handler.py

import torch
import locale
import os
from typing import Dict, List, Any
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, BitsAndBytesConfig
from langchain.llms import HuggingFacePipeline
from langchain.retrievers.document_compressors import LLMChainExtractor
from langchain.retrievers import ContextualCompressionRetriever
from langchain.vectorstores import Chroma
from langchain import PromptTemplate, LLMChain
from langchain.chains import RetrievalQA, ConversationalRetrievalChain
from langchain.prompts import PromptTemplate
from langchain.prompts.prompt import PromptTemplate
from langchain.memory import ConversationBufferMemory
from langchain.embeddings import HuggingFaceBgeEmbeddings
from langchain.document_loaders import WebBaseLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.chains.qa_with_sources import load_qa_with_sources_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import HumanMessage
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableLambda, RunnableBranch, RunnablePassthrough
from operator import itemgetter
from langchain.schema import format_document
from langchain.memory import ConversationBufferMemory
from langchain_core.messages import AIMessage, HumanMessage, get_buffer_string
from langchain_core.runnables import RunnableParallel
from typing import Optional
from langchain.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores.utils import DistanceStrategy

EMBEDDING_MODEL_NAME = "mixedbread-ai/mxbai-embed-large-v1"
MARKDOWN_SEPARATORS = [
    "\n#{1,6} ",
    "```\n",
    "\n\\*\\*\\*+\n",
    "\n---+\n",
    "\n___+\n",
    "\n\n",
    "\n",
    " ",
    "",
]

class EndpointHandler():

    def __init__(self, path=""):
        # Config LangChain
        os.environ["LANGCHAIN_TRACING_V2"] = "true"
        os.environ["LANGCHAIN_API_KEY"] = "ls__9834e6b2ff094d43a28418c9ecea2fd5"
        
        # Load Vector db
        urls = [
            "https://scholars.cityu.edu.hk/en/persons/man-hon-michael-cheung(0f913a96-a28d-47ea-848c-f444804c16f2).html",
            "https://scholars.cityu.edu.hk/en/persons/man-hon-michael-cheung(0f913a96-a28d-47ea-848c-f444804c16f2)/publications.html",
            "https://www.cityu.edu.hk/media/press-release/2022/05/17/cityu-council-announces-appointment-professor-freddy-boey-next-president",
            "https://www.cityu.edu.hk/media/press-release/2023/05/18/professor-freddy-boey-installed-5th-president-cityu",
        ]
        
        loader = WebBaseLoader(urls)
        docs = loader.load()
        
        text_splitter = RecursiveCharacterTextSplitter.from_huggingface_tokenizer(
            AutoTokenizer.from_pretrained(EMBEDDING_MODEL_NAME),
            chunk_size=512,
            chunk_overlap=int(512 / 10),
            add_start_index=True,
            strip_whitespace=True,
            separators=MARKDOWN_SEPARATORS,
        )

        docs_processed = []
        for doc in docs:
            docs_processed += text_splitter.split_documents([doc])

        # Remove duplicates
        unique_texts = {}
        docs_processed_unique = []
        for doc in docs_processed:
            if doc.page_content not in unique_texts:
                unique_texts[doc.page_content] = True
                docs_processed_unique.append(doc)

        embedding_model = HuggingFaceEmbeddings(
            model_name=EMBEDDING_MODEL_NAME,
            multi_process=True,
            model_kwargs={"device": "cuda"},
            encode_kwargs={"normalize_embeddings": True},  # set True for cosine similarity
        )
        
        self.vectorstore = FAISS.from_documents(
            docs_processed_unique, embedding_model, distance_strategy=DistanceStrategy.COSINE
        )

        # Create LLM
        READER_MODEL_NAME = path

        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.bfloat16,
        )
        model = AutoModelForCausalLM.from_pretrained(READER_MODEL_NAME, quantization_config=bnb_config)
        tokenizer = AutoTokenizer.from_pretrained(READER_MODEL_NAME)

        # Testing
        # tokenizer.pad_token = tokenizer.eos_token
        
        self.READER_LLM = pipeline(
            model=model,
            tokenizer=tokenizer,
            task="text-generation",
            do_sample=True,
            temperature=0.2,
            repetition_penalty=1.1,
            return_full_text=False,
            max_new_tokens=256,
        )
        
        prompt_in_chat_format = [
            {
                "role": "system",
                "content": """Using the information contained in the context.
        Respond only to the question asked, response should be concise and relevant to the question.
        If the answer cannot be deduced from the context, do not give an answer.""",
            },
            {
                "role": "user",
                "content": """Context: {context}
        Now here is the question you need to answer.
        Question: {question}""",
            },
        ]
        
        self.RAG_PROMPT_TEMPLATE = tokenizer.apply_chat_template(
            prompt_in_chat_format, tokenize=False, add_generation_prompt=True
        )

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        # get inputs
        inputs = data.pop("inputs",data)
        date = data.pop("date", None)

        retrieved_docs = self.vectorstore.similarity_search(query=inputs, k=2)
        
        retrieved_docs_text = [
            doc.page_content for doc in retrieved_docs
        ]  # we only need the text of the documents
        context = "\nExtracted documents:\n"
        context += "".join([f"Document {str(i)}:::\n" + doc for i, doc in enumerate(retrieved_docs_text)])

        final_prompt = self.RAG_PROMPT_TEMPLATE.format(
            question=inputs, context=context
        )
        
        # Redact an answer
        answer = self.READER_LLM(final_prompt)[0]["generated_text"]

        return answer