{"id":47449,"date":"2023-08-02T09:41:55","date_gmt":"2024-02-14T22:04:39","guid":{"rendered":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/"},"modified":"2024-04-30T16:36:39","modified_gmt":"2024-04-30T08:36:39","slug":"rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47","status":"publish","type":"post","link":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/","title":{"rendered":"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5)"},"content":{"rendered":"<p>2019\u5e747\u670816\u65e5\u66f4\u65b0\uff1a\u6700\u8fd1\uff0cpip install tensorflow-rocm\u5df2\u7ecf\u8db3\u591f\u53ef\u7528\uff0c\u56e0\u6b64\u5982\u679c\u53ea\u662f\u60f3\u5728ROCM\u4e2d\u4f7f\u7528TensorFlow\uff0c\u8bf7\u4f7f\u7528pip\u3002<\/p>\n<hr \/>\n<p>\u5982\u679c\u662f\u4e2d\u56fd\u4eba\uff0c\u5e94\u8be5\u4f1a\u5e0c\u671b\u5728\u4e2d\u534e\u6c11\u56fd\u7684\u73af\u5883\u4e0b\u4f7f\u7528TensorFlow\u5427\uff01<\/p>\n<p>\u8ba9\u6211\u4eec\u6311\u6218\u4e00\u4e0b\u5427\uff01<\/p>\n<p>\u8ffd\u52a0\u6ce8\uff1a\u81f32018\u5e749\u670823\u65e5\u4e3a\u6b62\uff0c\u4eceTensorFlow1.9\u7248\u672c\u5f00\u59cb\uff0c\u5df2\u7ecf\u63d0\u4f9b\u4e86\u9884\u7f16\u8bd1\u7248\u672c\uff0c\u53ef\u4ee5\u901a\u8fc7pip\u7b49\u65b9\u5f0f\u8fdb\u884c\u5b89\u88c5\u3002\u5982\u679c\u60a8\u60f3\u5148\u5c1d\u8bd5\u5728ROCm\u4e0a\u8fd0\u884cTensorFlow\uff0c\u6211\u4eec\u63a8\u8350\u4f7f\u7528\u9884\u7f16\u8bd1\u7248\u672c\u3002<\/p>\n<p>\u622a\u81f32018\u5e749\u670823\u65e5\uff0cROCm 1.9 + TensorFlow 1.10\u662f\u6700\u65b0\u7248\u672c\u3002<\/p>\n<p>\u622a\u81f32018\u5e747\u670821\u65e5\uff0cAMD ROCm\u5b98\u65b9\u53d1\u5e03\u7684\u662fTensorFlow 1.3\u7248\u672c\uff0c\u8fd9\u5df2\u7ecf\u6709\u70b9\u8fc7\u65f6\u4e86\u3002<\/p>\n<p>\u6709\u4e00\u4e2a\u540d\u4e3atensorflow-upstream\u7684\u4ee3\u7801\u5e93\uff0c\u9700\u8981\u6784\u5efa\u5b83\u3002<\/p>\n<p>\u53ef\u4ee5\u53c2\u8003tensorflow-upstream\/rocm_docs\u4e2d\u7684ROCm\u6784\u5efa\u6587\u6863\u7b49\u3002<\/p>\n<ul class=\"post-ul\">ROCm 1.8.2<\/ul>\n<p>ROCm \u6700\u521d\u53ea\u652f\u6301 PCI Gen3 \u7684\u67b6\u6784\uff0c\u4f46\u8003\u8651\u5230\u865a\u62df\u8d27\u5e01\u6316\u77ff\u7684\u9700\u6c42\uff0c\u73b0\u5728\u4e5f\u80fd\u5728 PCI Gen2 \u4e0a\u8fd0\u884c\u3002<br \/>\n\uff08\u4ee4\u4eba\u60ca\u8bb6\u7684\u662f\uff0c\u751a\u81f3\u65e7\u6b3e\u7684 FirePro W9100\uff08Hawaii\uff09GPU \u4e5f\u867d\u4e0d\u88ab\u63a8\u8350\u4f7f\u7528\uff0c\u4f46\u5374\u6210\u4e3a\u4e86\u517c\u5bb9\u7684GPU\uff09<\/p>\n<p>\u53ea\u8981\u5728\u6b64\u5728SandyBridge\u8ba1\u7b97\u673a\u4e0a\u5b89\u88c5\u65f6\uff0ctensorflow\u6784\u5efa\u65f6nasm\u51fa\u73b0\u975e\u6cd5\u6307\u4ee4\u9519\u8bef\u7b49\u95ee\u9898\u3002 \u5982\u679c\u7a0d\u5fae\u4fee\u6539tensorflow-upstream\u7684\u914d\u7f6e\uff0c\u6682\u65f6\u53ef\u4ee5\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff08\u8bf7\u53c2\u8003\u4e0b\u9762\u7684\u8bf4\u660e\uff09\u3002<\/p>\n<h2>\u6784\u6210<\/h2>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">Ubuntu 16.04<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">Python 3.5 (conda)<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">ROCm 1.8.2<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u305d\u3053\u305d\u3053\u6700\u65b0\u306e\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3(ROCm \u306e\u30da\u30fc\u30b8\u306b\u3042\u308b\u30b5\u30dd\u30fc\u30c8\u5bfe\u8c61\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3)\u306e CPU<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">RX 580(gfx803), VEGA(gtx900)<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">Bazel<\/ul>\n<h2>\u8bf7\u63d0\u4f9b\u76f8\u5173\u7684\u4fe1\u606f<\/h2>\n<p>\u8c22\u8c22\u4f60\u7684\u53c2\u8003\uff0c\u6211\u4f1a\u597d\u597d\u5229\u7528\u3002<\/p>\n<p>\u5982\u679c\u8981\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d\u8fdb\u884c\u8bad\u7ec3\uff0c\u53ea\u9009\u62e9NVIDIA\u5417\uff1f\u4e0d\uff0c\u8fd8\u6709AMD\u8fd9\u4e2a\u9009\u9879\u5462\u3002<\/p>\n<p>Note: The translation above is provided in simplified Chinese.<\/p>\n<p>\u4f7f\u7528AMD\u5236\u9020\u7684GPU\uff0c\u7ed3\u5408ROCm\u3001OpenCV\u548cTensorFlow\u6765\u521b\u5efaGPU\u8ba1\u7b97\u73af\u5883\u3002<\/p>\n<h2>\u5b89\u88c5<\/h2>\n<p>\u9996\u5148\uff0c\u53c2\u8003 ROCm \u7f51\u7ad9\uff0c\u5148\u5b89\u88c5 ROCm\u3002<\/p>\n<p>\u8bf7\u524d\u5f80\u4ee5\u4e0b\u7f51\u5740\u4e0b\u8f7dRadeonOpenCompute\/ROCm\u8f6f\u4ef6\u5305\uff1a<br \/>\nhttps:\/\/github.com\/RadeonOpenCompute\/ROCm \uff08\u5b89\u88c5\u6307\u4ee4\uff1aapt-get install rocm-dev\uff09<\/p>\n<p>\u6211\u4f1a\u52a0\u5165\u7c7b\u4f3c\u4e8eblas\u7684\u4e1c\u897f\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"nv\">$ <\/span><span class=\"nb\">sudo <\/span>apt <span class=\"nb\">install <\/span>rocrand rocfft rocblas\r\n<\/code><\/pre>\n<p>\u6211\u5c06\u5b89\u88c5MIOpen\uff08\u7c7b\u4f3c\u4e8ecuDNN\u7684\u8ba1\u7b97\u52a0\u901f\u5e93\uff09\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"nv\">$ <\/span><span class=\"nb\">sudo <\/span>apt <span class=\"nb\">install <\/span>miopen-hip miopengemm\r\n<\/code><\/pre>\n<p>\u7528Python 3.5\u51c6\u5907\u73af\u5883\uff08\u4e5f\u53ef\u80fd\u9002\u7528\u4e8e3.6\u7248\u672c\uff09<br \/>\n\u8fd9\u6b21\u4f7f\u7528\u4e86(mini)conda\u3002<\/p>\n<pre class=\"post-pre\"><code>$ conda create -n py35 python=3.5\r\n$ source activate py35\r\n<\/code><\/pre>\n<p>\u5982\u679c\u5728python\u6a21\u5757\u7684\u6784\u5efa\u8fc7\u7a0b\u4e2d\u5931\u8d25\uff0c\u8bf7\u9002\u5f53\u5b89\u88c5numpy\u548csix\u5e93\u3002<\/p>\n<pre class=\"post-pre\"><code>(py35) $ (pip install numpy six\r\n<\/code><\/pre>\n<p>\u5b89\u88c5Bazel beforehand.<\/p>\n<p>\u6211\u5c06\u786e\u8ba4tensorflow-upstream\u7684\u5206\u652f\u662fdevelop-upstream\u3002<\/p>\n<p>\u914d\u7f6e\u3002<\/p>\n<h3>\u53ef\u9009\u7684<\/h3>\n<p>\u5728tensorflow-upstream\u4e2d\uff0c\u9ed8\u8ba4\u7684\u4f18\u5316\u9009\u9879\u4e3a-march=haswell\u3002\u5c3d\u7ba1\u53ef\u4ee5\u5728configure\u4e2d\u66f4\u6539\u5b83\uff0c\u4f46\u7531\u4e8e\u5176\u4ed6\u5730\u65b9\u4f7f\u7528\u4e86-march=haswell\uff0c\u56e0\u6b64\u5728\u67d0\u4e9bCPU\u4e0a\u65e0\u6cd5\u6b63\u5e38\u5de5\u4f5c\uff08\u5373\u4f7f\u662f\u8f83\u5c0f\u578b\u7684Celeron\uff0c\u4f8b\u5982G3930\u4e4b\u7c7b\u7684haswell\u53ca\u5176\u540e\u7eed\u67b6\u6784\u7684CPU\u4e5f\u5b58\u5728\u90e8\u5206\u529f\u80fd\u7f3a\u5931\uff0c\u56e0\u6b64\u4f1a\u51fa\u73b0NG\uff09\u3002\u5177\u4f53\u6765\u8bf4\uff0c\u7528\u4e8e\u6784\u5efatensorflow\u6240\u9700\u7684\u4f9d\u8d56\u7a0b\u5e8fprotoc\u548cnasm\u5728\u8fd0\u884c\u65f6\u4f1a\u56e0\u975e\u6cd5\u6307\u4ee4\u800c\u5d29\u6e83\u3002<\/p>\n<p>\uff082018\u5e749\u670823\u65e5\u66f4\u65b0\uff09\u9488\u5bf9\u8fd9\u4e2a\u95ee\u9898\u7684\u4fee\u8865\u7a0b\u5e8f\u5df2\u7ecf\u88ab\u5408\u5e76\u5230 https:\/\/github.com\/ROCmSoftwarePlatform\/tensorflow-upstream\/pull\/81\uff0c\u5e76\u5f97\u5230\u4fee\u590d\u3002<\/p>\n<p>\u5728configure.py\u4e2d\u5c06-march=native\u548c&#8211;host_copt=-march=native\u914d\u7f6e\u4e3a\u4ee5\u4e0b\u9009\u9879\u53ef\u80fd\u66f4\u597d\u3002<\/p>\n<pre class=\"post-pre\"><code><span class=\"gh\">diff --git a\/configure.py b\/configure.py\r\nindex fc33325..45971da 100644\r\n<\/span><span class=\"gd\">--- a\/configure.py\r\n<\/span><span class=\"gi\">+++ b\/configure.py\r\n<\/span><span class=\"p\">@@ -486,7 +486,7 @@<\/span> def set_cc_opt_flags(environ_cp):\r\n   elif is_windows():\r\n     default_cc_opt_flags = '\/arch:AVX'\r\n   else:\r\n<span class=\"gd\">-    default_cc_opt_flags = '-march=haswell'\r\n<\/span><span class=\"gi\">+    default_cc_opt_flags = '-march=native'\r\n<\/span>   question = ('Please specify optimization flags to use during compilation when'\r\n               ' bazel option \"--config=opt\" is specified [Default is %s]: '\r\n              ) % default_cc_opt_flags\r\n<span class=\"p\">@@ -496,7 +496,7 @@<\/span> def set_cc_opt_flags(environ_cp):\r\n     write_to_bazelrc('build:opt --copt=%s' % opt)\r\n   # It should be safe on the same build host.\r\n   if not is_ppc64le() and not is_windows():\r\n<span class=\"gd\">-    write_to_bazelrc('build:opt --host_copt=-march=haswell')\r\n<\/span><span class=\"gi\">+    write_to_bazelrc('build:opt --host_copt=-march=native')\r\n<\/span>   write_to_bazelrc('build:opt --define with_default_optimizations=true')\r\n<\/code><\/pre>\n<p>\u5982\u679c\u5df2\u7ecf\u5728\u5176\u4ed6\u5730\u65b9\u6784\u5efa\u4e86 tensorflow\uff0c\u6700\u597d\u5220\u9664 $HOME\/.cache\/bazel \u76ee\u5f55\u4e0b\u7684\u7f13\u5b58\u6587\u4ef6\uff0c\u7136\u540e\u8fd0\u884c\u4ee5\u4e0b\u914d\u7f6e\u547d\u4ee4\u4ee5\u786e\u4fdd\u987a\u5229\u8fd0\u884c\u3002<\/p>\n<p>\u8fd0\u884c configure\uff0c\u8fdb\u884c\u521d\u59cb\u8bbe\u7f6e\u3002<br \/>\n\u5927\u4f53\u4e0a\u662f\u6309\u9ed8\u8ba4\u65b9\u5f0f\u8fdb\u884c\u7684\uff0c\u4f46\u6709\u4e9b\u4fee\u6539\u7684\u5730\u65b9\u662f\u5c06 XLA JIT \u8bbe\u7f6e\u4e3a n\uff08\u4e5f\u53ef\u4ee5\u4f7f\u7528 Y\uff09\u3002<\/p>\n<pre class=\"post-pre\"><code>$ \r\n$ .\/configure\r\nYou have bazel 0.15.2 installed.\r\nPlease specify the location of python. [Default is \/home\/syoyo\/miniconda3\/envs\/py35\/bin\/python]: \r\n\r\n\r\nFound possible Python library paths:\r\n  \/home\/syoyo\/miniconda3\/envs\/py35\/lib\/python3.5\/site-packages\r\nPlease input the desired Python library path to use.  Default is [\/home\/syoyo\/miniconda3\/envs\/py35\/lib\/python3.5\/site-packages]\r\n\r\nDo you wish to build TensorFlow with jemalloc as malloc support? [Y\/n]: Y\r\njemalloc as malloc support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with Google Cloud Platform support? [y\/N]: N\r\nNo Google Cloud Platform support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with Hadoop File System support? [y\/N]: N\r\nNo Hadoop File System support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with Amazon AWS Platform support? [y\/N]: N\r\nNo Amazon AWS Platform support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with Apache Kafka Platform support? [y\/N]: N\r\nNo Apache Kafka Platform support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with XLA JIT support? [Y\/n]: N\r\nNo XLA JIT support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with GDR support? [y\/N]: N\r\nNo GDR support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with VERBS support? [y\/N]: N\r\nNo VERBS support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with OpenCL SYCL support? [y\/N]: N \r\nNo OpenCL SYCL support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with ROCm support? [Y\/n]: Y\r\nROCm support will be enabled for TensorFlow.\r\n\r\nDo you wish to build TensorFlow with CUDA support? [y\/N]: N\r\nNo CUDA support will be enabled for TensorFlow.\r\n\r\nDo you wish to download a fresh release of clang? (Experimental) [y\/N]: N\r\nClang will not be downloaded.\r\n\r\nDo you wish to build TensorFlow with MPI support? [y\/N]: N\r\nNo MPI support will be enabled for TensorFlow.\r\n\r\nPlease specify optimization flags to use during compilation when bazel option \"--config=opt\" is specified [Default is -march=native]: \r\n\r\n\r\nWould you like to interactively configure .\/WORKSPACE for Android builds? [y\/N]: N\r\nNot configuring the WORKSPACE for Android builds.\r\n<\/code><\/pre>\n<p>\u4f7f\u7528 build_python3 \u811a\u672c\u8fdb\u884c\u6784\u5efa\u3002<\/p>\n<pre class=\"post-pre\"><code>(py35) $ .\/build_python3\r\n<\/code><\/pre>\n<p>\u5982\u679c\u5728build_python3\u811a\u672c\u7684\u6700\u540e\u4e00\u6b65\u5b89\u88c5pip wheel\u65f6\u51fa\u73b0\u6743\u9650\u9519\u8bef\uff0c\u8bf7\u5728&#8211;user\u5904\u8fdb\u884c\u5b89\u88c5\u3002<\/p>\n<pre class=\"post-pre\"><code>(py35) $ pip3 install --user \/tmp\/tensorflow_pkg\/tensorflow-1.9.0rc0-cp35-cp35m-linux_x86_64.whl\r\n<\/code><\/pre>\n<h2>\u786e\u8ba4\u52a8\u4f5c<\/h2>\n<p>\u4f7f\u7528GPU<br \/>\nhttps:\/\/www.tensorflow.org\/guide\/using_gpu<\/p>\n<p>\u6211\u5c06\u5c1d\u8bd5\u8fd0\u884c\u4e00\u4e2a\u7b80\u5355\u7684\u52a0\u6cd5\u7a0b\u5e8f\u3002<\/p>\n<pre class=\"post-pre\"><code>2018-07-21 22:05:25.506635: I tensorflow\/core\/platform\/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX512F\r\n2018-07-21 22:05:25.507760: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1500] Found device 0 with properties: \r\nname: Device 67df\r\nAMDGPU ISA: gfx803\r\nmemoryClockRate (GHz) 1.35\r\npciBusID 0000:65:00.0\r\nTotal memory: 8.00GiB\r\nFree memory: 7.75GiB\r\n2018-07-21 22:05:25.507806: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1611] Adding visible gpu devices: 0\r\n2018-07-21 22:05:25.507838: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1031] Device interconnect StreamExecutor with strength 1 edge matrix:\r\n2018-07-21 22:05:25.507854: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1037]      0 \r\n2018-07-21 22:05:25.507870: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1050] 0:   N \r\n2018-07-21 22:05:25.507938: I tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:1168] Created TensorFlow device (\/job:localhost\/replica:0\/task:0\/device:GPU:0 with 7539 MB memory) -&gt; physical GPU (device: 0, name: Device 67df, pci bus id: 0000:65:00.0)\r\n2018-07-21 22:05:25.523000: E tensorflow\/core\/common_runtime\/gpu\/gpu_device.cc:264] Illegal GPUOptions.experimental.num_dev_to_dev_copy_streams=0 set to 1 instead.\r\nDevice mapping:\r\n\/job:localhost\/replica:0\/task:0\/device:GPU:0 -&gt; device: 0, name: Device 67df, pci bus id: 0000:65:00.0\r\n2018-07-21 22:05:26.074124: I tensorflow\/core\/common_runtime\/direct_session.cc:288] Device mapping:\r\n\/job:localhost\/replica:0\/task:0\/device:GPU:0 -&gt; device: 0, name: Device 67df, pci bus id: 0000:65:00.0\r\n\r\nMatMul: (MatMul): \/job:localhost\/replica:0\/task:0\/device:GPU:0\r\n2018-07-21 22:05:26.075049: I tensorflow\/core\/common_runtime\/placer.cc:935] MatMul: (MatMul)\/job:localhost\/replica:0\/task:0\/device:GPU:0\r\na: (Const): \/job:localhost\/replica:0\/task:0\/device:GPU:0\r\n2018-07-21 22:05:26.075094: I tensorflow\/core\/common_runtime\/placer.cc:935] a: (Const)\/job:localhost\/replica:0\/task:0\/device:GPU:0\r\nb: (Const): \/job:localhost\/replica:0\/task:0\/device:GPU:0\r\n2018-07-21 22:05:26.075117: I tensorflow\/core\/common_runtime\/placer.cc:935] b: (Const)\/job:localhost\/replica:0\/task:0\/device:GPU:0\r\n[[22. 28.]\r\n [49. 64.]]\r\n<\/code><\/pre>\n<p>\u770b\u54ea~\uff01\u6709\u4e00\u4e9b\u5b9e\u9a8c\u6027\u9519\u8bef\u51fa\u73b0\u4e86\uff0c\u4f46\u662f\u5b83\u8fd8\u662f\u6b63\u5e38\u5de5\u4f5c\u4e86\uff01<\/p>\n<h2>CIFAR10 \u53ef\u4ee5\u88ab\u7b80\u6d01\u7684\u89e3\u91ca\u4e3a\u4e2d\u6587\u8bc6\u522b\u56fe\u50cf\u5341\u4e2a\u5206\u7c7b\u6570\u636e\u96c6\u3002<\/h2>\n<p>\u6211\u5c06\u4f7f\u7528CIFAR10\u8fdb\u884c\u5b66\u4e60\u3002<br \/>\n\u5c3d\u7ba1\u5728rocm-smi\u4e0b\u964d\u4e86\u65f6\u949f\u901f\u5ea6\uff0c\u4f46VEGA56\u4ee580W\u7684\u529f\u8017\u8fd0\u884c\uff0c\u6bcf\u79d2\u751f\u621011500\u4e2a\u6837\u4f8b\u3002<\/p>\n<blockquote><p>ROCe tensorflow-upstream \u9879\u76ee\u5728\u8bad\u7ec3 CIFAR10 \u65b9\u9762\u53d6\u5f97\u4e86\u6210\u529f\uff01\u3297\ufe0f\u3297\ufe0f\u3297\ufe0f\u3297\ufe0f\u3297\ufe0f\u3297\ufe0f\u3297\ufe0f??????????????????????????? VEGA56 setsclk 3 \u5728 80W \u4e0b\u8fbe\u5230\u4e86 1150 \u4e2a example\/\u79d2\u3002\u8c22\u8c22\u3002pic.twitter.com\/lpNfF5FDWE\u2014 Syoyo Fujita (@syoyo) July 22, 2018<\/p><\/blockquote>\n<p><script><\/script><\/p>\n<p>\u5728TF1.8\u4e2d\uff0c125W\u529f\u8017\u9650\u5236\u4e0b\uff0c1080 Ti\u7684\u6bcf\u79d2\u5904\u7406\u7684\u8303\u4f8b\u6570\u572810500\u81f311500\u4e4b\u95f4\u3002<\/p>\n<p>\u7531\u4e8e\u4e0eNV GPU\u76f8\u6bd4\uff0cVEGA\u7684\u6027\u80fd\u5dee\u4e0d\u591a\uff0c\u6240\u4ee5\u6027\u4ef7\u6bd4\u4f3c\u4e4e\u5f88\u9ad8\uff0c\u5c3d\u7ba1\u5b83\u7684\u5185\u5b58\u5bb9\u91cf\u8f83\u5c0f\uff08VEGA56\u4e3a8GB\uff0c1080 Ti\u4e3a11GB\uff09\uff0c\u4f46\u8fd9\u4e0d\u662f\u95ee\u9898\u3002\u56e0\u4e3aCIFAR10\u80fd\u6b63\u5e38\u8fd0\u884c\uff0c\u6240\u4ee5\u81f3\u5c11\u5305\u62ecCNN\u5728\u5185\u7684\u5b66\u4e60\u5e94\u8be5\u6ca1\u6709\u95ee\u9898\uff0c\u5e94\u8be5\u53ef\u4ee5\u5728AMD GPU\u4e0a\u8fd0\u884c\u3002<\/p>\n<h2>\u95ee\u9898\u6240\u5728<\/h2>\n<p>\u5bfc\u5165tensorflow\u592a\u91cd\u4e86\uff0c\u975e\u5e38\u56f0\u96be&#8230;(\u5927\u7ea6\u9700\u89815-6\u79d2). \u60f3\u77e5\u9053\u662f\u4e0d\u662f\u6709\u67d0\u4e2a\u6a21\u5757\u88ab\u8bbe\u7f6e\u6210\u4e86\u8c03\u8bd5\u7248\u672c?<\/p>\n<h2>\u6784\u5efa\u5931\u8d25\u793a\u4f8b<\/h2>\n<p>\u5982\u679c hipcc \u7684\u8bbe\u7f6e\uff08\u6216\u8005 clang \u7684\u5b89\u88c5\uff1f\uff09\u51fa\u73b0\u4e86\u95ee\u9898\uff0c\u90a3\u4e48\u6709\u53ef\u80fd\u51fa\u73b0 hipcc \u8c03\u7528 nvcc\uff08CUDA\uff09\u5bfc\u81f4\u6784\u5efa\u5931\u8d25\u7684\u60c5\u51b5\uff08\u5373\u4f7f\u5728\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf\u4e3a hcc \u7684\u60c5\u51b5\u4e0b\uff0c\u5728 bazel \u73af\u5883\u4e2d\u4f1a\u88ab\u6539\u5199\u6210 nvcc\uff09\u3002<\/p>\n<p>\u4f8b\u5982\uff0c\u5b58\u5728\u4ee5\u4e0b\u7c7b\u4f3c\u95ee\u9898\u3002<\/p>\n<p>\u5bf9\u4e8e\u8fd9\u4e2a\u95ee\u9898\u6211\u4e0d\u662f\u5f88\u6e05\u695a\u3002\u5982\u679c\u9047\u5230\u8fd9\u79cd\u60c5\u51b5\uff0c\u4f3c\u4e4e\u53ea\u80fd\u5c1d\u8bd5\u6e05\u9664\u7f13\u5b58\u6216\u91cd\u65b0\u5b89\u88c5\u5305\u3002<\/p>\n<h2>\u628a\u4ee5\u4e0b\u5185\u5bb9\u7528\u4e2d\u6587\u8fdb\u884c\u672c\u5730\u5316\u6539\u5199\uff0c\u53ea\u9700\u8981\u4e00\u79cd\u9009\u9879:<\/h2>\n<p>\u5f85\u529e\u4e8b\u9879<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">CIFAR10 \u5b66\u7fd2\u306e\u7cbe\u5ea6\u3092 NVIDIA GPU \u3068\u6bd4\u8f03\u3059\u308b.<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">\u4eee\u60f3\u901a\u8ca8\u30de\u30a4\u30cb\u30f3\u30b0\u30d7\u30ed\u30b0\u30e9\u30e0\u3068\u30ab\u30fc\u30cd\u30eb\u30d5\u30fc\u30b8\u30e7\u30f3\u3057, \u6a5f\u68b0\u5b66\u7fd2\u3057\u3064\u3064\u30de\u30a4\u30cb\u30f3\u30b0\u3059\u308b\u30b9\u30ad\u30fc\u30e0\u3092\u78ba\u7acb\u3057\u305f\u3044.<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">pytorch \u3092 ROCm \u3067\u30d3\u30eb\u30c9\u3059\u308b(\u30d3\u30eb\u30c9\u30aa\u30d7\u30b7\u30e7\u30f3\u306f\u3042\u308b\u304c, \u5b9f\u969b\u306b\u52d5\u304f\u304b\u3069\u3046\u304b)<\/ul>\n<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<ul class=\"post-ul\">\n<li style=\"list-style-type: none;\">\n<ul class=\"post-ul\">InfiniBand(VERBS) \u6709\u52b9\u306b\u3057, GPU \u30af\u30e9\u30b9\u30bf\u3067\u4e26\u5217\u5b66\u7fd2\u3057\u305f\u3044.<\/ul>\n<\/li>\n<\/ul>\n<p>\u512a\u79c0\u306a\u6a5f\u68b0\u5b66\u7fd2\u82e5\u4eba\u3055\u307e\u304c, \u512a\u79c0\u306a ROCm \u6a5f\u68b0\u5b66\u7fd2\u82e5\u4eba\u3055\u307e\u3078\u3068\u4eba\u985e\u53f2\u4e0a\u6700\u901f\u3067\u6607\u83ef\u306a\u3055\u308c\u308b\u30b9\u30ad\u30fc\u30e0\u3092\u6975\u3081\u308b\u65c5\u306b\u51fa\u305f\u3044.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>2019\u5e747\u670816\u65e5\u66f4\u65b0\uff1a\u6700\u8fd1\uff0cpip install tensorflow-rocm\u5df2\u7ecf\u8db3\u591f\u53ef\u7528\uff0c\u56e0\u6b64\u5982\u679c\u53ea [&hellip;]<\/p>\n","protected":false},"author":4,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-47449","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"yoast_head":"<!-- This site is optimized with the Yoast SEO Premium plugin v21.5 (Yoast SEO v21.5) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5) - Blog - Silicon Cloud<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-\u7684-tensorflow-upstream1-9-0-rc0-\u7f16\u8bd1\u5c1d\u8bd5\uff08cifar10-11500\u4e2a\u6837\u4f8b-\u79d2-vega56-80w-\u622a\u81f32018\u5e747\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5)\" \/>\n<meta property=\"og:description\" content=\"2019\u5e747\u670816\u65e5\u66f4\u65b0\uff1a\u6700\u8fd1\uff0cpip install tensorflow-rocm\u5df2\u7ecf\u8db3\u591f\u53ef\u7528\uff0c\u56e0\u6b64\u5982\u679c\u53ea [&hellip;]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-\u7684-tensorflow-upstream1-9-0-rc0-\u7f16\u8bd1\u5c1d\u8bd5\uff08cifar10-11500\u4e2a\u6837\u4f8b-\u79d2-vega56-80w-\u622a\u81f32018\u5e747\/\" \/>\n<meta property=\"og:site_name\" content=\"Blog - Silicon Cloud\" \/>\n<meta property=\"article:published_time\" content=\"2024-02-14T22:04:39+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-04-30T08:36:39+00:00\" \/>\n<meta name=\"author\" content=\"\u79d1, \u96c5\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"\u79d1, \u96c5\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"5 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/\",\"name\":\"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5) - Blog - Silicon Cloud\",\"isPartOf\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\"},\"datePublished\":\"2024-02-14T22:04:39+00:00\",\"dateModified\":\"2024-04-30T08:36:39+00:00\",\"author\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/41e222757cdd2a3365361328bd79970a\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.silicloud.com\/zh\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5)\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/\",\"name\":\"Blog - Silicon Cloud\",\"description\":\"\",\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/41e222757cdd2a3365361328bd79970a\",\"name\":\"\u79d1, \u96c5\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/1b2d3e00a7df03689797ebd4af8c5827ba5af936849a71050ec331f4cf902c5d?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/1b2d3e00a7df03689797ebd4af8c5827ba5af936849a71050ec331f4cf902c5d?s=96&d=mm&r=g\",\"caption\":\"\u79d1, \u96c5\"},\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/author\/keya\/\"},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#local-main-organization-logo\",\"url\":\"\",\"contentUrl\":\"\",\"caption\":\"Blog - Silicon Cloud\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5) - Blog - Silicon Cloud","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-\u7684-tensorflow-upstream1-9-0-rc0-\u7f16\u8bd1\u5c1d\u8bd5\uff08cifar10-11500\u4e2a\u6837\u4f8b-\u79d2-vega56-80w-\u622a\u81f32018\u5e747\/","og_locale":"zh_CN","og_type":"article","og_title":"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5)","og_description":"2019\u5e747\u670816\u65e5\u66f4\u65b0\uff1a\u6700\u8fd1\uff0cpip install tensorflow-rocm\u5df2\u7ecf\u8db3\u591f\u53ef\u7528\uff0c\u56e0\u6b64\u5982\u679c\u53ea [&hellip;]","og_url":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-\u7684-tensorflow-upstream1-9-0-rc0-\u7f16\u8bd1\u5c1d\u8bd5\uff08cifar10-11500\u4e2a\u6837\u4f8b-\u79d2-vega56-80w-\u622a\u81f32018\u5e747\/","og_site_name":"Blog - Silicon Cloud","article_published_time":"2024-02-14T22:04:39+00:00","article_modified_time":"2024-04-30T08:36:39+00:00","author":"\u79d1, \u96c5","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"\u79d1, \u96c5","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"5 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/","url":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/","name":"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5) - Blog - Silicon Cloud","isPartOf":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website"},"datePublished":"2024-02-14T22:04:39+00:00","dateModified":"2024-04-30T08:36:39+00:00","author":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/41e222757cdd2a3365361328bd79970a"},"breadcrumb":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.silicloud.com\/zh\/blog\/"},{"@type":"ListItem","position":2,"name":"ROCm 1.8.2 \u7684 TensorFlow upstream(1.9.0-rc0) \u7f16\u8bd1\u5c1d\u8bd5\uff08CIFAR10 11500\u4e2a\u6837\u4f8b\/\u79d2 @ VEGA56 80W. \u622a\u81f32018\u5e747\u670822\u65e5)"}]},{"@type":"WebSite","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website","url":"https:\/\/www.silicloud.com\/zh\/blog\/","name":"Blog - Silicon Cloud","description":"","inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/41e222757cdd2a3365361328bd79970a","name":"\u79d1, \u96c5","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/1b2d3e00a7df03689797ebd4af8c5827ba5af936849a71050ec331f4cf902c5d?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/1b2d3e00a7df03689797ebd4af8c5827ba5af936849a71050ec331f4cf902c5d?s=96&d=mm&r=g","caption":"\u79d1, \u96c5"},"url":"https:\/\/www.silicloud.com\/zh\/blog\/author\/keya\/"},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/rocm-1-8-2-%e7%9a%84-tensorflow-upstream1-9-0-rc0-%e7%bc%96%e8%af%91%e5%b0%9d%e8%af%95%ef%bc%88cifar10-11500%e4%b8%aa%e6%a0%b7%e4%be%8b-%e7%a7%92-vega56-80w-%e6%88%aa%e8%87%b32018%e5%b9%b47\/#local-main-organization-logo","url":"","contentUrl":"","caption":"Blog - Silicon Cloud"}]}},"_links":{"self":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/47449","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/comments?post=47449"}],"version-history":[{"count":2,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/47449\/revisions"}],"predecessor-version":[{"id":93041,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/47449\/revisions\/93041"}],"wp:attachment":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/media?parent=47449"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/categories?post=47449"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/tags?post=47449"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}