tfjs-node does not work with Node14 or Node 12 >= 12.17

[jeffrson]

TensorFlow.js version

1.7.4

Describe the problem or feature request

This is due to the new N-API-Version '6' which is missing from package.json.

The library builds apparently without issues (as N-API '5'), but the install.js script fails to symlink or copy tensorflow.dll (because folder for N-API '6' does not exist).

Code to reproduce the bug / link to feature request

With node14:

yarn init -y
yarn add "@tensorflow/tfjs-node" (also with "-gpu")
node -r "@tensorflow/tfjs-node"
dir ".\node_modules\@tensorflow\tfjs-node\lib"
dir ".\node_modules\@tensorflow\tfjs-node\lib\napi-v5"
jq .binary.napi_versions ".\node_modules\@tensorflow\tfjs-node\package.json"
(misses version 6)

[TheNewSound]

Same goes for node v13.12.0
gyp ERR! configure error gyp ERR! stack Error: `gyp` failed with exit code: 1 gyp ERR! stack at ChildProcess.onCpExit (C:\Users\Admin\AppData\Roaming\npm\node_modules\node-gyp\lib\configure.js:351:16) gyp ERR! stack at ChildProcess.emit (events.js:315:20) gyp ERR! stack at Process.ChildProcess._handle.onexit (internal/child_process.js:275:12) gyp ERR! System Windows_NT 10.0.18362 gyp ERR! command "C:\\Program Files\\nodejs\\node.exe" "C:\\Users\\Admin\\AppData\\Roaming\\npm\\node_modules\\node-gyp\\bin\\node-gyp.js" "configure" "--verbose" gyp ERR! cwd C:\Users\Admin\folder\node_modules\@tensorflow\tfjs-node gyp ERR! node -v v13.12.0 gyp ERR! node-gyp -v v6.1.0 gyp ERR! not ok

[kangyizhang]

I have created https://github.com/tensorflow/tfjs/pull/3242 to fix this. Please use node version smaller than 14 for now. The next release of this library will support the latest node version.

[jeffrson]

As it turns out, Node 12 changed to Napi6 as well starting from 12.17.0;
https://github.com/nodejs/node/blob/master/doc/changelogs/CHANGELOG_V12.md#stable-n-api-release-6 so you have to stick to 12.16.3 for now.

[TheNewSound]

I'm not 100% sure but TFJS-node 2.0.0 was released 2 weeks ago or so, I think it now has Napi6 support?

[jeffrson]

Well, maybe, yes. Unfortunately, currently it seems I still need to use tensorflow 1.15.

[davidkhanpk]

"@tensorflow/tfjs": "^2.6.0",
"@tensorflow/tfjs-node": "^2.6.0",
node_version: "12.19.0"

internal/modules/cjs/loader.js:1065
return process.dlopen(module, path.toNamespacedPath(filename));
^

Error: The specified module could not be found.
\?\C:\Users..\node_modules\@tensorflow\tfjs-node\lib\napi-v6\tfjs_binding.node

[rpritr]

"@tensorflow/tfjs": "^2.6.0",
"@tensorflow/tfjs-node": "^2.6.0",
node_version: "12.19.0"

internal/modules/cjs/loader.js:1065
return process.dlopen(module, path.toNamespacedPath(filename));
^

Error: The specified module could not be found.
\?\C:\Users..\node_modules@tensorflow\tfjs-node\lib\napi-v6\tfjs_binding.node

I used node 12.16.1 and It found the specified module, however I still get notice: Hi there. Looks like you are running TensorFlow.js in Node.js. To speed things up dramatically, install our node backend, which binds to TensorFlow C++, by running npm i @tensorflow/tfjs-node, or npm i @tensorflow/tfjs-node-gpu if you have CUDA. Then call require('@tensorflow/tfjs-node'); (-gpu suffix for CUDA) at the start of your program. Visit https://github.com/tensorflow/tfjs-node for more details.