What’s Out There
← Back to swiping
Language & textFreeCompany

PersonaHub

Tencent AI Lab (proj-persona)

Short persona descriptions that steer an LLM into generating varied synthetic data.

Released data from the paper on scaling synthetic data with a billion personas: personas distilled from web text, plus samples of math, instruction, and other synthetic data generated from them. One row is one persona or generated sample depending on subset. The Hub holds a subset of the paper's full set.

Access

Cost
FreeNo charge.
License
Open, non-commercial onlyCC BY-NC-SA 4.0
Sign-up
No sign-up
How you get it
API and Bulk downloadJSON Lines

What's in it

One row is
One row per persona description or generated sample
Covers
Not geographic
Years
2024
Updated
One-time release
Size
100,000 to 1 million rows on the Hub (a subset of the paper's collection)
persona descriptionsynthesized promptsubset typelanguage

Quality

3 of 5

A well-cited idea with released samples, but personas are LLM-made and the Hub release is a fraction of the paper's scale.

  • Non-commercial license
  • Personas are model-written, not real people
  • Released size is far below the billion in the title

Connects to

  • FinePersonas: Open set of 21 million personas using a similar approach
  • Nemotron-Personas-Korea: Personas grounded in real national statistics, contrasted with LLM-invented ones

Questions it could answer

  • How diverse are personas by occupation?
  • Do prompts from different personas actually differ?

Similar data sets

  • Nemotron-Personas-Korea

    Synthetic Korean personas whose ages, jobs, and regions follow official national statistics.

    DemographicsNVIDIAFree

  • Cosmopedia

    Synthetic textbooks, blog posts, and stories written by Mixtral to mimic what web pretraining text teaches.

    Language & textHugging Face (SmolLM team)Free

  • HH-RLHF (Helpful and Harmless)

    Human preference comparisons that helped define how chat assistants are trained to be helpful and harmless.

    Language & textAnthropicFree

  • Human Connectome Project Young Adult (HCP-YA)

    Hours of brain scans on 1,200 young adults, many of them twins, released to anyone who signs terms.

    Biology & genomicsWU-Minn HCP Consortium (Washington University in St. Louis and University of Minnesota)Free

  • UltraChat 200k

    A 200,000-dialogue filtered slice of ChatGPT-generated conversations that trained Zephyr-7B.

    Language & textHugging Face H4 team (filtered from Tsinghua's UltraChat)Free

  • WildChat-1M

    A million real ChatGPT conversations with state, country, and hashed-IP fields attached.

    Language & textAllen Institute for AI and University of WashingtonFree