Created by: L-lei

paddle版本如下: Name: paddlepaddle-gpu Version: 1.6.1.post107

第1个错误:DeepVoice3/deepvoice3_paddle/modules.py中371行 --------------------------------------------------------------------------------------------------self.embed._w.value().get_tensor().set( array, fluid.framework._current_expected_place())

self.embed._w后面是没有value函数的,我这边直接注释调了

第2个错误: DeepVoice3/deepvoice3_paddle/deepvoice3.py中223 x = self.embed(x) 错误信息如下:

PaddleCheckError: Expected ids_dims[ids_rank - 1] == 1, but received ids_dims[ids_rank - 1]:153 != 1:1. ShapeError: The last dimensions of the 'Ids' tensor must be 1. But received Ids's last dimensions = 153, Ids's shape = [16, 153]. at [/paddle/paddle/fluid/operators/lookup_table_op.cc:51]

Embedding 需要的最后一个维度是1,这个我在版本提交记录中看到被修复过了,不知道是不是没全部提交导致的.临时解决方式

#x = self.embed(fluid.layers.reshape(x,[x.shape[0],x.shape[1],1]))

版本记录地址: https://github.com/PaddlePaddle/models/pull/4048/files/264f2ba4a829e3cfef65b4d9ef2f9167756f840f

我本地数据集只有20个音频文件,也没法测试被我改动后是否还有效果