• tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[1,1424,2128,64]


    tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[1,1424,2128,64]

    1、开始以为是 是显卡显存不足,没有想到是gpu问题

    ref: https://www.cnblogs.com/heiao10duan/p/9083303.html

    Dense(1024, activation='relu'),  #将batch_size从4096改为1024因为我的显存不够
     
    2、开始以为是没有选中独立显卡
    # import os           #key: https://blog.csdn.net/littlehaes/article/details/82317220
    # os.environ["CUDA_VISIBLE_DEVICES"] = "2"    #ref: https://www.cnblogs.com/ying-chease/p/9473938.html
                                                #ref: https://www.cnblogs.com/darkknightzh/archive/2004/01/13/8576825.html
                                                # ref:http://www.voidcn.com/article/p-xhpiftyw-bte.html
                                                # ref:https://blog.csdn.net/weixin_43301333/article/details/108854504
    # os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'    # ref:https://blog.csdn.net/zqx951102/article/details/88897117
                                                # ref:https://www.imooc.com/article/details/id/289425
                                            
    3、 后来通过list_local_devices发现是没有装TensorFlow-gpu, install TensorFlow-gpu
    # from tensorflow.python.client import device_lib
    # print(device_lib.list_local_devices())
    # https://blog.csdn.net/littlehaes/article/details/82317220
    但是注意,安装命令应该使用:conda install -c anaconda tensorflow-gpu
    而不是 pip3 uninstall tensorflow-gpu
     
    测试是否安装成功
    方法一:
    import tensorflow as tf
    import tensorflow.compat.v1 as tfs
    sess = tfs.Session(config=tfs.ConfigProto(log_device_placement=True))
     
    输出中看到:

    2020-12-10 08:38:41.553101: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 3001 MB memory) -> physical GPU (device: 0, name: Quadro P1000, pci bus id: 0000:01:00.0, compute capability: 6.1)
    Device mapping:

    方法二:

    from tensorflow.python.client import device_lib
    print(device_lib.list_local_devices())
     
    输出中看到:

    ncarnation: 5727256362115513404
    , name: "/device:GPU:0"
    device_type: "GPU"
    memory_limit: 3147143577
    locality {
    bus_id: 1
    links {
    }
    }

     
    4、安装成功后,代码运行报错

    from tensorflow.python._pywrap_tensorflow_internal import *
    ImportError: DLL load failed: 找不到指定的模块。

    https://www.jb51.net/article/178940.htm

  • 相关阅读:
    [APIO2018]Circle selection
    [CF1025D]Recovering BST
    [CF1023D]Array Restoration
    [POI2006]PRO-Professor Szu
    [CF888G]Xor-MST
    可持久化并查集小结
    [国家队集训]最短路
    [POI2012]FES-Festival
    [SDOI2010]所驼门王的宝藏
    洛谷 P1025 & [NOIP2001提高组] 数的划分(搜索剪枝)
  • 原文地址:https://www.cnblogs.com/carl2380/p/14111746.html
Copyright © 2020-2023  润新知