metadata
dataset_info:
splits:
- name: biomed
- name: law
- name: finance
features:
- name: text
type: string
license: apache-2.0
task_categories:
- text-generation
language:
- en
Dataset Description
This dataset contains the test splits used to evaluate the Memory Decoder model across three specialized domains: biomedical, legal, and finance. The test data was randomly sampled from publicly available datasets to assess the model's performance in domain-specific language understanding.
GitHub: https://github.com/LUMIA-Group/MemoryDecoder
Dataset Sources
The test data is sampled randomly from the following source datasets:
Biomedical Domain
- Source: Medilora/mimic_iii_diagnosis_anonymous
- Description: Anonymized clinical discharge summaries from the MIMIC-III database containing medical diagnoses, patient histories, and treatment procedures
Legal Domain
- Source: clairebarale/AsyLex
- Description: Refugee status determination documents from Canada (1996-2022) containing legal judgments and case analyses
- Paper Reference: ACL 2023 Findings
Finance Domain
- Source: jyanimaulik/yahoo_finance_stockmarket_news
- Description: Stock market news articles from Yahoo Finance covering company earnings, market analysis, and financial commentary
Dataset Structure
Data Splits
The dataset consists of test splits only, randomly sampled from the source datasets:
├── biomed/ # Biomedical domain test split
├── law/ # Legal domain test split
└── finance/ # Finance domain test split
Citation
@article{cao2025memory,
title={Memory decoder: A pretrained, plug-and-play memory for large language models},
author={Cao, Jiaqi and Wang, Jiarui and Wei, Rubin and Guo, Qipeng and Chen, Kai and Zhou, Bowen and Lin, Zhouhan},
journal={arXiv preprint arXiv:2508.09874},
year={2025}
}
Contact
For questions and support: maximus.cao@outlook.com