cryptolab-playground/gas-centroids

GAS Indexing Artifacts

Dataset Description

This dataset contains pre-computed deterministic centroids and associated geometric metadata generated using our GAS (Geometry-Aware Selection) algorithm. These artifacts are designed to benchmark Approximate Nearest Neighbor (ANN) search performance in privacy-preserving or dynamic vector database environments.

Purpose

To serve as a standardized benchmark resource for evaluating the efficiency and recall of vector databases implementing the GAS architecture. It is specifically designed for integration with VectorDBBench.

Dataset Summary

• Source Data: Wikipedia (Public Dataset)
• Embedding Model: google/embeddinggemma-300m

Dataset Structure

For each embedding model, the directory contains two key file:

Data	Description
centroids.npy	centroids as followed IVF
tree_info.pkl	tree metadata with parent and leaf info

Data Fields

Centroids: centroids.npy

• Purpose: Finding the nearest clusters for IVF (Inverted File Index)
• Type: NumPy array (np.ndarray)
• Shape: [32768, 768]
• Description: 768-dimensional vectors representing 32,768 cluster centroids
• Normalization: L2-normalized (unit norm)
• Format: float32

Tree Metadata: tree_info.pkl

• Purpose: Finding virtual clusters following hierarchical tree structure for efficient GAS search
• Type: Python dictionary (pickle)
• Keys:

  • node_parents: Dictionary mapping each node ID to its parent node ID

    • Format: {node_id: parent_node_id, ...}
    • Contains parent-child relationships for all nodes in the tree

  • leaf_ids: List of leaf node IDs

    • Format: [leaf_id_1, leaf_id_2, ..., leaf_id_32768]
    • Total 32,768 leaf nodes (corresponding to 32,768 centroids)

  • leaf_to_centroid_idx: Mapping from leaf node IDs to centroid indices in centroids.npy

    • Format: {leaf_node_id: centroid_index, ...}
    • Maps each leaf node to its corresponding row index in centroids.npy
    • Important: Leaf IDs in leaf_ids are ordered sequentially, so the i-th leaf corresponds to the i-th centroid

Dataset Creation

Source Data

Source dataset is a large public dataset, Wikipedia: mixedbread-ai/wikipedia-data-en-2023-11.

Preprocessing

1. Create Centroids by GAS approach:

   Description TBD

2. Chunking: For texts exceeding 2048 tokens:

   • Split into chunks with ~100 token overlap
   • Embedded each chunk separately
   • Averaged chunk embeddings for final representation

3. Normalization: All embeddings are L2-normalized

Embedding Generation

• Model: google/embeddinggemma-300m
• Dimension: 768
• Max Token Length: 2048
• Normalization: L2-normalized

Usage

import wget

def download_centroids(embedding_model: str, dataset_dir: str) -> None:
    """Download pre-computed centroids and tree info for GAS."""
    dataset_link = "https://huggingface.co/datasets/cryptolab-playground/gas-centroids/resolve/main/embeddinggemma-300m"
    wget.download(f"{dataset_link}/centroids.npy", out="centroids.npy")
    wget.download(f"{dataset_link}/tree_info.pkl", out="tree_info.pkl")

License

Apache 2.0

Citation

If you use this dataset, please cite:

@dataset{gas-centroids,
  author = {CryptoLab, Inc.},
  title = {GAS Centroids},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/cryptolab-playground/gas-centroids}
}

Source Dataset Citation

@dataset{wikipedia_data_en_2023_11,
  author = {mixedbread-ai},
  title = {Wikipedia Data EN 2023 11},
  year = {2023},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/mixedbread-ai/wikipedia-data-en-2023-11}
}

Embedding Model Citation

@misc{embeddinggemma,
  title={Embedding Gemma},
  author={Google},
  year={2024},
  url={https://huggingface.co/google/embeddinggemma-300m}
}

Acknowledgments

• Original dataset: mixedbread-ai/wikipedia-data-en-2023-11
• Embedding model: google/embeddinggemma-300m
• Benchmark framework: VectorDBBench