---
dataset_info:
  features:
  - name: sentence1
    dtype: string
  - name: sentence2
    dtype: string
  - name: labels
    dtype: int64
  - name: metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 486486046
    num_examples: 183053
  - name: validation
    num_bytes: 55616011
    num_examples: 21790
  - name: test
    num_bytes: 75316832
    num_examples: 28605
  download_size: 237057532
  dataset_size: 617418889
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: mit
source_datasets:
- napsternxg/wands
---

# wands

WANDS native three-grade product relevance with query-disjoint internal holdouts.

Original Wayfair data: https://github.com/wayfair/WANDS (MIT).
Uses the pinned joined napsternxg/wands Hub copy; its random row splits overlap
on queries and are replaced by deterministic 80/10/10 normalized-query groups.
All native judgments are retained. Inputs contain query and product content;
popularity/rating fields and relevance labels never enter model inputs.

Original data: [napsternxg/wands](https://huggingface.co/datasets/napsternxg/wands). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/](https://github.com/sileod/tasksource/tree/main/scripts/repackage_dataset).