6/22/2026
Fusing a whole model into one CUDA kernel: graph building, task lowering, scheduling, and lessons from leaning too much on AI agents.
4/21/2026
Memory optimization for distributed training. The three ZeRO stages explained with math and notation.
4/6/2026
Math behind the Retention mechanism, recurrent, parallel, and chunkwise forms from the RetNet paper.
4/5/2026
Control plane, nodes, pods, services, kubectl, minikube, and the reconciliation loop explained.
4/5/2026
Complete UV cheatsheet, the fast Python package manager that replaces pip, virtualenv, and poetry.