On June 3–4, 2026, the TornadoVM team from UNIMAN participated in jPrime 2026 in Sofia, Bulgaria.
The presentation showcased the latest developments in TornadoVM and GPULlama3, demonstrating seamless GPU acceleration for local AI inference in Java applications using Quarkus and LangChain4j.
The talk highlighted TornadoVM’s ability to transparently offload compute-intensive workloads to GPUs, enabling Java developers to leverage hardware acceleration without requiring GPU programming expertise.
The work aligns with the objectives of the AERO project, particularly in advancing hardware acceleration technologies for the European Processor Initiative ecosystem. Live demonstrations featured GPULlama3 integrated with Quarkus and LangChain4j, showcasing accelerated local LLM inference and end-to-end AI workflows running on GPU-accelerated Java applications.
