Keras，Tensorflow，CuDDN无法初始化

Question

我有一个非常强大的Windows个人电脑（运行windows10），有112GB内存，16核和3xgeforcertx2070（不支持SLI等）。它运行的是cudn7.5+Tensorflor 1.13+python3.7

我的问题是，当我试图运行Keras模型进行培训或对矩阵进行预测时，我会得到下面的错误。一开始我以为只有当我同时运行多个程序时才会发生这种情况，但事实并非如此，现在当我只运行Keras的一个实例时，我也遇到了错误（通常-但不总是）

2019-06-15 19:33:17.878911: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:2 with 6317 MB memory) -> physical GPU (device: 2, name: GeForce RTX 2070, pci bus id: 0000:44:00.0, compute capability: 7.5) 2019-06-15 19:33:23.423911: I tensorflow/stream_executor/dso_loader.cc:152] successfully opened CUDA library cublas64_100.dll locally 2019-06-15 19:33:23.744678: E tensorflow/stream_executor/cuda/cuda_blas.cc:510] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED 2019-06-15 19:33:23.748069: E tensorflow/stream_executor/cuda/cuda_blas.cc:510] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED 2019-06-15 19:33:23.751235: E tensorflow/stream_executor/cuda/cuda_blas.cc:510] failed to create cublas handle: CUBLAS_STATUS_ALLOC_FAILED 2019-06-15 19:33:25.267137: E tensorflow/stream_executor/cuda/cuda_dnn.cc:334] Could not create cudnn handle: CUDNN_STATUS_ALLOC_FAILED 2019-06-15 19:33:25.270582: E tensorflow/stream_executor/cuda/cuda_dnn.cc:334] Could not create cudnn handle: CUDNN_STATUS_ALLOC_FAILED Exception: 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. [[{{node conv2d_1/convolution}}]] [[{{node dense_3/Sigmoid}}]]

Answer 1

在代码中添加以下内容

from keras.backend.tensorflow_backend import set_session
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = True  # dynamically grow the memory used on the GPU
config.log_device_placement = True  # to log device placement (on which device the operation ran)
sess = tf.Session(config=config)
set_session(sess)  # set this TensorFlow session as the default session for Keras