PLAID

PLAID index. The PLAID index is the most scalable type of index for multi-vector search and leverage PQ-IVF as well as custom kernel for decompression.

Parameters

• index_folder ('str') – defaults to indexes

  The folder where the index will be stored.

• index_name ('str') – defaults to colbert

  The name of the index.

• override ('bool') – defaults to False

  Whether to override the collection if it already exists.

• embedding_size ('int') – defaults to 128

  The number of dimensions of the embeddings.

• nbits ('int') – defaults to 2

  The number of bits to use for the quantization.

• nranks ('int') – defaults to 1

• kmeans_niters ('int') – defaults to 4

  The number of iterations to use for the k-means clustering.

• index_bsize ('int') – defaults to 1

• ndocs ('int') – defaults to 8192

  The number of candidate documents

• centroid_score_threshold ('float') – defaults to 0.35

  The threshold scores for centroid pruning.

• ncells ('int') – defaults to 8

  The number of cells to consider for search.

• search_batch_size ('int') – defaults to 262144

  The batch size to use when searching.

Methods

call

Query the index for the nearest neighbors of the queries embeddings.

Parameters

• queries_embeddings ('np.ndarray | torch.Tensor')
• k ('int') – defaults to 10

add_documents

Add documents to the index.

Parameters

• documents_ids ('str | list[str]')
• documents_embeddings ('list[np.ndarray | torch.Tensor]')
• batch_size ('int') – defaults to 2000

get_documents_embeddings

remove_documents

Remove documents from the index.

Parameters

• documents_ids ('list[str]')