tensorflow-gpu报错：tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found

laxyer

已于 2023-03-27 12:59:29 修改

阅读量1.7k

点赞数 1

CC 4.0 BY-SA版权

文章标签： tensorflow python 深度学习

于 2023-03-27 12:57:27 首次发布

本文链接：https://blog.youkuaiyun.com/laxyer/article/details/129792942

本文详细记录了解决tensorflow-gpu运行时遇到的UnknownError问题的过程。尝试了检查tensorflow、CUDA和cuDNN版本一致性，分配显存及添加特定代码等方法，最终成功继续训练。尽管有人认为某代码可能使训练转为CPU，但对作者有效。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

一、报错内容

tensorflow.python.framework.errors_impl.UnknownError: 2 root error(s) found.
  (0) Unknown: 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_24/convolution}}]]
  (1) Unknown: 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_24/convolution}}]]
	 [[conv2d_46/truediv/_417]]
0 successful operations.
0 derived errors ignored.

二、解决方案

import keras
import tensorflow as tf
import os

os.environ["CUDA_VISIBLE_DEVICES"] = "1"#因为是单GPU训练，所以需要加上这一行代码
config = tf.ConfigProto()#t配置tf.Session的运算方式
config.gpu_options.allow_growth = True  # TensorFlow按需分配显存
config.gpu_options.per_process_gpu_memory_fraction