sxjscience opened a new issue #10453: Bug of CuDNN RNN with variable sequence 
length
URL: https://github.com/apache/incubator-mxnet/issues/10453
 
 
   ## Description
   Segfault will be triggered by the following code:
   
   ```python
   from mxnet.gluon.rnn import LSTM
   import mxnet as mx
   import numpy as np
   
   ctx = mx.gpu()
   lstm = LSTM(num_layers=1, hidden_size=200, dropout=0.5)
   lstm.initialize(ctx=ctx)
   batch_size = 32
   for seq_len in range(500, 10, -1):
       for repeat in range(10):
           print(seq_len, repeat)
           inputs_nd = mx.nd.random.normal(0, 1, shape=(seq_len, batch_size, 
200), ctx=ctx)
           out = lstm(inputs_nd)
           print(out[0].sum().asscalar())
           mx.nd.waitall()
   ```
   
   I'm using V100 + cuda 9.0 + cudnn 7.0.4 (P3 instance).  The GPU memory keeps 
increasing and finally raises seg fault.
   
   Also, the same script + configuration has not triggered an error in M60 (g3 
instance).
   
   @eric-haibin-lin @DickJC123 @szha @szhengac 

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services

Reply via email to