即使我安装了directml（我在amd上），为什么我的VSCode仍试图使用cuda？

Question

我有一个张量流对象检测项目，我想构建并阅读它在 cpu 上会很慢。那时有人告诉我使用 directml，因为我有一个 AMD gpu 而不是 NVIDIA 的。

我创建了一个名为“directml”的 anaconda 环境，并在其上安装了 tensorflow 和 directml（参见 picture）。如果我现在尝试运行我从本教程 (https://docs.microsoft.com/en-us/windows/ai/directml/gpu-tensorflow-windows) 中找到的测试应用程序：

import tensorflow.compat.v1 as tf 

tf.enable_eager_execution(tf.ConfigProto(log_device_placement=True)) 

print(tf.add([1.0, 2.0], [3.0, 4.0]))

我没有得到想要的输出：

2020-06-15 11:27:18.240065: I tensorflow/core/common_runtime/dml/dml_device_factory.cc:32] DirectML: creating device on adapter 0 (AMD Radeon VII) 

2020-06-15 11:27:18.323949: I tensorflow/stream_executor/platform/default/dso_loader.cc:60] Successfully opened dynamic library DirectMLba106a7c621ea741d2159d8708ee581c11918380.dll 

2020-06-15 11:27:18.337830: I tensorflow/core/common_runtime/eager/execute.cc:571] Executing op Add in device /job:localhost/replica:0/task:0/device:DML:0 

tf.Tensor([4. 6.], shape=(2,), dtype=float32)

但我却得到了这个：

2021-09-16 17:15:03.700209: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'cudart64_100.dll'; dlerror: cudart64_100.dll not found
2021-09-16 17:15:03.700418: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2021-09-16 17:15:05.192685: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'nvcuda.dll'; dlerror: nvcuda.dll not found
2021-09-16 17:15:05.192902: E tensorflow/stream_executor/cuda/cuda_driver.cc:318] failed call to cuInit: UNKNOWN ERROR (303)
2021-09-16 17:15:05.197503: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: DESKTOP-N3L36AL
2021-09-16 17:15:05.197857: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: DESKTOP-N3L36AL
2021-09-16 17:15:05.198376: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2021-09-16 17:15:05.202832: I tensorflow/core/common_runtime/eager/execute.cc:571] Executing op Add in device /job:localhost/replica:0/task:0/device:CPU:0
tf.Tensor([4. 6.], shape=(2,), dtype=float32)

在我看来，tensorflow 似乎在尝试使用 cuda 而不是 directml，但我不知道为什么会这样。我的 Windows 以及我的 AMD 驱动程序都是最新的。

Answer 1

你不应该只安装 tensorflow tensorflow-directml。因为现在 python 正在导入 tensorflow 而不是 tensorflow-directml。卸载 tensorflow，它应该会修复导入。