From Java VMs and GPU Acceleration to Motorcycle Electronics

An airhacks.fm conversation with Christos Kotselidis (@CKotselidis (twitter.com/CKotselidis)) about:
early experiences with computers and programming,
transition to studying Java (java.com/en) and virtual machines at university,
work on en.wikipedia.org/wiki/Jikes compiler and distributed software transactional memory for PhD,
current roles as professor at University of Manchester and working on motorcycle electronics at KTM,
overview of tornadovm.org project for accelerating Java on GPUs and other hardware,
discussion of recent Java implementations of LLMs like jlama (github.com/tjake/Jlama) and llama3 java (github.com/mukel/llama3.java),
potential for TornadoVM to accelerate model inference,
challenges around quantized types for large models,
integration with Project Panama (openjdk.java.net/projects/panama) for improved native interop,
importance of performance and energy efficiency for enterprise Java applications,
potential for Java Flight Recorder (docs.oracle.com/javacomponents/jmc-5-4/jfr-run…) to provide power consumption metrics,
need for standardized quantized types in Java,
opportunities for Java in AI/ML workloads,
invitation for companies to reach out about using Tornado VM for their use cases

Christos Kotselidis on twitter: @CKotselidis (twitter.com/CKotselidis)