Original link: GPULlama3.java Brings GPU-Accelerated LLM Inference to Pure Java / InfoQ.
You take LLama3 (an LLM), JLama (a Java API for this), GraalVM (AOT compliation for Java) and TornadoVM (GPU calculation support for Java) and get GPU accellerated LLM usage for Java and other JVM languages.
Interesting, but I don't know how much demand there is for such a solution.