[Showcase] Torch-RecHub-Scala3: Full 1:1 industrial recommendation framework ported based on JavaCPP-PyTorch & LibTorch

[mullerhai]

Hi @saudet and JavaCPP-PyTorch maintainers,

 We’re glad to announce the official completion of Torch-RecHub-Scala3, a fully industrial-grade recommendation framework built entirely on top of JavaCPP-PyTorch and compiled native LibTorch binaries for JVM platform.

Project Info

Original Python upstream repo: https://github.com/datawhalechina/torch-rechub (well-known open-source CTR/CVR recommendation framework for search & advertising business)
Our Scala3 full-port repo: https://github.com/mullerhe/torch-rechub-scala

Core Achievement
We finished a complete 1:1 API reproduction of original Torch-RecHub after half a year of development. Thanks to mature C++ LibTorch binding provided by JavaCPP-PyTorch, the core recommendation algorithm stack for internet search, recommendation and promotion business is now fully available on JVM ecosystem, opening a brand-new era for industrial RecSys deployment on Java/Scala.
The entire framework supports cross-JVM languages out of box:
Java / Scala3 / Kotlin / Clojure / Groovy

All core training & inference APIs from original PyTorch version are fully aligned without functional missing.

Current Status

There remain minor incomplete details for further polishing, but we’ve accomplished the core implementation from scratch. This long-expected JVM-native recommendation runtime finally lands with the help of JavaCPP-PyTorch’s reliable PyTorch binding capability.
We’d like to express sincere gratitude to:
Original Torch-RecHub contributors for the outstanding Python recommendation baseline implementation
JavaCPP & JavaCPP-PyTorch team for enabling JVM to access native PyTorch/LibTorch ecosystem, which makes this cross-language port possible
Invitation
Welcome all community developers to clone, test and use our Torch-RecHub-Scala3 project:
👉 https://github.com/mullerhe/torch-rechub-scala
We’re also willing to share our port experience and feedback any JavaCPP-PyTorch related issues we found during development if needed.

I want to donate this project to bytedeco ！
Best regards,