TENSORFLOW PROCESS FINISHED WITH EXIT CODE -1073741819 (0XC0000005)

一时失言乱红尘 2022-08-31 09:55 81阅读 0赞

这次报错与是否gpu没有关系：

## TENSORFLOW 导入失败：PROCESS FINISHED WITH EXIT CODE -1073741819 (0XC0000005) ##

测试脚本：

import tensorflow as tf
    hello = tf.constant('Hello, TensorFlow!')
    sess = tf.Session()

这个异常，我的解决方法是，tf2.0一下版本，只支持cuda10.0。

重新安装cuda10.0，tf1.15.0版本，后来这个问题就解决了。

预训练下面有个：checkpoint文件，

checkpoint

model\_checkpoint\_path: "snap-43001"  
all\_model\_checkpoint\_paths: "snap-33501"  
all\_model\_checkpoint\_paths: "snap-35501"  
all\_model\_checkpoint\_paths: "snap-37001"  
all\_model\_checkpoint\_paths: "snap-43001"

实际上 最终模型是snap-37001

![20210725092910788.png][]

改一下checkpoint

model\_checkpoint\_path: "snap-37001"  
all\_model\_checkpoint\_paths: "snap-29001"  
all\_model\_checkpoint\_paths: "snap-33501"  
all\_model\_checkpoint\_paths: "snap-35501"  
all\_model\_checkpoint\_paths: "snap-37001"

ok,然后加载预训练ok!

加载预训练方法1：

tf 1.15版本以前加载预训练：

saver.restore(sess, restore_snap)

tf1.15版本：

aaa = tf.train.latest_checkpoint(restore_snap)
                #saver.restore(sess, restore_snap)
                print("start restore------------------", restore_snap,aaa)
                self.saver.restore(self.sess, aaa)
                print("Model restored from " + restore_snap)

加载预训练方法2：

参考：[https://blog.csdn.net/wjc1182511338/article/details/82111790][https_blog.csdn.net_wjc1182511338_article_details_82111790]

def build_graph():
        w1 = tf.Variable([1,3,10,15],name='W1',dtype=tf.float32)
        w2 = tf.Variable([3,4,2,18],name='W2',dtype=tf.float32)
        w3 = tf.placeholder(shape=[4],dtype=tf.float32,name='W3')
        w4 = tf.Variable([100,100,100,100],dtype=tf.float32,name='W4')
        add = tf.add(w1,w2,name='add')
        add1 = tf.add(add,w3,name='add1')
        return w3,add1
    
    with tf.Session() as sess:
        ckpt_state = tf.train.get_checkpoint_state('./temp/')
        if ckpt_state:
            w3,add1=build_graph()
            saver = tf.train.Saver()
            saver.restore(sess, ckpt_state.model_checkpoint_path)
        else:
            w3,add1=build_graph()
            init_op = tf.group(tf.global_variables_initializer(),tf.local_variables_initializer())
            sess.run(init_op)
            saver = tf.train.Saver()
        a = sess.run(add1,feed_dict={
                w3:[1,2,3,4]
            })
        print(a)
        saver.save(sess,'./temp/model')

[20210725092910788.png]: /images/20220829/4d75dbbd3b8741c2b96397fd85ceeb55.png
[https_blog.csdn.net_wjc1182511338_article_details_82111790]: https://blog.csdn.net/wjc1182511338/article/details/82111790