在训练的时候报CUDNN_STATUS_EXECUTION_FAILED错误
Created by: yeyupiaoling
环境
- PaddlePaddle 1.2.0 GPU版本
- Ubuntu 16.04
- CUDA 9.0 CUDNN 7.4
- 模型MobileNet V1
问题
在使用MobileNet V1训练图像分类的是报错,错误如下:
W0102 15:54:15.308672 4412 device_context.cc:203] Please NOTE: device: 0, CUDA Capability: 75, Driver Version: 10.0, Runtime Version: 9.0
W0102 15:54:15.308725 4412 device_context.cc:210] device: 0, cuDNN Version: 7.0.
Traceback (most recent call last):
File "train.py", line 66, in <module>
fetch_list=[avg_cost, accuracy, param_name])
File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/executor.py", line 472, in run
self.executor.run(program.desc, scope, 0, True, True)
paddle.fluid.core.EnforceNotMet: CUDNN_STATUS_EXECUTION_FAILED at [/paddle/paddle/fluid/operators/conv_cudnn_op.cu.cc:230]
PaddlePaddle Call Stacks:
在跑官方的房价预测例子种正常运行。