There is a desire to do the initial steps in tfjs using nodejs. At the moment, for tests, I can only use a computer with the following configuration: Windows 7 SP1 8Gb Ram e7500 (no AVX) GeForce 750Ti node v12.19.0
When using tfjs-node, I get the error:
return process.dlopen (module, path.toNamespacedPath (filename));
As far as I understand this is due to the fact that the processor is very old, without AVX. Can I somehow rebuild tfjs-node to work on my processor, it would be ideal to build tfjs-node-gpu. If there is such an opportunity, what should I do for this?
I've come across assemblies from fo40225 (https://github.com/fo40225), but they are for Python.
Solved a problem. First I tried changing Windows to Win10, it didn't help. Therefore, I decided to rebuild tensorflow.dll. After many attempts, I came up with this setup: Basel 3.1 Python 3.8 NumPy installed globally VS BuildTools 2019 Tensorflow branch 2.3 compiled (bazel build -c opt // tensorflow / tools / lib_package: libtensorflow)
After that I copied the dll to the folder with node_modules \ @tensorflow \ tfjs-node \ lib \ napi-v6