backend/semantic_search.py

import typing

import lancedb
import os
import gradio as gr
from sentence_transformers import SentenceTransformer
from FlagEmbedding import FlagReranker

db = lancedb.connect(".lancedb")

TABLE = db.open_table(os.getenv("TABLE_NAME"))
VECTOR_COLUMN = os.getenv("VECTOR_COLUMN", "vector")
TEXT_COLUMN = os.getenv("TEXT_COLUMN", "text")
TOP_K = int(os.getenv("TOP_K", 5))
TOP_K_RERANK = int(os.getenv("TOP_K_RERANK", 2))
BATCH_SIZE = int(os.getenv("BATCH_SIZE", 32))
RERANK_MODEL = os.getenv("RERANK_MODEL", "BAAI/bge-reranker-large")

retriever = SentenceTransformer(os.getenv("EMB_MODEL"))
reranker = FlagReranker(RERANK_MODEL,
                        use_fp16=True)  # Setting use_fp16 to True speeds up computation with a slight performance degradation


def rerank(query: str, documents: typing.List[str], k: int):
    data_for_reranker = [(query, document) for document in documents]
    scores = reranker.compute_score(data_for_reranker, batch_size=BATCH_SIZE)
    indices_scores = [(i, score) for (i, score) in enumerate(scores)]
    indices_scores.sort(key=lambda x: x[1], reverse=True)
    best_indices = list(map(lambda x: x[0], indices_scores[:k]))
    return [documents[i] for i in best_indices]


def retrieve(query, k):
    query_vec = retriever.encode(query)
    try:
        documents = TABLE.search(query_vec, vector_column_name=VECTOR_COLUMN).limit(k).to_list()
        documents = [doc[TEXT_COLUMN] for doc in documents]
        documents = rerank(query, documents, TOP_K_RERANK)
        return documents

    except Exception as e:
        raise gr.Error(str(e))


if __name__ == "__main__":
    retrieve("What is RAG?", TOP_K)