如何解决 Tensorflow -GPU 的警告？答案

【问题标题】：How do I solve warnings for Tensorflow -GPU?如何解决 Tensorflow -GPU 的警告？
【发布时间】：2020-08-04 22:30:47
【问题描述】：

我正在尝试在 python、ubuntu 18.04 中安装 tensorflow-gpu，使用 pip 命令作为 pip install tensorflow-gpu==2.1.0 运行此命令时：import tensorflow as tf 我收到以下错误：

>>> import tensorflow as tf
2020-08-04 16:06:41.659287: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer.so.6'; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.1/lib64
2020-08-04 16:06:41.659398: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer_plugin.so.6'; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.1/lib64
2020-08-04 16:06:41.659413: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.

当我运行 tf.test.is_gpu_available() 时，我得到了 True，但我收到了很多如下警告：

WARNING:tensorflow:From <stdin>:1: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.config.list_physical_devices('GPU')` instead.
2020-08-04 16:08:33.381085: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-08-04 16:08:33.423734: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2199995000 Hz
2020-08-04 16:08:33.425574: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x41e6da0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2020-08-04 16:08:33.425653: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): Host, Default Version
2020-08-04 16:08:33.432949: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcuda.so.1
2020-08-04 16:08:33.580069: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.580423: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x4282490 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
2020-08-04 16:08:33.580441: I tensorflow/compiler/xla/service/service.cc:176]   StreamExecutor device (0): GeForce GTX 1060, Compute Capability 6.1
2020-08-04 16:08:33.580561: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.580785: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1555] Found device 0 with properties: 
pciBusID: 0000:01:00.0 name: GeForce GTX 1060 computeCapability: 6.1
coreClock: 1.733GHz coreCount: 10 deviceMemorySize: 5.94GiB deviceMemoryBandwidth: 178.99GiB/s
2020-08-04 16:08:33.583471: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-08-04 16:08:33.641250: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcublas.so.10
2020-08-04 16:08:33.675210: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcufft.so.10
2020-08-04 16:08:33.683125: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcurand.so.10
2020-08-04 16:08:33.744338: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusolver.so.10
2020-08-04 16:08:33.753194: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcusparse.so.10
2020-08-04 16:08:33.763988: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudnn.so.7
2020-08-04 16:08:33.764318: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.765495: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.766332: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1697] Adding visible gpu devices: 0
2020-08-04 16:08:33.766456: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library libcudart.so.10.1
2020-08-04 16:08:33.769895: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1096] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-08-04 16:08:33.769951: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1102]      0 
2020-08-04 16:08:33.769973: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] 0:   N 
2020-08-04 16:08:33.770273: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.771146: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:981] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2020-08-04 16:08:33.771798: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1241] Created TensorFlow device (/device:GPU:0 with 4847 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1060, pci bus id: 0000:01:00.0, compute capability: 6.1)
True

我不确定这些警告/错误是否会中断 GPU 的使用。我该如何解决？

【问题讨论】：

你是否安装了GPU安装的tensorflow软件要求？，如果没有按照这个：tensorflow.org/install/gpu
要求是指CUDA吗？如果是的话，我安装了 CUDA 10.1
你用的是什么版本的tensorflow？
首先，没有错误。您称之为“错误”的第一组事情实际上是警告 - 它们前面是W。下一个块以警告开头，然后其余的都是信息性的 - 它们以 I 开头。那么让我们来看看这三件事。 1.前3个警告与TensorRT有关（libnvinfer是TRT的一部分）。如果您不打算使用 TRT，可以忽略这些警告。无论如何，它只会在模型评估期间使用，而不是在训练期间使用。 2. 下一个警告只是对你大喊大叫，让你使用其他东西来测试 GPU 是否存在。
3.最后一组信息输出I 在我看来都很好。它告诉您：它成功打开了一堆库，numa 节点查询存在一些问题，因此假设您只有 1 个 numa 节点，这可能是正确的，并且它正确响应您的 GPU 查询 - 告诉您是的，你有一个 GPU (True)，它是一个 GTX1060。说了这么多，如果你想“解决”TRT 的问题，那就安装 TensorRT。我想你应该可以找到instructions to do that。

标签： tensorflow gpu ubuntu-18.04 nvidia

【解决方案1】：

在您的第一个日志中，它只是说您没有安装 TensorRT。如果您不想使用 tensorflow 的 Feature，请忽略此警告。

那么，你gpu测试，你的log没有问题，可以关注gpu矩阵部分

# tf.test.is_gpu_available() on my machine, which has three gpu
   0  1  2
0: N  N  N
1: N  N  N
2: N  N  N

最后的输出

True

顺便说一句，如果你想关闭这些无聊的日志，你可以设置如下。

''' TF_CPP_MIN_LOG_LEVEL
0 = all messages are logged (default behavior)
1 = INFO messages are not printed
2 = INFO and WARNING messages are not printed
3 = INFO, WARNING, and ERROR messages are not printed
'''
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'

【讨论】：