GV

GeezerVison

The World Wide Web, restored to 800 pixels of glory.

Home|About|Gaming cabinet|Archives|How it works|Starter sites

On this site

  • Home
  • Research
  • Publications
  • Open Source
  • Updates
  • Scholar
  • GitHub
  • X

This view

  • Original page
  • New address

GeezerVison is showing sinclaircoder.github.io — Zengzhi Wang | GAIR Lab, SJTU

Zengzhi Wang 王增志

PhD student at GAIR Lab, Shanghai Jiao Tong University.
Advised by Prof. Pengfei Liu.

I work on the engineering and science of data for foundation language models. Specifically:

  • Data Science: building scaling ladders, evaluating how data recipes scale, and designing scalable data recipes.
  • Data Engineering: parsing, refining, auditing, organizing, and synthesizing data.

Currently, I am a research intern on an industry pre-training team.

Always happy to connect—feel free to reach out.

Zengzhi Wang smiling in an ornate interior in the UAE.
We should dream big.
  • OctoThinker

    An early systematic study of why mid-training matters for scalable reinforcement learning.

    Connects data quality and format to RL performance, with a two-stage recipe that prepares base models for stronger downstream RL.

    ↗
  • MegaMath

    The largest open math pre-training corpus at release.

    Used in NVIDIA Nemotron 3 Nano, AMD Instella-MoE, Apertus, and SmolLM3.

    ~968K total downloads on Hugging Face as of Sep 5, 2026
    ↗
  • ProX

    Making expert-style data refinement scalable with small language models.

    Generates and executes edits tailored to each training example to improve pre-training efficiency.

    ↗

Build broad knowledge Connect diverse knowledge units into a general representation.

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.