Korok is a lightweight hybrid search and reranking package.
To install the package, run the following command:
make installThe following code snippet demonstrates how to use Korok for nearest neighbor search with a Model2Vec encoder:
from model2vec import StaticModel
from korok import Pipeline
# Create texts to encode
texts = ["triforce", "master sword", "hylian shield", "boomerang", "hookshot"]
# Initialize the encoder and pipeline
encoder = StaticModel.from_pretrained("minishlab/potion-base-8M")
pipeline = Pipeline.fit(texts=texts, encoder=encoder)
# Query for nearest neighbors
query_text = "sword"
results = pipeline.query([query_text], k=3)The following code snippet demonstrates how to use Korok with a reranker:
from model2vec import StaticModel
from korok import Pipeline
from korok.rerankers import CrossEncoderReranker
# Create texts to encode
texts = ["triforce", "master sword", "hylian shield", "boomerang", "hookshot"]
# Initialize the encoder, reranker and pipeline
encoder = StaticModel.from_pretrained("minishlab/potion-base-8M")
reranker = CrossEncoderReranker()
pipeline = Pipeline.fit(texts=texts, encoder=encoder, reranker=reranker)
# Query for nearest neighbors with reranking
query_text = "sword"
results = pipeline.query([query_text], k=3)