Dataset Overview

The CLaRa (Continuous Latent Reasoning) dataset is an open-source benchmark released by Apple in 2025, designed for retrieval-augmented generation and continuous latent reasoning. The related research paper can be found at [insert link or citation here].CLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning", aiming to provide training and evaluation data to support model performance in Question Answering under compressed document representations.

This dataset contains three major categories of data: pre-training, instruction fine-tuning, and end-to-end tuning. It covers construction and evaluation sets for various mainstream QA benchmarks, such as 2WikiMultihopQA, HotpotQA, MuSiQue, and Natural Questions. The dataset is large-scale (ranging from 10 GB to 100 GB), uniformly formatted in JSONL, supports both positive-document scenarios (Oracle) and standard retrieval settings, making it suitable for research on Retrieval-Augmented Generation (RAG), document compression, and complex QA tasks.

Dataset Composition

The dataset primarily consists of the following three parts, with all data stored in JSONL format:

• Pretraining Data: Pre-training data used to learn the document compressor. Contains fields:

  • data_type: Data type (e.g., qa)
  • question: List of questions
  • answers: List of answers
  • docs: List of document contexts

• Fine-Tuning Data: Instruction fine-tuning data used for QA training based on compressed document representations. Contains fields:

  • question: String containing the question
  • docs: List of retrieved documents
  • gold_answer: Ground truth answer
  • answer: Model-generated answer

Additionally, there are evaluation sets tailored for different benchmarks (2wiki, hotpotqa, musique, nq). These include an Oracle setting that includes positive documents (inst_eval*_with_pos) and a normal setting without them (inst_eval*_no_pos).

• End-to-End Tuning Data: Complete workflow data used for end-to-end model training. Contains fields:
  • question: String containing the question
  • answer: String containing the answer
  • docs: List of retrieved documents
  • pos_index: Index of the positive document within the docs list

These also consist of separate training and evaluation sets. For each benchmark test (ours_2wiki, ours_hotpotqa, ours_musique, ours_nq), two configurations are provided: one including positive documents (with_pos) and another excluding them (no_pos).

Citation```bibtex

@misc{he2025clarabridgingretrievalgeneration, title={CLaRa: Bridging Retrieval and Generation with Continuous Latent Reasoning}, author={Jie He and Richard He Bai and Sinead Williamson and Jeff Z. Pan and Navdeep Jaitly and Yizhe Zhang}, year={2025}, eprint={2511.18659}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2511.18659}, }