... tensorflow/stream_executor/cuda/cuda_blas.cc:238] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED ... ... tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_ALLOC_FAILED ... ... Unknown: Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
软件版本:
python: 3.7.7
keras: 2.3.1
tensorboard: 2.1.0
tensorflow: 2.1.0
显卡版本:
gpu: RTX 2070
cuda: 10.1
cudnn: 7.8
解决办法:
参考tensorflow的 pull #34888 : “Failed to get convolution algorithm. This is probably because cuDNN failed to initialize”
The changes introduced in 06d8f77 are not compatible with standalone Keras (they are compatible with tf.Keras). a keras.Model does not have a _get_distribution_strategy method, which is now assumed for the Tensorboard callback.