support_langchain.py

import json
import pandas as pd
from langchain import PromptTemplate
from langchain.chains import LLMChain, SequentialChain  # import LangChain libraries
from langchain.llms import OpenAI  # import OpenAI model
from langchain.prompts import load_prompt


def answer_question(question: str):
	llm = OpenAI(temperature=0.7)

	# Chain 1: Generating a rephrased version of the user's question
	prompt_template = load_prompt("rephrase_prompt.json")
	question_chain = LLMChain(llm=llm, prompt=prompt_template, output_key="statement")

	# Chain 2: Generating assumptions made in the statement
	prompt_template = load_prompt("assumptions_prompt.json")
	assumptions_chain = LLMChain(llm=llm, prompt=prompt_template, output_key="assertions")

	# Chain 3: Fact checking the assumptions
	prompt_template = load_prompt("fact_check_prompt.json")
	fact_checker_chain = LLMChain(llm=llm, prompt=prompt_template, output_key="facts")

	# Final Chain: Generating the final answer to the user's question based on the facts and assumptions
	prompt_template = load_prompt("answer_prompt.json")
	answer_chain = LLMChain(llm=llm, prompt=prompt_template, output_key="answer")
	overall_chain = SequentialChain(
		chains=[question_chain, assumptions_chain, fact_checker_chain, answer_chain],
		input_variables=["question"],
		output_variables=["answer", "assertions", "facts"],
		verbose=True
	)
	results = overall_chain({"question": question})

	print(f"Facts: \n{results['facts']}")
	print(f"\n\nAnswer:{results['answer']}\n\n")

	return results


# modelop.score
def action(data: dict):
	result = answer_question(data['question'])
	yield {'facts': result['facts'], 'answer': result['answer']}


# modelop.metrics
def metrics(data: pd.DataFrame):
	results = {"Questions, Facts, and Answers": []}
	for index, row in data.iterrows():
		llm = OpenAI(temperature=0.7)
		questions = llm.predict(row['prompt'])
		for question in iter(questions.splitlines()):
			result = next(action({'question': question}))
			results["Questions, Facts, and Answers"].append(
				{"question": question, "answer": result["answer"], "facts": result["facts"]})

	yield results


def main():
	# generate_prompt_files()
	print(json.dumps(next(action({'question': 'What was the town of tomichi in colorado like?'})), indent=2))
	print(json.dumps(next(metrics(pd.DataFrame.from_dict([
		{'prompt': 'generate a list of random financial questions without line numbers and one per line'}]))), indent=2))


if __name__ == '__main__':
	main()