Dataset · baize
Baize self-chat data
Builder: UC San Diego / Sun Yat-sen University (Baize authors)
Availability: available · checked 2026-09-24 · source
Data lives in the project's GitHub repo, not on Hugging Face; the repo data folder returned HTTP 200. No HF dataset repo id was confirmed (project-baize/baize-chatbot is not a dataset).
Raw record: /data/datasets/baize.json
Fields
- id
- baize
- identifiers
- builder
- UC San Diego / Sun Yat-sen University (Baize authors)
- release_date
- 2023-04-03partial · sourcenote: arXiv v1 date of the Baize paper.
- availability
- available · checked 2026-09-24 · sourcenote: Data lives in the project's GitHub repo, not on Hugging Face; the repo data folder returned HTTP 200. No HF dataset repo id was confirmed (project-baize/baize-chatbot is not a dataset).
- content
- Paper: dialogues produced by having ChatGPT chat with itself ('self-chat') from seed questions from Quora and Stack Overflow; 'we collect a total of 111.5k dialogues through self-chat, using ∼55k questions' for the v1 family. Repo README lists 54K/57K/47K dialogs from Quora, StackOverflow and MedQuAD questions. Falcon-7B/40B-instruct's card lists Baize as part of their fine-tuning mix.recorded · sourcenote: Counts are the paper's and the README's; the released files were not counted.
- primary_sources
- record_history
- date:2026-09-24 · change:created from primary sources (tranche 2 dataset records) · by:wilson-pruitt + claude ·
Parents
Influence without weights
- distilled_from_outputs → ChatGPT (Nov. 2022 launch model) declared source Paper: 'leveraging ChatGPT (gpt-3.5-turbo)' for self-chat; snapshot not_recorded.
Children
Training data
- ← trained_on Baize v2 7B declared source Card: trained on self-chat data (SFT + SDF); paper: 'leveraging ChatGPT to engage in a conversation with itself'. The `baize` dataset record describes the v1 corpus; the README ships separate v1 and v2 collection scripts (collect.py, collect_v2.py), so v2's exact dialogues are not pinned down. Reviewer: confirm the family-level edge is acceptable.
Read in
No station on the reading path has touched this record yet.