README.md

metadata

dataset_info:
  features:
    - name: query
      dtype: string
    - name: positive
      dtype: string
    - name: negative
      dtype: string
  splits:
    - name: train
      num_bytes: 2766980301
      num_examples: 1391986
  download_size: 1589194354
  dataset_size: 2766980301
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Training Data For Paper: "Pooling And Attention: What Are Effective Designs For LLM-Based Embedding Models?"

Citation:

@misc{poolingattentioneffectivedesigns,
      title={Pooling And Attention: What Are Effective Designs For LLm-Based Embedding Models?}, 
      author={Yixuan Tang and Yi Yang},
      year={2024},
      eprint={2409.02727},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2409.02727}, 
}