Skip to content

Vector Database with support for late interaction and token level embeddings.

License

Notifications You must be signed in to change notification settings

DeployQL/LintDB

Repository files navigation

icon

LintDB

LintDB is a multi-vector database meant for Gen AI. LintDB natively supports late interaction like ColBERT and PLAID.

Key Features

  • Multi vector support: LintDB stores multiple vectors per document id and calculates the max similarity across vectors to determine relevance.
  • Bit-level Compression: LintDB fully implements PLAID's bit compression, storing 128 dimension embeddings in as low as 16 bytes.
  • Embedded: LintDB can be embedded directly into your Python application. No need to setup a separate database.
  • Full Support for PLAID and ColBERT: LintDB is built around PLAID and ColBERT.
  • Filtering: LintDB supports filtering on any field in the schema.

Installation

LintDB relies on OpenBLAS for accerlated matrix multiplication. To smooth the process of installation, we only support conda.

conda install lintdb -c deployql -c conda-forge

Usage

LintDB makes it easy to upload data, even if you have multiple tenants.

Below shows creating a database. LintDB defines a schema for a given database that can be used to index embeddings, floats, strings, even dates. Fields can be indexed, stored, or used as a filter.

from lintdb.core import (
  Schema,
  ColbertField,
  QuantizerType,
  Configuration,
  IndexIVF
)

schema = Schema(
  [
    ColbertField('colbert', DataType.TENSOR, {
      'dimensions': 128,
      'quantization': QuantizerType.BINARIZER,
      "num_centroids": 32768,
      "num_iterations": 10,
    })
  ]
)
config = Configuration()
index = IndexIVF(index_path, schema, config)
)

And querying the database. We can query any of the data fields we indexed.

from lintdb.core import (
Query,
VectorQueryNode
)
for id, query in zip(data.qids, data.queries):
  embedding = checkpoint.queryFromText(query)
e = np.squeeze(embedding.cpu().numpy().astype('float32'))

query = Query(
  VectorQueryNode(
    TensorFieldValue('colbert', e)
  )
)
results = index.search(0, query, 10)
print(results)

Late Interaction Model Support

LintDB aims to support late interaction and more advanced retrieval models.

  • ColBERTv2 with PLAID
  • XTR

Roadmap

LintDB aims to be a retrieval platform for Gen AI. We believe that to do this, we must support flexible retrieval and scoring methods while maintaining a high level of performance.

  • Improving performance and scalability
  • Improved benchmarks
  • Support CITADEL for scalable late interaction
  • Support learnable query adapters in the retrieval pipeline
  • Enhance support for arbitrary retrieval and ranking functions
  • Support learnable ranking functions

Comparison with other Vector Databases

LintDB is one of two databases that support token level embeddings. The other being Vespa.

Token Level Embeddings

Vespa

Vespa is a robust, mature search engine with many features. However, the learning curve to get started and operate Vespa is high. With embedded LintDB, there's no setup required. conda install lintdb -c deployql and get started.

Embedded

Chroma

Chroma is an embedded vector database available in Python and Javascript. LintDB currently only supports Python.

However, unlike Chroma, LintDB offers multi-tenancy support.

Documentation

For detailed documentation on using LintDB, refer to the official documentation

License

LintDB is licensed under the Apache 2.0 License. See the LICENSE file for details.

We want to offer a managed service

We need your help! If you'd want a managed LintDB, reach out and let us know.

Book time on the founder's calendar: https://calendar.app.google/fsymSzTVT8sip9XX6