Model components / Training & synthetic datasets

FLAN Collection (v2)

The FLAN Collection (v2) is Google Research's instruction-tuning compilation of 1,800+ tasks (P3, Super-NaturalInstructions, chain-of-thought data) that trained Flan-T5 and Flan-PaLM. It remains a canonical instruction-data building block, folded into Tulu 3, OLMo 3's Dolci, and Marin's Dolmino mixes.

Openness

4 medium confidence

4.0

access: ungated(conversion)
license: Apache-2.0(recipe)+per-task
dataset_card: partial

Released as an Apache-2.0 recipe/repo; the widely used HF conversion (ai2-adapt-dev) is ungated. Underlying task datasets keep their own licenses.

https://github.com/google-research/FLAN recorded 2026-07-04
Apache-2.0 FLAN v2 recipe
https://huggingface.co/datasets/ai2-adapt-dev/flan_v2_converted recorded 2026-07-04
ungated HF conversion used by Tulu/OLMo mixes

Adoption

1 high confidence

1.0

924 monthly downloads on Hugging Face, graded on the training-corpus bands.

https://huggingface.co/api/datasets/ai2-adapt-dev/flan_v2_converted recorded 2026-07-04
downloads field = 924 (30-day)

Capability

3 high confidence

3.0

Strong historical comparative ablations (Flan 2022 beats P3/T0/SNI on the same T5-XL) and still a component in Tulu 3, Dolci, and Marin's mixes, but dated as a standalone chat-era recipe.

https://arxiv.org/abs/2301.13688 recorded 2026-07-04
The Flan Collection: Flan 2022 leads P3++/Super-NatInst on MMLU/BBH (same T5-XL)
https://huggingface.co/datasets/allenai/Dolci-Instruct-SFT recorded 2026-07-04
OLMo 3 Dolci still consumes FLAN v2 (89,981 prompts)

Unchanged since 2026-07-04 (last edited, not re-checked)