Ml-Infrastructure

Building a Pure-Python GGUF Weight Loader with Dequantization for Llama Models

A hands-on guide to building a pure-Python GGUF weight loader with full dequantization support for Llama models — covering binary parsing, quantized tensor reconstruction, and a roadmap to production-grade infrastructure.

Diagram of a transformer training loop showing activations dropped at checkpoint boundaries and recomputed during backward pass.

Building a Gradient Checkpointing Memory Scheduler From Scratch: A Portfolio Project That Actually Signals ML Systems Skill

A practical build guide for a portfolio-grade gradient checkpointing scheduler that trades compute for memory in mini GPT training. Includes runnable PyTorch code, a scheduler design, tests, and a roadmap to production-flavored upgrades.

Abstract diagram of GPU scheduling queues feeding a rack of accelerators.

GPU Scheduling at Scale: How Modern Systems Stop Fighting Over Silicon

GPUs are no longer exotic, but scheduling them well is still hard. This post walks through the policies, primitives, and production patterns that decide who gets the silicon.

Stylized dashboard showing GPU utilization bars alongside cost and revenue metrics.

GPU Utilization Is Becoming a Business Metric

GPUs are no longer a hidden infrastructure cost. They are a tracked business metric, tied to revenue, margins, and team accountability. This post unpacks why utilization moved from engineering dashboards to board decks, and what it means for the way you build, schedule, and bill ML workloads.