Dataset · allenai
Tulu 3 SFT Mixture for instructions
allenai/tulu-3-sft-mixture
939,344 examples from the dataset mixture used for SFT training of the Tulu 3 model family.
What's inside
A mixture of text SFT examples used to train the Tulu 3 series, totaling 939,344 samples. It includes CoCoNot, FLAN v2, and other datasets; subset licenses vary, and some data is non-commercial. Languages listed in the metadata include amh, arb, ary, ars, acq, arz, and others.
What you can do with it
- SFT training of instruction-following LLMs
- Evaluating instruction following
- Reproducing or benchmarking against Tulu 3
- Tasks
- Other
- Languages
- Amharic, Arabic, Moroccan Arabic, Najdi Arabic, acq, Egyptian Arabic, apc, Bangla and 4 more
- Size
- 100K–1M rows
- Downloads this month
- 31.3K
- Likes
- 286
Models trained on it
- dphn/Dolphin3.0-Llama3.1-8B-GGUF 61.4K
- bartowski/cognitivecomputations_Dolphin3.0-Mistral-24B-GGUFText generation 48.7K
- bartowski/Dolphin3.0-Llama3.1-8B-GGUFText generation 28.8K
- bartowski/Dolphin3.0-Llama3.2-3B-GGUFText generation 20.5K
- bartowski/cognitivecomputations_Dolphin3.0-R1-Mistral-24B-GGUFText generation 10.8K
- allenai/Llama-3.1-Tulu-3-8B-SFTText generation 9.1K
- mradermacher/Qwen3.5-2B-MultiDistill-Megamix-6xMerge-UNCENSORED-heretic-i1-GGUFImage text → text 9K
- bartowski/Dolphin3.0-Qwen2.5-3b-GGUFText generation 8.7K
How to load it
from datasets import load_dataset
ds = load_dataset("allenai/tulu-3-sft-mixture")