Jifeng Wu's Personal Website

Memory models govern shared-memory visibility and ordering; forward-progress guarantees govern whether concurrent work can continue executing.

Study notes on OpenMP shared-memory parallelism, data scoping, tasks, and GPU offloading.

Dorm-friendly adaptations of Three Sisters stew, corn soup, strawberry drink, toasted seeds, and staghorn sumac tea.

A concise introduction to Lean's Type and Prop through Python's object–class–metaclass relationship.

A warp is a bundle of 32 threads that a GPU executes in lockstep under a single instruction. This post explains what a warp is, where the name comes from, how thread IDs and private registers let one

A personal act for the lifelong preservation of physical vigor, mental vitality, human connection, and the capacity for renewal.

This post introduces Nadaraya-Watson kernel regression from a localized squared-error objective, then connects it to the normalized, similarity-weighted averaging performed by attention.

llama.cpp supports 90+ model architectures through its convert_hf_to_gguf.py conversion script. The architecture is auto-detected from the model’s config.json. Run python convert_hf_to_gguf.py --print

2026-08 Financial Statement

Compile NEFF Executables from PyTorch Code