Optimized implementation of PLAID index for fast ColBERT retrieval, providing 10x storage compression and sub-200ms latency. Default index backend for PyLate library, enabling efficient multi-vector late interaction retrieval.
FastPLAID is an optimized implementation of the PLAID (Performance-optimized Late Interaction Driver) index specifically designed for fast ColBERT retrieval. It serves as the default index backend in the PyLate library.
FastPLAID is a purpose-built index for fast ColBERT retrieval, specifically optimized for late interaction models that retain all token representations.
FastPLAID is significantly faster than the original Stanford PLAID implementation. While the Stanford PLAID is primarily for research or comparison purposes, FastPLAID is designed for production use with practical optimizations.
FastPLAID is seamlessly integrated into PyLate, making it the default choice for users building ColBERT-based retrieval systems. Users can also opt for the original Stanford PLAID implementation if needed for research purposes.
##Implementation Example
from pylate import indexes
index = indexes.PLAID(
index_folder="pylate-colbert-index",
index_name="my_documents",
override=True, # Use FastPLAID by default
)
For complete, high-performance multi-vector search pipelines, FastPLAID can be paired with pylate-rs, a lightweight Rust-based implementation for production use.
Free and open-source as part of PyLate.
Loading more......