Ctrlk
BlogSlackSignup/Login
x-twitter
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Search & Retrieval

Use the advanced search and retrieval capabilities of Spice

Spice provides advanced search capabilities that go beyond standard SQL queries, offering both traditional SQL search patterns and Vector-Similarity Search functionality.

SQL-Based Search​

Spice supports basic search patterns directly through SQL, leveraging its SQL query features. For example, you can perform a text search within a table using SQL's LIKE clause:

SELECT id, text_column
FROM my_table
WHERE
    LOWER(text_column) LIKE '%search_term%'
  AND
    date_published > '2021-01-01'

Vector Search​

Spice also provides advanced Vector-Similarity Search capabilities, enabling more nuanced and intelligent searches. The runtime supports both:

  1. Local embedding models, e.g. sentence-transformers/all-MiniLM-L6-v2.

  2. Remote embedding providers, e.g. OpenAI.

See Model Providers to view all supported providers

Embedding models are defined in the spicepod.yaml file as top-level components.

embeddings:
  - from: openai
    name: remote_service
    params:
      openai_api_key: ${ secrets:SPICE_OPENAI_API_KEY }

  - name: local_embedding_model
    from: huggingface:huggingface.co/sentence-transformers/all-MiniLM-L6-v2

Datasets can be augmented with embeddings targeting specific columns, to enable search capabilities through similarity searches.

By defining embeddings on the body column, Spice is now configured to execute similarity searches on the dataset.

For more details, see the API reference for /v1/search.

Spice also supports vector search on datasets with preexisting embeddings. See below for compatibility details.

Chunking Support​

Spice supports chunking of content before embedding, which is useful for large text columns such as those found in Document Tables. Chunking ensures that only the most relevant portions of text are returned during search queries. Chunking is configured as part of the embedding configuration.

The body column will be divided into chunks of approximately 512 tokens, while maintaining structural and semantic integrity (e.g. not splitting sentences).

Document Retrieval​

When performing searches on datasets with chunking enabled, Spice returns the most relevant chunk for each match. To retrieve the full content of a column, include the embedding column in the additional_columns list.

For example:

Response:

Pre-Existing Embeddings​

Datasets that already include embeddings can utilize the same functionalities (e.g., vector search) as those augmented with embeddings using Spice. To ensure compatibility, these table columns must adhere to the following constraints:

  1. Underlying Column Presence:

    • The underlying column must exist in the table, and be of string Arrow data type .

  2. Embeddings Column Naming Convention:

    • For each underlying column, the corresponding embeddings column must be named as <column_name>_embedding. For example, a customer_reviews table with a review column must have a review_embedding column.

  3. Embeddings Column Data Type:

    • The embeddings column must have the following Arrow data type when loaded into Spice:

      1. FixedSizeList[Float32 or Float64, N], where N is the dimension (size) of the embedding vector. FixedSizeList is used for efficient storage and processing of fixed-size vectors.

      2. If the column is chunked, use List[FixedSizeList[Float32 or Float64, N]].

  4. Offset Column for Chunked Data:

    • If the underlying column is chunked, there must be an additional offset column named <column_name>_offsets with the following Arrow data type:

      1. List[FixedSizeList[Int32, 2]], where each element is a pair of integers [start, end] representing the start and end indices of the chunk in the underlying text column. This offset column maps each chunk in the embeddings back to the corresponding segment in the underlying text column.

      2. For instance, [[0, 100], [101, 200]] indicates two chunks covering indices 0–100 and 101–200, respectively.

By following these guidelines, you can ensure that your dataset with pre-existing embeddings is fully compatible with the vector search and other embedding functionalities provided by Spice.

Example​

A table sales with an address column and corresponding embedding column(s).

The same table if it was chunked:

Last updated

Was this helpful?