CalvinXKY/InfraTech

What it solves

InfraTech provides a structured learning path and practical code examples for understanding the complex infrastructure behind AI models. It bridges the gap between high-level model usage and the low-level systems engineering required to train and run large-scale AI models efficiently.

How it works

The project is organized as a curated collection of Jupyter notebooks and technical documentation. It covers core concepts like distributed training, inference optimization, and hardware acceleration. Users can explore specific implementations—such as vLLM scheduling, SGLang memory management, or custom PyTorch training loops—to see how these systems function under the hood.

Who it’s for

This repository is designed for developers, researchers, and AI engineers who want to specialize in AI Infrastructure. It is ideal for those looking to master the mechanics of model deployment, performance tuning, and the internal architecture of popular frameworks.

Highlights

• Practical Jupyter notebooks for hands-on learning of inference and training internals.
• Deep dives into popular frameworks like vLLM and SGLang, including custom scheduler and memory management implementations.
• Comprehensive coverage of performance optimization techniques, such as parallel strategies (TP/PP/SP/EP), quantization, and speculative decoding.
• Explanations of distributed training foundations, including collective communication and reinforcement learning (RL) training-inference co-location.
• Curated technical articles and documentation on hardware-aware AI engineering.