
/opt/conda/lib/python3.6/site-packages/PASE-0.1.1.dev0-py3.6.egg/pase/models/frontend.py in wf_builder(cfg_path)
21 with open(cfg_path, 'r') as cfg_f:
22 cfg = json.load(cfg_f)
---> 23 return wf_builder(cfg)
24 elif isinstance(cfg_path, dict):
25 if "name" in cfg_path.keys():
/opt/conda/lib/python3.6/site-packages/PASE-0.1.1.dev0-py3.6.egg/pase/models/frontend.py in wf_builder(cfg_path)
34 raise TypeError('Unrecognized frontend type: ', model_name)
35 else:
---> 36 return WaveFe(**cfg_path)
37 else:
38 TypeError('Unexpected config for WaveFe')
/opt/conda/lib/python3.6/site-packages/PASE-0.1.1.dev0-py3.6.egg/pase/models/frontend.py in __init__(self, num_inputs, sincnet, kwidths, strides, dilations, fmaps, norm_type, pad_mode, sr, emb_dim, rnn_dim, activation, rnn_pool, rnn_layers, rnn_dropout, rnn_type, vq_K, vq_beta, vq_gamma, norm_out, tanh_out, resblocks, denseskips, densemerge, name)
192 rnn_type=rnn_type,
193 bidirectional=True,
--> 194 dropout=rnn_dropout)
195 self.W = nn.Conv1d(rnn_dim, emb_dim, 1)
196 else:
/opt/conda/lib/python3.6/site-packages/PASE-0.1.1.dev0-py3.6.egg/pase/models/modules.py in build_rnn_block(in_size, rnn_size, rnn_layers, rnn_type, bidirectional, dropout, use_cuda)
57 bidirectional=bidirectional)
58 else:
---> 59 raise TypeError('Unrecognized rnn type: ', rnn_type)
60 return rnn
61
TypeError: ('Unrecognized rnn type: ', 'qrnn')
I experienced the same error. I think it is something related with the installation of git+https://github.com/salesforce/pytorch-qrnn and cupy-cuda101
as @jcvasquezc said, you need to install cupy (your cuda version) + https://github.com/salesforce/pytorch-qrnn
^this resolves the error
Hi @delip ,
indeed as others correctly point out, you need to install a proper cupy version (in the requirements.txt you'll see the cuda 10.1 version cupy-cuda101, adjust as needed like cupy-cuda100 for your setup ) in addition to the official Pytorch QRNN repo.
