SKU SFT-2003 · Sold by External
SmolTalk
Product specifications
| SKU | SFT-2003 |
|---|---|
| Data type | SFT dataset |
| Volume | 1043917 rows |
| Size on disk | 4.15 GB |
| Format | Parquet (Hugging Face) |
| Access model | PUBLIC LICENSE |
| Pricing | Free · open-source license |
| Quality score | — |
| License | Apache-2.0 |
An open synthetic instruction-tuning mixture built by Hugging Face's TB team to train SmolLM2-Instruct. Its 'all' config combines newly generated subsets — Smol-Magpie-Ultra (~400k), Smol-Constraints, Smol-Rewrite and Smol-Summarize — with curated public data such as OpenHermes-2.5, MetaMathQA, NuminaMath-CoT, Self-OSS-Instruct (code), system-chats and a function-calling subset. Rows are chat conversations with a role/content message list plus a source tag.