Spaces:
Running
Running
File size: 5,843 Bytes
5e433de |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 |
"""
To preprocess the data and create a vector database using docling and langchain,
openai embeddings.
"""
import getpass
import os
from dotenv import load_dotenv
import itertools
from uuid import uuid4
import faiss
from langchain_community.docstore.in_memory import InMemoryDocstore
from langchain_community.vectorstores import FAISS
from langchain_openai import OpenAIEmbeddings
from docling.document_converter import DocumentConverter
from langchain_huggingface import HuggingFaceEmbeddings
from transformers import AutoTokenizer
from docling_core.transforms.chunker.hybrid_chunker import HybridChunker
from docling_core.types.doc.document import TableItem,PictureItem
from docling_core.types.doc.labels import DocItemLabel
from langchain_core.documents import Document
import logging
load_dotenv()
def adding_metadata_chunks(chunks: HybridChunker, file_name: str, speciality: str) -> list[Document]:
"""Adding metadata to the chunks
This function processes a list of chunks and adds metadata to each chunk.
Args:
chunks (Hybridchunker): The chunks to be processed.
file_name (str): The name of the file from which the chunks were created.
specality (str): specalization of the book.
Returns:
List[Document]: A list of Document objects with added metadata.
"""
documents = []
for idx, chunk in enumerate(chunks):
items = chunk.meta.doc_items
if len(items) == 1 and isinstance(items[0], TableItem):
# If the chunk is a table, we can skip it
continue
main_ref = " ".join([item.get_ref().cref for item in items])
parent_ref = " ".join([item.parent.get_ref().cref for item in items])
child_ref = " ".join([str(child) for sublist in [item.children for item in items] for child in sublist])
text = chunk.text # The text of the chunk
metadata = {
"source": file_name,
"specilization": speciality,
"chunk_index": idx,
"self_ref": main_ref,
"parent_ref": parent_ref,
"child_ref": child_ref,
"chunk_type": "text",
}
document = Document(page_content=text, metadata=metadata)
documents.append(document)
return documents
def modifying_tables(docling_document, file_name: str, speciality: str) -> list[Document]:
"""Extract the tables from the converted document and add metadata.
Args:
document (Document): converted document.
file_name (str): file name.
specality (str): specalization of the book.
Returns:
list[TableItem]: A list of documents containing table data with
reference IDs in the metadata.
"""
tables: list[Document] = []
for table in docling_document.tables:
if table.label in [DocItemLabel.TABLE]:
main_ref = table.get_ref().cref
parent_ref = table.parent.get_ref().cref
child_ref = table.children
text = table.export_to_markdown()
metadata = {
"source": file_name,
"chunk_index": None,
"self_ref": main_ref,
"parent_ref": parent_ref,
"child_ref": child_ref,
"chunk_type": "table",
}
document = Document(page_content=text, metadata=metadata)
tables.append(document)
return tables
def dataloader(file_path:str, embeddings_model:str) -> list[Document]:
logging.info("Converting the document to docling format...")
docling_document = DocumentConverter().convert(source=file_path).document
file_name = file_path.split("\\")[-1].split(".")[0]
# Create a hybrid chunker to chunk the document
embeddings_tokenizer = AutoTokenizer.from_pretrained(embeddings_model)
logging.info("Chunking the document...")
chunks = HybridChunker(tokenizer=embeddings_tokenizer).chunk(docling_document)
# Add metadata to the chunks
logging.info("Adding metadata to the chunks...")
texts = adding_metadata_chunks(chunks, file_name)
logging.info("Modifying tables...")
tables = modifying_tables(docling_document, file_name)
# Combine the text and table documents into a single list
documents = list(itertools.chain(texts, tables))
logging.info(f"Loaded {len(documents)} documents from {file_name}.")
return documents
def create_vector_database(documents: list[Document]) -> FAISS:
"""Create a vector database from the documents.
Args:
file_path (str): The path to the document file.
embeddings_model (str): The model name for embeddings.
Returns:
list[Document]: A list of Document objects with embeddings.
"""
logging.info("Creating the vector database...")
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
index = faiss.IndexFlatL2(len(embeddings.embed_query("hello world")))
vector_store = FAISS(
embedding_function=embeddings,
index=index,
docstore=InMemoryDocstore(),
index_to_docstore_id={},
)
uuids = [str(uuid4()) for _ in range(len(documents))]
vector_store.add_documents(documents=documents, ids=uuids)
logging.info("Vector database created successfully.")
def main(file_path:str, embeddings_model:str) -> FAISS:
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
documents = dataloader(file_path, embeddings_model)
create_vector_database(documents)
if __name__ == "__main__":
file_path = r"converted\ROBBINS-&-COTRAN-PATHOLOGIC-BASIS-OF-DISEASE-10TH-ED-with-image-refs.md"
embeddings_model = "ibm-granite/granite-embedding-125m-english"
main(file_path, embeddings_model) |