datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
akeel-thought-injection-20k
Akeel Thought Injection Dataset (20K)
Single-turn reasoning traces for training thought injection models
A QRK Labs Research Dataset
Overview
This dataset contains 20,000 single-turn examples of thought injection reasoning traces. Each example demonstrates a model:
Receiving a question
Reasoning about what it knows (in <think> tags)
Requesting external knowledge when needed (<knowledge> tags)
Integrating retrieved context… See the full description on the dataset page: https://huggingface.co/datasets/qrk-labs/akeel-thought-injection-20k.akeel-thought-injection-50k-augmented
Akeel Thought Injection Dataset (50K Augmented)
Multi-turn augmented reasoning traces for robust thought injection training
A QRK Labs Research Dataset
Overview
This dataset contains 50,000 augmented examples designed to prevent overfitting when training thought injection models. It expands the base 20K dataset through:
Shuffled originals (17,500) — Base samples randomly reordered
System prompt variations (17,500) — Same Q&A with different… See the full description on the dataset page: https://huggingface.co/datasets/qrk-labs/akeel-thought-injection-50k-augmented.akeel-thought-injection-62k-with-refusals