Internal assert error in awd_qrnn

[sky1ove]

To reproduce the error:https://colab.research.google.com/drive/19WzwYTzs78DNvQ7VTw_37hbXq3l5ozsk?usp=sharing

My input is 872 embeddings (from integer 0 to 872), output is n=206 classifications. AWD_LSTM works fine. However, when I changed to AWD_QRNN, it shows the bug below.

config = awd_qrnn_clas_config.copy()
config.update({'bidir':True})

model =get_text_classifier(AWD_QRNN, 872, 206, seq_len=872, config=config)
print(model)
SequentialRNN(
(0): SentenceEncoder(
(module): AWD_QRNN(
(encoder): Embedding(872, 400, padding_idx=1)
(encoder_dp): EmbeddingDropout(
(emb): Embedding(872, 400, padding_idx=1)
)
(rnns): ModuleList(
(0): QRNN(
(layers): ModuleList(
(0): QRNNLayer(
(linear): WeightDropout(
(module): Linear(in_features=800, out_features=2328, bias=True)
)
)
)
(layers_bwd): ModuleList(
(0): QRNNLayer(
(linear): Linear(in_features=800, out_features=2328, bias=True)
)
)
)
(1): QRNN(
(layers): ModuleList(
(0): QRNNLayer(
(linear): WeightDropout(
(module): Linear(in_features=1552, out_features=2328, bias=True)
)
)
)
(layers_bwd): ModuleList(
(0): QRNNLayer(
(linear): Linear(in_features=1552, out_features=2328, bias=True)
)
)
)
(2): QRNN(
(layers): ModuleList(
(0): QRNNLayer(
(linear): WeightDropout(
(module): Linear(in_features=1552, out_features=2328, bias=True)
)
)
)
(layers_bwd): ModuleList(
(0): QRNNLayer(
(linear): Linear(in_features=1552, out_features=2328, bias=True)
)
)
)
(3): QRNN(
(layers): ModuleList(
(0): QRNNLayer(
(linear): WeightDropout(
(module): Linear(in_features=1552, out_features=600, bias=True)
)
)
)
(layers_bwd): ModuleList(
(0): QRNNLayer(
(linear): Linear(in_features=1552, out_features=600, bias=True)
)
)
)
)
(input_dp): RNNDropout()
(hidden_dps): ModuleList(
(0): RNNDropout()
(1): RNNDropout()
(2): RNNDropout()
(3): RNNDropout()
)
)
)
(1): PoolingLinearClassifier(
(layers): Sequential(
(0): LinBnDrop(
(0): BatchNorm1d(1200, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): Dropout(p=0.4, inplace=False)
(2): Linear(in_features=1200, out_features=50, bias=False)
(3): ReLU(inplace=True)
)
(1): LinBnDrop(
(0): BatchNorm1d(50, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(1): Dropout(p=0.1, inplace=False)
(2): Linear(in_features=50, out_features=206, bias=False)
)
)
)
)

x, y = dls.one_batch()
model(x)

Everything works well with AWD_LSTM, but not with AWD_QRNN. Please fix it. Thanks.

[jph00]

Please provide a complete reproducer. A colab link would be most helpful. The reproducer needs to use a publicly available dataset - preferably a small one that's provided in fastai's URLs class.

[sky1ove]

Please provide a complete reproducer. A colab link would be most helpful. The reproducer needs to use a publicly available dataset - preferably a small one that's provided in fastai's URLs class.

https://colab.research.google.com/drive/19WzwYTzs78DNvQ7VTw_37hbXq3l5ozsk?usp=sharing
Please check this colab