plaidml-keras：OSError：异常：堆栈溢出

Question

我正在使用plaidml-keras通过Radeon RX 480 GPU训练循环神经网络。但是我一直遇到

  

OSError：异常：堆栈溢出

我正在任务管理器中监视我的RAM和GPU，并且内存没有地方满。我正在使用Jupyter Notebook。但是，我能够训练较小的模型。

笔记本电脑会在错误发生之前立即输出一长串警告。

谢谢。

警告看起来像这样

n1 [label="I0\nembedding_1_input\ Placeholder\ FLOAT32\(<tile\.Value\ SymbolicDim\ UINT64\(\)>,\ 300\)" shape=circle];
n2 [label="I1\ndense_1_target\ Placeholder\ FLOAT32\(<tile\.Value\ SymbolicDim\ UINT64\(\)>,\ <tile\.Value\ SymbolicDim\ UINT64\(\)>\)" shape=circle];
n3 [label="I2\ndense_1_sample_weights\ Placeholder\ FLOAT32\(<tile\.Value\ SymbolicDim\ UINT64\(\)>,\)" shape=circle];
n4 [label="O0\nMean\ FLOAT32\(\)" shape=doublecircle];
n5 [label="O1\nMean\ FLOAT32\(\)" shape=doublecircle];
n6 [label="RMSprop/iterations\ Tensor\ INT64\(\)" shape=circle];
n7 [label="Add\ INT64\(\)" shape=circle];
n7 -> n6 [style=dotted];
...
n77 [label="Gradients/O0\ FLOAT32\(100,\ 256\)" shape=box];
n76 -> n77 [label="O0"];
n78 [label="Gradients/O1\ FLOAT32\(64,\ 256\)" shape=box];
n76 -> n78 [label="O1"];
n79 [label="Gradients/O2\ FLOAT32\(256,\)" shape=box];
n76 -> n79 [label="O2"];
n80 [label="Gradients/O3\ FLOAT32\(100,\ 256\)" shape=box];
n76 -> n80 [label="O3"];
n81 [label="Gradients/O4\ FLOAT32\(64,\ 256\)" shape=box];
n76 -> n81 [label="O4"];
n82 [label="Gradients/O5\ FLOAT32\(256,\)" shape=box];
...