MeetingBank-LLMCompressed is a dataset released by Microsoft in 2024 for building training data for the LLMLingua-2 compressor, with the associated paper titled "LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression", aimed at providing efficient and faithful task-agnostic prompt compression training samples.

The dataset contains 5,169 samples from the MeetingBank training set, each including the original meeting transcript text and its version compressed independently by GPT-4 in chunks, with a data size of approximately 246 MB.

Dataset Composition

The dataset includes the following 6 fields:

1. idx: The index number of the instance
2. prompt: The original text of the meeting transcript
3. prompt_list: The list of original text chunks corresponding to the prompt
4. compressed_prompt_list: The list of texts after each chunk is independently compressed by GPT-4
5. compressed_prompt: The complete compressed text formed by concatenating all compressed chunks
6. summary: The meeting transcript summary from MeetingBank

Citation

@inproceedings{pan2024llmlingua2,
    title={LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression},
    author={Zhuoshi Pan and Qianhui Wu and Huiqiang Jiang and Menglin Xia and Xufang Luo and Jue Zhang and Qingwei Lin and Victor Rühle and Yuqing Yang and Chin-Yew Lin and H. Vicky Zhao and Lili Qiu and Dongmei Zhang},
    year={2024},
    booktitle = {Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics},
    publisher = {Association for Computational Linguistics}
}