Jifeng Wu's Personal Website

Local CUDA vLLM Setup for Python-Only Development Using a Precompiled Wheel

How to run vLLM directly from a local clone with editable Python code while using precompiled native libraries from a nightly wheel — no full source compilation required.

2026-05-19

AI and Machine Learning

#reference #systems #machine-learning #python #llm #linux #vllm

Compile NEFF Executables from NKI Kernels

Compile NEFF Executables from NKI Kernels

2026-05-14

Systems

#reference #systems #machine-learning #python #shell #neuron

Type Theory Concepts: A to Z

An introduction to type theory concepts using Python, C++, and layman's terms.

2026-05-13

Programming

#reference #programming-languages #type-theory

What is S3?

S3 is a giant distributed hashmap exposed as both a public HTTP server and an authenticated REST API.

2026-05-13

Systems

#reference #systems #software-engineering

vLLM Internals — PagedAttention and Custom Accelerator Compilation

How vLLM applies OS-style paging to the KV cache for efficient memory management, enables continuous batching and memory sharing, and routes computation graphs to custom compiler backends for new acce

2026-05-12

AI and Machine Learning

#reference #systems #machine-learning #llm #inference #pytorch #vllm

Our Cognitive Profile and a Personal Playbook for the Agentic Era

A personal analysis of the twice-exceptional neurocognitive profile — the intersection of high intelligence, executive dysfunction, and perfectionism — and a strategic playbook for thriving in the age

2026-05-10

Reflections

#llm #essay #career

Exporting Compute Graphs, LLM Shape Dynamics, and Serving Runtimes

Why CNNs export to one graph while LLMs need many, how Hugging Face hides this complexity through eager execution, and why vLLM-style systems act as specialized inference operating systems rather than

2026-05-03

AI and Machine Learning

#reference #systems #machine-learning #llm #inference #pytorch #vllm

Schedules in Machine Learning Computation: What They Are and Who Needs to Know About Them

A schedule is the plan for how a computation is executed efficiently on hardware. This post maps out the schedule-specification spectrum across ML frameworks, explains who needs to care about scheduli

2026-05-03

AI and Machine Learning

#reference #systems #machine-learning #computer-architecture

Important Locations on Jailbroken iOS

Important Locations on Jailbroken iOS

2026-04-26

Systems

#reference #systems #c++

Learning MLIR and HLO by Building a Tiny StableHLO-to-LLVM IR Compiler

This project explored how to lower machine-learning-oriented IR into low-level compiler IR by building a small MLIR-based compiler. The tool reads StableHLO MLIR, applies a sequence of lowering and cl

2026-04-26

Systems

#reference #systems #c++ #build-systems #machine-learning