AI Potluck
Model components / Training & synthetic datasets

OpenAssistant Conversations (OASST1)

OpenAssistant

Crowdsourced human dialogue corpus from the 2023 OpenAssistant project, holding about 161,000 messages across 35 languages contributed by the LAION community. It remains a staple human-data ingredient in open post-training mixes, from Guanaco through Tulu 3 to OLMo 3's Dolci.

Verified 2026-08-13 via the OpenAssistant/oasst1 dataset card on Hugging Face.

Openness

5 high confidence
5.0
license
Apache-2.0
gated
false
dataset_card
present

Apache-2.0, ungated, full dataset card.

Adoption

3 high confidence
3.0

20,967 downloads in the trailing 30 days for OpenAssistant/oasst1, which bands at 10K-100K, level 3 on the dataset adoption scale.

Capability

3 high confidence
3.0

Proven, with ongoing downstream use (Guanaco, Tulu 3, OLMo 3 Dolci), but dated, reduced to a fractional ingredient, and with no comparative results isolating its value. The QLoRA paper finetunes Guanaco 65B on a variant of OASST1 - 9,209 examples, taking the top reply at each level of the conversation tree - and describes the result as the best-performing open-source chatbot model of its time.

  • https://arxiv.org/pdf/2305.14314 recorded 2026-08-14

    QLoRA paper full text, where OASST1 occurs 19 times. "the top QLoRA tuned model, Guanaco 65B, which we finetune on a variant of OASST1, is the best-performing open-source chatbot model and offers performance competitive to ChatGPT"; the dataset appendix gives the variant as 9,209 examples - "we only use the top reply at each level in the conversation tree. This limits the dataset to 9,209 examples." The arXiv abstract page carries none of this.

  • https://huggingface.co/datasets/allenai/Dolci-Instruct-SFT recorded 2026-07-04

    OpenAssistant Guanaco appears as a small fraction of OLMo 3 Dolci SFT

Verified 2026-08-13