SKU SFT-2406 · Sold by External
OpenAssistant Conversations v2 (OASST2)
Product specifications
| SKU | SFT-2406 |
|---|---|
| Data type | SFT dataset |
| Volume | 135,174 messages (128,575 train + 6,599 validation) |
| Size on disk | ~66.7 MB Parquet (download_size 66,674,129 bytes; in-memory dataset_size 170,771,665 bytes per datasets-server info; the HF card YAML reports dataset_size 166,813,577 bytes) |
| Format | Parquet |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open dataset (HF) |
| Quality score | — |
| License | Apache License 2.0 |
OASST2 is the second release of the OpenAssistant Conversations dataset, a crowdsourced collection of message trees where volunteers alternately wrote prompter and assistant turns and then reviewed, labeled, and ranked them. Each row is a single message with tree-structure fields (message_id, parent_id, message_tree_id), the message text, its role (prompter/assistant), language, and rich human annotations including quality/helpfulness labels, reviewer counts, per-sibling rank, and Detoxify toxicity scores. The Hugging Face release contains 135,174 messages split into a 128,575-message train split and a 6,599-message validation split, stored as Parquet. Messages span 35 language codes (per the HF card language list) and the data snapshot was finalized on 2023-11-05 and published on Hugging Face on 2023-12-24. It is distinct from the earlier oasst1 release, and is licensed under Apache 2.0.