Trying hard to install tfjs-node-gpu on Windows 10

[theodorDiaconu]

Using python2.7
Installed windows utilities

Stuck here:

ENOENT 'C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\build\Release\tensorflow.dll' 

gyp info it worked if it ends with ok
gyp info using node-gyp@3.8.0
gyp info using node@10.15.3 | win32 | x64
gyp info spawn C:\Program Files (x86)\Microsoft Visual Studio\2017\BuildTools\MSBuild\15.0\Bin\MSBuild.exe
gyp info spawn args [ 'build/binding.sln',
gyp info spawn args   '/clp:Verbosity=minimal',
gyp info spawn args   '/nologo',
gyp info spawn args   '/p:Configuration=Release;Platform=x64' ]
Building the projects in this solution one at a time. To enable parallel build, please add the "/m" switch.
  deps-stage
  (node:9872) UnhandledPromiseRejectionWarning: Error: ENOENT: no such file or directory, rename 'C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\deps\lib\tensorflow.dl
  l' -> 'C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\build\Release\tensorflow.dll'
  (node:9872) UnhandledPromiseRejectionWarning: Unhandled promise rejection. This error originated either by throwing inside of an async function without a catch block, or by rej
  ecting a promise which was not handled with .catch(). (rejection id: 1)
  (node:9872) [DEP0018] DeprecationWarning: Unhandled promise rejections are deprecated. In the future, promise rejections that are not handled will terminate the Node.js process
   with a non-zero exit code.
  generate_def
  fs.js:114
      throw err;
      ^

CUSTOMBUILD : error : ENOENT: no such file or directory, open 'C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\build\Release\tensorflow.dll' [C:\Projects\tsflow\node_mo
dules\@tensorflow\tfjs-node-gpu\build\tfjs_binding.vcxproj]
      at Object.openSync (fs.js:438:3)
      at Object.readFileSync (fs.js:343:35)
      at files.map (C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\scripts\generate_defs.js:23:40)
      at Array.map (<anonymous>)
      at Object.<anonymous> (C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\scripts\generate_defs.js:23:23)
      at Module._compile (internal/modules/cjs/loader.js:701:30)
      at Object.Module._extensions..js (internal/modules/cjs/loader.js:712:10)
      at Module.load (internal/modules/cjs/loader.js:600:32)
      at tryModuleLoad (internal/modules/cjs/loader.js:539:12)
      at Function.Module._load (internal/modules/cjs/loader.js:531:3)
gyp ERR! build error
gyp ERR! stack Error: `C:\Program Files (x86)\Microsoft Visual Studio\2017\BuildTools\MSBuild\15.0\Bin\MSBuild.exe` failed with exit code: 1
gyp ERR! stack     at ChildProcess.onExit (C:\Users\Pc\AppData\Roaming\npm\node_modules\node-gyp\lib\build.js:262:23)
gyp ERR! stack     at ChildProcess.emit (events.js:189:13)
gyp ERR! stack     at Process.ChildProcess._handle.onexit (internal/child_process.js:248:12)
gyp ERR! System Windows_NT 10.0.17763
gyp ERR! command "C:\\Program Files\\nodejs\\node.exe" "C:\\Users\\Pc\\AppData\\Roaming\\npm\\node_modules\\node-gyp\\bin\\node-gyp.js" "build"
gyp ERR! cwd C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu
gyp ERR! node -v v10.15.3
gyp ERR! node-gyp -v v3.8.0
gyp ERR! not ok```

[nkreeger]

Can you try deleting your node_modules folder and trying again? We also have some troubleshooting tips here:

https://github.com/tensorflow/tfjs-node/blob/master/WINDOWS_TROUBLESHOOTING.md

[theodorDiaconu]

@nkreeger yes I followed those. I made sure the msvs_version of npm is 2013, I have python 2.7 installed. I ran this also: npm --add-python-to-path='true' --debug install --global windows-build-tools

I'm still getting:

* Building TensorFlow Node.js bindings
C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node\scripts\install.js:165
      throw new Error('node-gyp rebuild failed with: ' + err);
      ^

Error: node-gyp rebuild failed with: Error: Command failed: node-gyp rebuild
gyp ERR! build error
gyp ERR! stack Error: `C:\Program Files (x86)\MSBuild\14.0\bin\msbuild.exe` failed with exit code: 1
gyp ERR! stack     at ChildProcess.onExit (C:\Program Files\nodejs\node_modules\npm\node_modules\node-gyp\lib\build.js:262:23)
gyp ERR! stack     at ChildProcess.emit (events.js:189:13)
gyp ERR! stack     at Process.ChildProcess._handle.onexit (internal/child_process.js:248:12)
gyp ERR! System Windows_NT 10.0.17763
gyp ERR! command "C:\\Program Files\\nodejs\\node.exe" "C:\\Program Files\\nodejs\\node_modules\\npm\\node_modules\\node-gyp\\bin\\node-gyp.js" "rebuild"
gyp ERR! cwd C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node
gyp ERR! node -v v10.15.3
gyp ERR! node-gyp -v v3.8.0
gyp ERR! not ok

    at cp.exec (C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node\scripts\install.js:165:13)
    at ChildProcess.exithandler (child_process.js:301:5)
    at ChildProcess.emit (events.js:189:13)
    at maybeClose (internal/child_process.js:970:16)
    at Process.ChildProcess._handle.onexit (internal/child_process.js:259:5)
npm WARN rollup-plugin-visualizer@1.1.1 requires a peer of rollup@>=0.60.0 but none is installed. You must install peer dependencies yourself.
npm WARN tsflow@1.0.0 No description
npm WARN tsflow@1.0.0 No repository field.

npm ERR! code ELIFECYCLE
npm ERR! errno 1
npm ERR! @tensorflow/tfjs-node@1.1.2 install: `node scripts/install.js`
npm ERR! Exit status 1
npm ERR!
npm ERR! Failed at the @tensorflow/tfjs-node@1.1.2 install script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.

npm ERR! A complete log of this run can be found in:
npm ERR!     C:\Users\Pc\AppData\Roaming\npm-cache\_logs\2019-05-17T16_47_58_637Z-debug.log

[theodorDiaconu]

What I did and it worked for tfjs-node:

npm i -S @tensorflow/tfjs-node -f

Let it fail. Go to node_modules folder. And run node-gyp rebuild it worked.

[theodorDiaconu]

However, my plan is to run training with the GPU, to have 10x speeds. But if I apply the same strategy there here's what I get:

$ node-gyp rebuild
gyp info it worked if it ends with ok
gyp info using node-gyp@3.8.0
gyp info using node@10.15.3 | win32 | x64
gyp info spawn C:\Python27\python.EXE
gyp info spawn args [ 'C:\\Users\\Pc\\AppData\\Roaming\\npm\\node_modules\\node-gyp\\gyp\\gyp_main.py',
gyp info spawn args   'binding.gyp',
gyp info spawn args   '-f',
gyp info spawn args   'msvs',
gyp info spawn args   '-G',
gyp info spawn args   'msvs_version=2015',
gyp info spawn args   '-I',
gyp info spawn args   'C:\\Projects\\tsflow\\node_modules\\@tensorflow\\tfjs-node-gpu\\build\\config.gypi',
gyp info spawn args   '-I',
gyp info spawn args   'C:\\Users\\Pc\\AppData\\Roaming\\npm\\node_modules\\node-gyp\\addon.gypi',
gyp info spawn args   '-I',
gyp info spawn args   'C:\\Users\\Pc\\.node-gyp\\10.15.3\\include\\node\\common.gypi',
gyp info spawn args   '-Dlibrary=shared_library',
gyp info spawn args   '-Dvisibility=default',
gyp info spawn args   '-Dnode_root_dir=C:\\Users\\Pc\\.node-gyp\\10.15.3',
gyp info spawn args   '-Dnode_gyp_dir=C:\\Users\\Pc\\AppData\\Roaming\\npm\\node_modules\\node-gyp',
gyp info spawn args   '-Dnode_lib_file=C:\\Users\\Pc\\.node-gyp\\10.15.3\\<(target_arch)\\node.lib',
gyp info spawn args   '-Dmodule_root_dir=C:\\Projects\\tsflow\\node_modules\\@tensorflow\\tfjs-node-gpu',
gyp info spawn args   '-Dnode_engine=v8',
gyp info spawn args   '--depth=.',
gyp info spawn args   '--no-parallel',
gyp info spawn args   '--generator-output',
gyp info spawn args   'C:\\Projects\\tsflow\\node_modules\\@tensorflow\\tfjs-node-gpu\\build',
gyp info spawn args   '-Goutput_dir=.' ]
gyp info spawn C:\Program Files (x86)\Microsoft Visual Studio\2017\BuildTools\MSBuild\15.0\Bin\MSBuild.exe
gyp info spawn args [ 'build/binding.sln',
gyp info spawn args   '/clp:Verbosity=minimal',
gyp info spawn args   '/nologo',
gyp info spawn args   '/p:Configuration=Release;Platform=x64' ]
Building the projects in this solution one at a time. To enable parallel build, please add the "/m" switch.
  deps-stage
  generate_def
CUSTOMBUILD : FATAL error : CALL_AND_RETRY_LAST Allocation failed - JavaScript heap out of memory [C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu\build\tfjs_binding.vcxproj]
   1: 00007FF6AFE4F04A v8::internal::GCIdleTimeHandler::GCIdleTimeHandler+5114
   2: 00007FF6AFE2A0C6 node::MakeCallback+4518
   3: 00007FF6AFE2AA30 node_module_register+2032
   4: 00007FF6B00B20EE v8::internal::FatalProcessOutOfMemory+846
   5: 00007FF6B00B201F v8::internal::FatalProcessOutOfMemory+639
   6: 00007FF6B05D2BC4 v8::internal::Heap::MaxHeapGrowingFactor+9556
   7: 00007FF6B05D10CB v8::internal::Heap::MaxHeapGrowingFactor+2651
   8: 00007FF6B018CA3B v8::internal::Factory::AllocateRawWithImmortalMap+59
   9: 00007FF6B018F4ED v8::internal::Factory::NewRawTwoByteString+77
  10: 00007FF6B018E67E v8::internal::Factory::NewStringFromUtf8+238
  11: 00007FF6B00CD8FB v8::String::NewFromUtf8+363
  12: 00007FF6AFE184C1 node::Buffer::New+24225
  13: 00007FF6B062C122 v8::internal::OptimizingCompileDispatcher::Unblock+59890
  14: 00007FF6B062D5AD v8::internal::OptimizingCompileDispatcher::Unblock+65149
  15: 00007FF6B062C606 v8::internal::OptimizingCompileDispatcher::Unblock+61142
  16: 00007FF6B062C4EB v8::internal::OptimizingCompileDispatcher::Unblock+60859
  17: 0000021BD9CDC5C1
gyp ERR! build error
gyp ERR! stack Error: `C:\Program Files (x86)\Microsoft Visual Studio\2017\BuildTools\MSBuild\15.0\Bin\MSBuild.exe` failed with exit code: 1
gyp ERR! stack     at ChildProcess.onExit (C:\Users\Pc\AppData\Roaming\npm\node_modules\node-gyp\lib\build.js:262:23)
gyp ERR! stack     at ChildProcess.emit (events.js:189:13)
gyp ERR! stack     at Process.ChildProcess._handle.onexit (internal/child_process.js:248:12)
gyp ERR! System Windows_NT 10.0.17763
gyp ERR! command "C:\\Program Files\\nodejs\\node.exe" "C:\\Users\\Pc\\AppData\\Roaming\\npm\\node_modules\\node-gyp\\bin\\node-gyp.js" "rebuild"
gyp ERR! cwd C:\Projects\tsflow\node_modules\@tensorflow\tfjs-node-gpu
gyp ERR! node -v v10.15.3
gyp ERR! node-gyp -v v3.8.0
gyp ERR! not ok

[adwellj]

@theodorDiaconu, does 0.3.2 install?
I (along with others) struggled with installing tfjs-node-gpu on Windows, but had success with 0.3.2 Trying any of the versions after would result in installation errors.

#1098 (comment)

[theodorDiaconu]

@adwellj I managed to be able to install that, by doing npm i -S {package}--force, then going manually inside it and running node-gyp configure and node-gyp rebuild.

Still with 1.1.2 node-gyp:
This is the error:

CUSTOMBUILD : FATAL error : CALL_AND_RETRY_LAST Allocation failed - JavaScript heap out of memory [C:\Projects\tsflow2\node_modules\@tensorflow\tfjs-node-gpu\build\tfjs_binding.vcxproj]
   1: 00007FF61F62F04A v8::internal::GCIdleTimeHandler::GCIdleTimeHandler+5114
   2: 00007FF61F60A0C6 node::MakeCallback+4518
   3: 00007FF61F60AA30 node_module_register+2032
   4: 00007FF61F8920EE v8::internal::FatalProcessOutOfMemory+846
   5: 00007FF61F89201F v8::internal::FatalProcessOutOfMemory+639
   6: 00007FF61FDB2BC4 v8::internal::Heap::MaxHeapGrowingFactor+9556
   7: 00007FF61FDB10CB v8::internal::Heap::MaxHeapGrowingFactor+2651
   8: 00007FF61F96CA3B v8::internal::Factory::AllocateRawWithImmortalMap+59
   9: 00007FF61F96F4ED v8::internal::Factory::NewRawTwoByteString+77
  10: 00007FF61F96E67E v8::internal::Factory::NewStringFromUtf8+238
  11: 00007FF61F8AD8FB v8::String::NewFromUtf8+363
  12: 00007FF61F5F84C1 node::Buffer::New+24225
  13: 00007FF61FE0C122 v8::internal::OptimizingCompileDispatcher::Unblock+59890
  14: 00007FF61FE0D5AD v8::internal::OptimizingCompileDispatcher::Unblock+65149
  15: 00007FF61FE0C606 v8::internal::OptimizingCompileDispatcher::Unblock+61142
  16: 00007FF61FE0C4EB v8::internal::OptimizingCompileDispatcher::Unblock+60859
  17: 000003C9C1B5C5C1

[theodorDiaconu]

I am now happily using 0.3.2. The experience to get it working was absolutely awful. Still I was expecting better improvements node vs node-gpu. For a sample set that I used the difference is 16000ms vs 8000ms. So almost twice. Not bad but was expecting 10x.

[adwellj]

@theodorDiaconu, glad to hear you got it working! Agreed, GPU installation is a very poor experience at this time. Fingers crossed that with enough feedback the team will be able to fix the GPU installations for the new versions.

For future reference, I didn't have to apply any special flags or go back and do any rebuilds. I simply install npm i @tensorflow/tfjs-node-gpu@0.3.2.

Regarding the GPU performance, I'm going from 550 sec/epoch with CPU to 103 sec/epoch with GPU. I would imagine that some models will see a larger benefit with GPU than others. For instance: #468 (comment)

[theodorDiaconu]

Understood, indeed I had a small model. I'm still learning about this. Very glad that I know when I have to do serious training I can use my GPU.

This ticket should still be open. Building doesn't work on ^0.3.2

[kangyizhang]

With the latest tfjs-node/tfjs-node-gpu package, we are shipping pre-compile addon and there should be no compilation on your machine. This issue should have been fixed.