summary.py

from langchain.prompts import load_prompt
from langchain.chains.summarize import load_summarize_chain
from langchain_core.documents.base import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter

def summary(sum_llm, text):
	""" Return summary of giving text """

	# Prompts for summary
	map_prompt = load_prompt('lc://prompts/summarize/map_reduce/map.yaml')
	combine_prompt = load_prompt('lc://prompts/summarize/map_reduce/combine.yaml')

	documents = Document(page_content=text, metadata={"source": "local"}) # Convert text to langchain Document
	
	text_splitter = RecursiveCharacterTextSplitter(
		chunk_size = 7000,
		chunk_overlap  = 0,
		length_function = len,
		is_separator_regex = False
	)
	documents = text_splitter.split_documents([documents]) # Split text
	
	sum_chain = load_summarize_chain(
		sum_llm, chain_type="map_reduce", 
		map_prompt=map_prompt,
		combine_prompt=combine_prompt,
		verbose=False
	)
	
	summary = sum_chain.invoke({"input_documents": documents}) # Taking summary
	
	return summary