[FEATURE] Distribution size of app that uses langchain4j with in-process embedding models

[InAnYan]

Is your feature request related to a problem? Please describe.
I've been developing an app that uses langchain4j.

We have found a problem, that the distribution size of our app is too big.

I was studying the modules inside distribution to see why app changed it's size so much.

It turns out onnxruntime module takes up 376M:

(Windows takes up the most space)

It seems there are several versions inside 1 package for different OSes.

Describe the solution you'd like
Is there any way to remove unused modules for OS-arch?

Alternatives
Currently, I'm investigating the issue, maybe there are other ways to solve this

[InAnYan]

OMG they put .pdb inside a release???

And it's been like this for years?

Similar issue: microsoft/onnxruntime#12084

[InAnYan]

UPDATE: I've successfully used Deep Java Library for embedding generation and the distribution size is very small.

I could try to make a new langchain4j package for in-process models that uses not ONNX, but djl

[langchain4j]

@InAnYan hmm, that is interesting, where can I see/run the code?

[InAnYan]

@langchain4j you can take a look at this example: https://github.com/deepjavalibrary/djl-demo/blob/master/huggingface/nlp/src/main/java/com/examples/TextEmbedding.java

And also available models: deepjavalibrary/djl#3357 (comment)

[langchain4j]

@InAnYan looks nice! I have nothing against a new package, but I have lots of other things right now and I won't be able to help an/or commit to maintain it afterwards. If you plan to maintain it, go ahead and we can add it on https://github.com/langchain4j/awesome-langchain4j page and in the documentation