from transformers import AutoTokenizer
from langchain.document_loaders import BSHTMLLoader, DirectoryLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import Chroma
from langchain import HuggingFacePipeline

def prepare_data(db_path, llm_path):
    bshtml_dir_loader = DirectoryLoader('./data/', loader_cls=BSHTMLLoader)
    data = bshtml_dir_loader.load()
    bloomz_tokenizer = AutoTokenizer.from_pretrained("bigscience/bloomz-1b7")
    text_splitter = CharacterTextSplitter.from_huggingface_tokenizer(bloomz_tokenizer, chunk_size=100, chunk_overlap=0, separator="\n")
    documents = text_splitter.split_documents(data)

    embeddings = HuggingFaceEmbeddings()

    llm = HuggingFacePipeline.from_model_id(
        model_id="bigscience/bloomz-1b7",
        task="text-generation",
        model_kwargs={"temperature" : 0, "max_length" : 500})

    vectordb = Chroma.from_documents(documents=documents, embedding=embeddings, persist_directory=db_path)
    vectordb.persist()

    return llm