【发布时间】:2016-08-13 10:37:50
【问题描述】:
我在 Theano 中做卷积:
theano.tensor.nnet.conv.conv2d(x,h, border_mode='full')
内存不足,我收到以下消息:
RuntimeError: GpuCorrMM failed to allocate working memory of 3591 x 319086
Apply node that caused the error: GpuCorrMM_gradInputs{valid, (1, 1)}(GpuContiguous.0, GpuContiguous.0)
Inputs types: [CudaNdarrayType(float32, (True, False, True, False)), CudaNdarrayType(float32, (False, True, False, False))]
Inputs shapes: [(1, 513, 1, 7), (1, 1, 513, 622)]
Inputs strides: [(0, 7, 0, 1), (0, 0, 622, 1)]
Inputs values: ['not shown', 'not shown']
我已尝试将 theano 标志设置为“optimizer_ exclude=conv_dnn”,但仍然没有成功。有没有办法解决这个问题?
【问题讨论】:
标签: machine-learning neural-network theano conv-neural-network convolution