Oct 10, 2026

From 15% to 90% GPU Utilization: Fix the Data Pipeline, Not the Model

Tarun Sunkaraneni from Amazon AGI shows how a multimodal SFT pipeline that spent about 85% of its time waiting on data was brought to roughly 90% GPU utilization. He walks through concurrency with asyncio and Ray actors, prefetching, zero-copy transfer via Ray's object store, and a network-card bottleneck that appears only at scale, and argues that the hardest part of training throughput is often keeping GPUs fed rather than optimizing the model.

Oct 08, 2025

Build a Local LLM App in Python with Just 2 Lines of Code

Distinguished Engineer Chris Hay demonstrates how to run and program Large Language Models (LLMs) locally in just two lines of Python code. The tutorial covers setting up a local environment with Ollama and UV, using a custom library for simplified interaction, and explores advanced topics like asynchronous streaming, persona customization with system prompts, and managing multi-turn conversations.