datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
think-32k
think-32k
A general reasoning dataset from this collection.
List of categories:
general_qa
code
science_qa
math
creative_writing
brainstorming
summarization
information_extraction
classification
I was definitely sane when I wrote this config:
Creative-Writing-Thinking
Creative-Writing-Thinking
Using essays-creative-writing-prompts and Qwen3-14b to generate the reasoning traces and answers. We created this reasoning dataset.
Suitable for LLM post-training, especially RL.
think-20k
think-20k
A general reasoning dataset from this collection.
List of categories:
general_qa
code
science_qa
math
creative_writing
brainstorming
summarization
information_extraction
classification
think-10k
think-10k
A dataset with extract rows the dataset in this collection.
List of categories:
general_qa
code
science_qa
math
creative_writing
brainstorming
summarization
information_extraction
classification
Dataset structure
main/train.csv -- the full 10k training datasft/train.csv -- 2k rows for SFT warmup before RLrl/train.csv -- 8k forws for RL
brainstorming-thinking
brainstorming-thinking
Using Qwen3-14b to synthetically generate the reasoning traces and answers for Explore_Instruct_Brainstorming_10k
Suitable for LLM post-training, especially RL.
code-thinking
code-thinking
Using mbpp and using Qwen3-14b to generate the reasoning traces for the datasets.
Suitable for training small LLMs for python code generation.
Aurora-Think-1.0