Posts

A simple mathematical derivation of the algorithm construction process from the paper “Fast Inference from Transformers via Speculative Decoding”.

Diffusion models are the de facto standard for image generation. Lilian Weng’s “What Are Diffusion Models?” is an excellent introduction to it, but readers without a solid mathematical background may struggle. This article fills that gap with clear, step‑by‑step derivations and explanations.

Python—a language well-suited for fast prototyping in the AI field—has become increasingly popular these days. Creating a well-designed Python package at the right time can make your work more impactful and more likely to be adopted. Since October 2024, I have been learning about Python package development. Since then, I have published my own work IAAR-Shanghai/UHGEval as a Python package, participated in the early architecture design of MemTensor/MemOS , and deeply studied and contributed to gaogaotiantian/dowhen . These experiences have given me some insights into Python package design, which I document in this post. ...

I used to blog for a while, but mostly just study notes. Now I plan to start blogging again, preferring to write in-depth long articles, especially in research and life experiences. The future of research won’t belong to the peer-review system—which is already crumbling, in my view. Plus, blogging has become an integral part of academic discourse; for instance, ICLR introduced blog post tracks in 2021 1. So, I hope to create cool stuff, write fascinating content, and hopefully catch your attention through blogging. 🌟 ...

在慕课网学习 Scrapy 时所作的笔记，后来不断发展，内容更丰富，不仅限于 Scrapy 的内容。