import lancedb
import os
import gradio as gr
from sentence_transformers import SentenceTransformer
from sentence_transformers import CrossEncoder
# 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")
BATCH_SIZE = int(os.getenv("BATCH_SIZE", 32))

retriever = SentenceTransformer(os.getenv("EMB_MODEL"))
# reranker = FlagReranker(os.getenv("RERANKER_MODEL", 'BAAI/bge-reranker-large'), use_fp16=True)
reranker = CrossEncoder(os.getenv("RERANKER_MODEL", 'cross-encoder/ms-marco-MiniLM-L-6-v2'), max_length=512)

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]

        return documents

    except Exception as e:
        raise gr.Error(str(e))
        
def rerank(documents, query, k):
    try:
        query_pairs = [[query, doc] for doc in documents]
        scores = reranker.predict(query_pairs)
        scored_documents = list(zip(documents, scores))
        scored_documents.sort(key=lambda x: x[1], reverse=True)
        top_k_documents = [doc for doc, _ in scored_documents[:k]]
        return top_k_documents

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