636: Red Hat's James Haung

Links

James on LinkedIn (https://www.linkedin.com/in/jahuang/)

Mike on LinkedIn (https://www.linkedin.com/in/dominucco/)

Mike's Blog (https://dominickm.com)

Show on Discord (https://discord.com/invite/k8e7gKUpEp)

Alice Promo (https://go.alice.dev/data-migration-offer-hands-on)

• AI on Red Hat Enterprise Linux (RHEL)

Trust and Stability: RHEL provides the mission-critical foundation needed for workloads where security and reliability cannot be compromised.

Predictive vs. Generative: Acknowledging the hype of GenAI while maintaining support for traditional machine learning algorithms.

Determinism: The challenge of bringing consistency and security to emerging AI technologies in production environments.

• Rama-Llama & Containerization

Developer Simplicity: Rama-Llama helps developers run local LLMs easily without being "locked in" to specific engines; it supports Podman, Docker, and various inference engines like Llama.cpp and Whisper.cpp.

Production Path: The tool is designed to "fade away" after helping package the model and stack into a container that can be deployed directly to Kubernetes.

Behind the Firewall: Addressing the needs of industries (like aircraft maintenance) that require AI to stay strictly on-premises.

• Enterprise AI Infrastructure

Red Hat AI: A commercial product offering tools for model customization, including pre-training, fine-tuning, and RAG (Retrieval-Augmented Generation).

Inference Engines: James highlights the difference between Llama.cpp (for smaller/edge hardware) and vLLM, which has become the enterprise standard for multi-GPU data center inferencing.