Free🇺🇸
Hugging Face pretraining dataset of 1.3 trillion tokens of educational web pages, selected from Common Crawl by an AI quality classifier.
Tulu 3 SFT Mixture overview and key facts (October 2026)
Instruction-tuning mix of 939,344 examples from the Allen Institute for AI, used to train the Tulu 3 models.
Free
What is Tulu 3 SFT Mixture?
The Tulu 3 SFT Mixture is an open collection of instruction and chat examples for supervised fine-tuning. It shows how the Allen Institute for AI turned base models into its Tulu 3 assistants.
The mixture combines 18 sources, among them FLAN v2, WildChat, OpenAssistant, NuminaMath, Aya, SciRIFF, table data and several synthetic persona datasets written for Tulu 3. Each example has an id, a list of messages and its source. The Allen Institute for AI in Seattle publishes it as a research artifact under ODC-By 1.0, while each subset keeps its own license. It was used to train Tulu 3 on Llama 3.1 8B and 70B.
Tulu 3 SFT Mixture AI features and use cases
- Fine-tunes base models into assistantsTeams with their own base model can use the mixture for supervised fine-tuning so the model follows instructions and answers in chat format. It is a documented starting point instead of collecting examples from scratch.
- Adds safety behaviorSubsets such as CoCoNot, WildGuardMix and WildJailbreak teach a model to refuse harmful or inappropriate requests. This helps when a fine-tuned model will face employees or customers. Test the behavior with your own prompts as well.
- Serves as a template for own dataThe source field shows how each skill was covered. Data teams can replace or extend subsets with their own company examples and keep the rest of the mix. The mixture then becomes a documented base for your own training data.
What sets Tulu 3 SFT Mixture apart
- Each example records its source dataset, so subsets with unwanted licenses can be removed.
- Documented as the exact data used for the published Tulu 3 models.
Where Tulu 3 SFT Mixture is strongest
United States Available internationally
The Allen Institute for AI is based in Seattle; the examples are mostly in English, with 100,000 prompts from the multilingual Aya dataset.
Questions to ask before you buy Tulu 3 SFT Mixture
Some subsets, such as No Robots, are licensed for non-commercial use only; remove them before commercial training. WildChat contains real user conversations, so review data protection for your use. The mixture is mainly English, so German performance needs additional data.
Tulu 3 SFT Mixture reviews
No reviews yet. Be the first to share how your team uses it.
Every review shows its verification level. We read each one before it goes live. How we check reviews
Questions about Tulu 3 SFT Mixture
Can we use the mixture commercially?
The collection is under ODC-By 1.0, but subsets keep their own licenses and some are non-commercial. Filter by the source field before commercial use.
How large is it?
It contains 939,344 examples in chat format, each with an id, messages and source.
Which models were trained with it?
The Allen Institute for AI used it to train the Tulu 3 series on Llama 3.1 8B and 70B; the models are listed in a Hugging Face collection.
Tulu 3 SFT Mixture update historyno updates yet
No changes recorded for Tulu 3 SFT Mixture yet.
Changes the vendor made and confirmed user reports, each checked against the vendor's own pages. We review every tool monthly.
Alternatives to Tulu 3 SFT Mixture
Tools in the same category (Training data) with the most overlapping markets.
Free🇺🇸
Multilingual pretraining dataset from Hugging Face covering more than 1,000 languages, including a large German subset.
Free🇫🇷
Open pretraining corpus of 2.27 trillion tokens from Pleias in Paris, containing only uncopyrighted or freely licensed documents.
Contact for pricing · Free planAvailable internationally
Free, open-source prompt library, formerly Awesome ChatGPT Prompts, available as a website, MCP server or self-hosted company library.