utils.py

import chromadb
from sentence_transformers import CrossEncoder, SentenceTransformer
import json

def chroma_client_setup():
    print("Setup client")
    chroma_client = chromadb.Client()
    collection = chroma_client.create_collection(
        name="food_collection",
        metadata={"hnsw:space": "cosine"} # l2 is the default
    )
    return collection

def load_data():
    print("load data")
    with open("test_json.json", "r") as f:
        data = json.load(f)
    return data

def embedding_function(items_to_embed: list[str]):
    print("embedding")
    sentence_model = SentenceTransformer(
        "mixedbread-ai/mxbai-embed-large-v1"
    )
    embedded_items = sentence_model.encode(
        items_to_embed
    )
    print(len(embedded_items))
    print(type(embedded_items[0]))
    print(type(embedded_items[0][0]))
    embedded_list = [item.tolist() for item in embedded_items]
    print(len(embedded_list))
    print(type(embedded_list[0]))
    print(type(embedded_list[0][0]))
    return embedded_list

def chroma_upserting(collection, payload:list[dict]):
    print('upserting')
    print("printing item:")
    embedding = embedding_function([item['doc'] for item in payload])
    print(type(embedding))
    collection.add(
        documents=[item['doc'] for item in payload],
        embeddings=embedding,
        #metadatas=item,
        ids=[f"id_{idx}" for idx, _ in enumerate(payload)]
        )

def search_chroma(collection, query:str):
    results = collection.query(
        query_embeddings=embedding_function([query]),
        n_results=5
    )
    return results

def reranking_results(query: str, top_k_results: list[str]):
    # Load the model, here we use our base sized model
    rerank_model = CrossEncoder("mixedbread-ai/mxbai-rerank-xsmall-v1")
    reranked_results = rerank_model.rank(query, top_k_results, return_documents=True)
    return reranked_results