chain.py

import json
import os
import pathlib
import pickle
from typing import Dict, List, Tuple

from langchain import PromptTemplate
from langchain.chains import LLMChain
from langchain.chains.base import Chain
from langchain.chains.combine_documents.base import BaseCombineDocumentsChain
from langchain.chains.conversation.memory import ConversationBufferMemory
from langchain.chains.question_answering import load_qa_chain
from langchain.prompts import FewShotPromptTemplate, PromptTemplate
from langchain.prompts.example_selector import \
    SemanticSimilarityExampleSelector
from langchain.vectorstores import FAISS, Weaviate
from pydantic import BaseModel

class CustomChain(Chain, BaseModel):

    vstore: FAISS
    chain: BaseCombineDocumentsChain
    key_word_extractor: Chain

    @property
    def input_keys(self) -> List[str]:
        return ["question"]

    @property
    def output_keys(self) -> List[str]:
        return ["answer"]

    def _call(self, inputs: Dict[str, str]) -> Dict[str, str]:
        question = inputs["question"]
        chat_history_str = _get_chat_history(inputs["chat_history"])
        if chat_history_str:
            new_question = self.key_word_extractor.run(
                question=question, chat_history=chat_history_str
            )
        else:
            new_question = question
        print(new_question)
        docs = self.vstore.similarity_search(new_question, k=3)
        new_inputs = inputs.copy()
        new_inputs["question"] = new_question
        new_inputs["chat_history"] = chat_history_str
        answer, _ = self.chain.combine_docs(docs, **new_inputs)

        ## Dedupe source list
        source_list = [doc.metadata['source'] for doc in docs]

        source_string = "\n\n*Sources:* "
        for i, source in enumerate(set(source_list)):
            source_string += f"[[{i}](https://{source})]"

        final_answer = answer + source_string
        return {"answer": final_answer}

def get_new_chain1(vectorstore, rephraser_llm, final_output_llm, isFlan) -> Chain:
    _eg_template = """## Example:

    Chat History:
    {chat_history}
    Follow Up Input: {question}
    Standalone question: {answer}"""
    _eg_prompt = PromptTemplate(
        template=_eg_template,
        input_variables=["chat_history", "question", "answer"],
    )

    _prefix = """Given the following conversation and a follow up question, rephrase the follow up question to be a standalone question. You should assume that the question is related to Hugging Face Code."""
    _suffix = """## Example:

    Chat History:
    {chat_history}
    Follow Up Input: {question}
    Standalone question:"""

    #### LOAD VSTORE WITH REPHRASE EXAMPLES
    with open("rephrase_eg.pkl", 'rb') as f:
        rephrase_example_selector = pickle.load(f)

    prompt = FewShotPromptTemplate(
        prefix=_prefix,
        suffix=_suffix,
        example_selector=rephrase_example_selector,
        example_prompt=_eg_prompt,
        input_variables=["question", "chat_history"],
    )

    key_word_extractor = LLMChain(llm=rephraser_llm, prompt=prompt)

    EXAMPLE_PROMPT = PromptTemplate(
        template=">Example:\nContent:\n---------\n{page_content}\n----------\nSource: {source}",
        input_variables=["page_content", "source"],
    )

    gpt_template = """You are an AI assistant for the open source transformers library provided by Hugging Face. The documentation is located at https://huggingface.co/docs/transformers.
- You are given extracted parts of a long document and a question. 
- Provide a conversational answer with a hyperlink to the documentation based on the "source". 
- Do NOT add .html to the end of links. Make sure to bold link text.
- You should only use hyperlinks that are explicitly listed as a source in the context. Do NOT make up a hyperlink that is not listed.
- If the question includes a request for code, provide a code block directly from the documentation.
- If you don't know the answer, just say "Hmm, I'm not sure." Don't try to make up an answer.
- If the question is not about Hugging Face Transformers, politely inform them that you are tuned to only answer questions about Transformers.

For example, if someone asks how to install Transformers, you should say:

You can install with pip: 
'''py
pip install transformers
'''
**(Source)**[https://huggingface.co/docs/transformers/main/en/installation]

Question: {question}
=========
{context}
=========
Answer in Markdown:"""

    flan_template = """
    {context}
    Based on the above documentation, answer the user's question in markdown: {question}"""

    PROMPT = PromptTemplate(template=gpt_template, input_variables=["question", "context"])

    if isFlan:
        PROMPT = PromptTemplate(template=flan_template, input_variables=["question", "context"])

    doc_chain = load_qa_chain(
        final_output_llm,
        chain_type="stuff",
        prompt=PROMPT,
        document_prompt=EXAMPLE_PROMPT,
        verbose=True
    )
    return CustomChain(chain=doc_chain, vstore=vectorstore, key_word_extractor=key_word_extractor)


def _get_chat_history(chat_history: List[Tuple[str, str]]):
    buffer = ""
    for human_s, ai_s in chat_history[-2:]:
        human = f"Human: " + human_s
        ai = f"Assistant: " + ai_s
        buffer += "\n" + "\n".join([human, ai])

    
    return buffer