Maen Suleiman

Supervised fine-tuning teaches a model to follow instructions, and reinforcement learning (RL) teaches it to reason. In this chalk talk, explore what RLHF and GRPO add beyond fine-tuning and how to run them on AWS Trainium, purpose-built for price performance.

The Neuron software stack has evolved into an open source-first architecture. Developers can read, modify, and contribute to the components that compile, optimize, and execute ML workloads on Trainium.

Running distributed AI workloads in production requires solving three problems: package compatibility, hardware abstraction, and network configuration. AWS Neuron Deep Learning Containers (DLCs) address all three by providing open-source, production-ready images for Trainium and Inferentia. This...

As PyTorch-based LLM serving matures, the challenge shifts from monolithic inference stacks to integrating diverse hardware accelerators efficiently. This session explores how modular plugin architectures enable PyTorch models to run optimally across backends—demonstrating AWS Trainium integration...

Dive deep into AWS Trainium's Neuron Kernel Interface (NKI) and unlock its potential for high-performance machine learning (ML). This session explores hardware foundations, kernel basics, and advanced techniques for optimizing ML workloads.