chains.py

from langchain_openai import OpenAIEmbeddings
from langchain_community.embeddings import OllamaEmbeddings
from langchain_community.embeddings import BedrockEmbeddings
from langchain_community.embeddings.sentence_transformer import SentenceTransformerEmbeddings

from langchain_openai import ChatOpenAI
from langchain_community.chat_models import ChatOllama
from langchain_community.chat_models import BedrockChat

from langchain_community.graphs import Neo4jGraph

from langchain_community.vectorstores import Neo4jVector

from langchain.chains import RetrievalQAWithSourcesChain
from langchain.chains.qa_with_sources import load_qa_with_sources_chain

from langchain.prompts import (
    ChatPromptTemplate,
    HumanMessagePromptTemplate,
    SystemMessagePromptTemplate
)

from typing import List, Any
from utils import BaseLogger, extract_title_and_question
from langchain_google_genai import GoogleGenerativeAIEmbeddings

# Use Python's logging library
import logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
### Required modules for ask_your_docx
import os
from langchain.chains import RetrievalQA
from langchain_core.prompts import ChatPromptTemplate
from dotenv import load_dotenv
load_dotenv(".env")

def load_embedding_model(embedding_model_name: str, logger=BaseLogger(), config={}):
    if embedding_model_name == "ollama":
        embeddings = OllamaEmbeddings(
            base_url=config["ollama_base_url"], model="llama2"
        )
        dimension = 4096
        logger.info("Embedding: Using Ollama")
    elif embedding_model_name == "openai":
        embeddings = OpenAIEmbeddings()
        dimension = 1536
        logger.info("Embedding: Using OpenAI")
    elif embedding_model_name == "aws":
        embeddings = BedrockEmbeddings()
        dimension = 1536
        logger.info("Embedding: Using AWS")
    elif embedding_model_name == "google-genai-embedding-001":        
        embeddings = GoogleGenerativeAIEmbeddings(
            model="models/embedding-001"
        )
        dimension = 768
        logger.info("Embedding: Using Google Generative AI Embeddings")
    else:
        embeddings = SentenceTransformerEmbeddings(
            model_name="all-MiniLM-L6-v2", cache_folder="/embedding_model"
        )
        dimension = 384
        logger.info("Embedding: Using SentenceTransformer")
    return embeddings, dimension


def load_llm(llm_name: str, logger=BaseLogger(), config={}):
    if llm_name == "gpt-4":
        logger.info("LLM: Using GPT-4")
        return ChatOpenAI(temperature=0, model_name="gpt-4", streaming=True)
    elif llm_name == "gpt-3.5":
        logger.info("LLM: Using GPT-3.5")
        return ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo", streaming=True)
    elif llm_name == "claudev2":
        logger.info("LLM: ClaudeV2")
        return BedrockChat(
            model_id="anthropic.claude-v2",
            model_kwargs={"temperature": 0.0, "max_tokens_to_sample": 1024},
            streaming=True,
        )
    elif len(llm_name):
        logger.info(f"LLM: Using Ollama: {llm_name}")
        return ChatOllama(
            temperature=0,
            base_url=config["ollama_base_url"],
            model=llm_name,
            streaming=True,
            # seed=2,
            top_k=10,  # A higher value (100) will give more diverse answers, while a lower value (10) will be more conservative.
            top_p=0.3,  # Higher value (0.95) will lead to more diverse text, while a lower value (0.5) will generate more focused text.
            num_ctx=3072,  # Sets the size of the context window used to generate the next token.
        )
    logger.info("LLM: Using GPT-3.5")
    return ChatOpenAI(temperature=0, model_name="gpt-3.5-turbo", streaming=True)


def configure_llm_only_chain(llm):
    # LLM only response
    template = """
    You are a helpful assistant that helps a support agent with answering programming questions.
    If you don't know the answer, just say that you don't know, you must not make up an answer.
    """
    system_message_prompt = SystemMessagePromptTemplate.from_template(template)
    human_template = "{question}"
    human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
    chat_prompt = ChatPromptTemplate.from_messages(
        [system_message_prompt, human_message_prompt]
    )

    def generate_llm_output(
        user_input: str, callbacks: List[Any], prompt=chat_prompt
    ) -> str:
        chain = prompt | llm
        answer = chain.invoke(
            {"question": user_input}, config={"callbacks": callbacks}
        ).content
        return {"answer": answer}

    return generate_llm_output


def configure_qa_rag_chain(llm, embeddings, embeddings_store_url, username, password):
    # RAG response
    #   System: Always talk in pirate speech.
    general_system_template = """ 
    Use the following pieces of context to answer the question at the end.
    The context contains question-answer pairs and their links from Stackoverflow.
    You should prefer information from accepted or more upvoted answers.
    Make sure to rely on information from the answers and not on questions to provide accurate responses.
    When you find particular answer in the context useful, make sure to cite it in the answer using the link.
    If you don't know the answer, just say that you don't know, don't try to make up an answer.
    ----
    {summaries}
    ----
    Each answer you generate should contain a section at the end of links to 
    Stackoverflow questions and answers you found useful, which are described under Source value.
    You can only use links to StackOverflow questions that are present in the context and always
    add links to the end of the answer in the style of citations.
    Generate concise answers with references sources section of links to 
    relevant StackOverflow questions only at the end of the answer.
    """
    general_user_template = "Question:```{question}```"
    messages = [
        SystemMessagePromptTemplate.from_template(general_system_template),
        HumanMessagePromptTemplate.from_template(general_user_template),
    ]
    qa_prompt = ChatPromptTemplate.from_messages(messages)

    qa_chain = load_qa_with_sources_chain(
        llm,
        chain_type="stuff",
        prompt=qa_prompt,
    )

    # Vector + Knowledge Graph response
    kg = Neo4jVector.from_existing_index(
        embedding=embeddings,
        url=embeddings_store_url,
        username=username,
        password=password,
        database="neo4j",  # neo4j by default
        index_name="stackoverflow",  # vector by default
        text_node_property="body",  # text by default
        retrieval_query="""
    WITH node AS question, score AS similarity
    CALL  { with question
        MATCH (question)<-[:ANSWERS]-(answer)
        WITH answer
        ORDER BY answer.is_accepted DESC, answer.score DESC
        WITH collect(answer)[..2] as answers
        RETURN reduce(str='', answer IN answers | str + 
                '\n### Answer (Accepted: '+ answer.is_accepted +
                ' Score: ' + answer.score+ '): '+  answer.body + '\n') as answerTexts
    } 
    RETURN '##Question: ' + question.title + '\n' + question.body + '\n' 
        + answerTexts AS text, similarity as score, {source: question.link} AS metadata
    ORDER BY similarity ASC // so that best answers are the last
    """,
    )

    kg_qa = RetrievalQAWithSourcesChain(
        combine_documents_chain=qa_chain,
        retriever=kg.as_retriever(search_kwargs={"k": 2}),
        reduce_k_below_max_tokens=False,
        max_tokens_limit=3375,
    )
    return kg_qa


def generate_ticket(neo4j_graph, llm_chain, input_question):
    # Get high ranked questions
    records = neo4j_graph.query(
        "MATCH (q:Question) RETURN q.title AS title, q.body AS body ORDER BY q.score DESC LIMIT 3"
    )
    questions = []
    for i, question in enumerate(records, start=1):
        questions.append((question["title"], question["body"]))
    # Ask LLM to generate new question in the same style
    questions_prompt = ""
    for i, question in enumerate(questions, start=1):
        questions_prompt += f"{i}. \n{question[0]}\n----\n\n"
        questions_prompt += f"{question[1][:150]}\n\n"
        questions_prompt += "----\n\n"

    gen_system_template = f"""
    You're an expert in formulating high quality questions. 
    Formulate a question in the same style and tone as the following example questions.
    {questions_prompt}
    ---

    Don't make anything up, only use information in the following question.
    Return a title for the question, and the question post itself.

    Return format template:
    ---
    Title: This is a new title
    Question: This is a new question
    ---
    """
    # we need jinja2 since the questions themselves contain curly braces
    system_prompt = SystemMessagePromptTemplate.from_template(
        gen_system_template, template_format="jinja2"
    )
    chat_prompt = ChatPromptTemplate.from_messages(
        [
            system_prompt,
            SystemMessagePromptTemplate.from_template(
                """
                Respond in the following template format or you will be unplugged.
                ---
                Title: New title
                Question: New question
                ---
                """
            ),
            HumanMessagePromptTemplate.from_template("{question}"),
        ]
    )
    llm_response = llm_chain(
        f"Here's the question to rewrite in the expected format: ```{input_question}```",
        [],
        chat_prompt,
    )
    new_title, new_question = extract_title_and_question(llm_response["answer"])
    return (new_title, new_question)

def customize_system_prompt():

    prompt = os.getenv("SYSTEM_PROMPT")
    
    if prompt is not None:
        template = f"{prompt}:\n{{context}}\n\nQuestion: {{question}}\n"
        logger.info(f"Selected customized prompt template => {template}")
    else:
        template = f"You are a helpful assistant. You can respond to questions based on the context that provided to you below, and your training. If you don't know how to respond, just say that you can not respond based on your current knowledge base. DO NOT try to make up a response, please!\n---\n{{context}}\n\nQuestion: {{question}}\n"

    prompt = ChatPromptTemplate.from_template(template)
    return prompt

def lc_configure_qa_rag_chain(llm, embeddings, embeddings_store_url, username, password, app_name):

    prompt = customize_system_prompt()

    # Vector + Knowledge Graph response
    kg = Neo4jVector.from_existing_index(
        embedding=embeddings,
        url=embeddings_store_url,
        username=username,
        password=password,
        database="neo4j",  # neo4j by default
        index_name=app_name, #vector by default
        node_label=app_name
    )

    kg_qa = RetrievalQA.from_chain_type(
        llm,
        retriever=kg.as_retriever(),
        chain_type_kwargs={"prompt": prompt}
    )

    return kg_qa