|
--- |
|
license: apache-2.0 |
|
task_categories: |
|
- sentence-similarity |
|
pretty_name: Wikipedia UForm Embeddings for Nearest Neighbors Search |
|
size_categories: |
|
- 1M<n<10M |
|
--- |
|
|
|
## Dataset Summary |
|
|
|
This dataset contains 256-dimensional vectors for a 1M sample of Wikipedia for Approximate Nearest Neighbors Search benchmarks. |
|
|
|
### Usage |
|
|
|
``` |
|
git lfs install |
|
git clone https://huggingface.co/datasets/unum-cloud/ann-wiki-1m |
|
``` |
|
|
|
### Dataset Structure |
|
|
|
The dataset contains three matrices: |
|
|
|
- base: `base.1M.fbin` with 1M vectors to construct the index. |
|
- query: `query.public.100K.fbin` with 100K vectors to lookup in the index. |
|
- truth: `groundtruth.public.100K.ibin` with 10x results for every one of the 100K queries. |
|
|
|
Use the [ashvardanian/read_matrix.py](https://gist.github.com/ashvardanian/301b0614252941ac8a3137ac72a18892) Gist to parse the files. |
|
|
