SKU SFT-2001 · Sold by External
Tulu 3 SFT Mixture
Product specifications
| SKU | SFT-2001 |
|---|---|
| Data type | SFT dataset |
| Volume | 939344 rows |
| Size on disk | 1.41 GB |
| Format | Parquet (Hugging Face) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | ODC-BY-1.0 |
A single-split SFT mixture assembled by Allen AI to train the Tulu 3 model family. It combines public sources (FLAN v2, OpenAssistant, No Robots, CoCoNot) with newly generated synthetic data (WildChat GPT-4, Persona math/instruction-following/code, NuminaMath-TIR) and safety data (WildGuardMix, WildJailbreak). Each row is a chat-formatted conversation with a role/content message list and a source tag identifying its origin.