Dataset · flan-v2

FLAN-v2 Collection (Flan 2022)

Builder: Longpre et al., github.com/google-research/FLAN
Availability: available · checked 2026-09-24 · source
HTTP 200 on the repo page. Only the repo page was checked, not each file. Third-party HF mirror Muennighoff/flan also exists.

Raw record: /data/datasets/flan-v2.json

Fields

id

flan-v2

builder

Longpre et al., github.com/google-research/FLAN

release_date

2023-01-31partial · source

note:  arXiv v1 date of the Flan Collection paper.

availability

available · checked 2026-09-24 · source

note:  HTTP 200 on the repo page. Only the repo page was checked, not each file. Third-party HF mirror Muennighoff/flan also exists.

content

Orca 2 paper: 'the FLAN-v2 Collection [33], which consists of five sub-collections, namely, CoT, NiV2, T0, Flan 2021 and Dialogue'; Orca 2 was first finetuned 'on the train split of FLAN-v2 dataset for one epoch'. Reference [33] in that paper is Longpre et al., 'The Flan Collection: Designing Data and Methods for Effective Instruction Tuning' (arXiv 2301.13688).recorded · source

note:  Recorded because orca-2-7b trained on it directly (Orca 2 paper section 4.2). Human-written task data, so no model parents recorded; the Flan Collection paper itself was not read in full.

primary_sources

https://arxiv.org/abs/2311.11045

https://arxiv.org/abs/2301.13688

https://github.com/google-research/FLAN

record_history

date:2026-09-24 · change:created from primary sources (needed by orca-2-7b trained_on edge; panel ruling 2/3 dataset routing) · by:wilson-pruitt + claude (Sonnet 5) ·

Parents

No edges recorded.

Children

Training data

• ← trained_on Orca 2 7B declared source Orca 2 paper 4.2: 'We start with LLaMA-2-7B or LLaMA-2-13B checkpoint and finetune it on the train split of FLAN-v2 dataset for one epoch.'

Read in

No station on the reading path has touched this record yet.