app.py

from langchain_community.document_loaders import (
    PyPDFLoader,
    TextLoader,
    DirectoryLoader,
    CSVLoader,
    UnstructuredExcelLoader,
    Docx2txtLoader,
)
from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
import tiktoken
import gradio as gr

import os, tempfile, glob, random
from pathlib import Path
#from IPython.display import Markdown
from PIL import Image
from getpass import getpass
import numpy as np
from itertools import combinations
import pypdf

# LLM: openai and google_genai
import openai
from langchain_openai import OpenAI, OpenAIEmbeddings, ChatOpenAI
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_google_genai import GoogleGenerativeAIEmbeddings


# LLM: HuggingFace
from langchain_community.embeddings import HuggingFaceInferenceAPIEmbeddings
from langchain_community.llms import HuggingFaceHub

# langchain prompts, memory, chains...
from langchain.prompts import PromptTemplate, ChatPromptTemplate
from langchain.chains import ConversationalRetrievalChain
from langchain_community.chat_message_histories import StreamlitChatMessageHistory
from operator import itemgetter
from langchain_core.runnables import RunnableLambda, RunnableParallel, RunnablePassthrough
from langchain.schema import Document, format_document
from langchain_core.messages import AIMessage, HumanMessage, get_buffer_string

from langchain_google_genai import (
    ChatGoogleGenerativeAI,
    HarmBlockThreshold,
    HarmCategory,
)

# OutputParser
from langchain_core.output_parsers import StrOutputParser

# Chroma: vectorstore
from langchain_community.vectorstores import Chroma

# Contextual Compression
from langchain.retrievers.document_compressors import DocumentCompressorPipeline
from langchain.text_splitter import CharacterTextSplitter
from langchain_community.document_transformers import EmbeddingsRedundantFilter,LongContextReorder
from langchain.retrievers.document_compressors import EmbeddingsFilter
from langchain.retrievers import ContextualCompressionRetriever

from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import CohereRerank
from langchain_community.llms import Cohere

from langchain.memory import ConversationSummaryBufferMemory,ConversationBufferMemory


from langchain.schema import Document


# Cohere
from langchain.retrievers.document_compressors import CohereRerank
from langchain_community.llms import Cohere

openai_api_key = os.environ['openai_key']
google_api_key = os.environ['gemini_key']
HF_key = os.environ['HF_token']
cohere_api_key = os.environ['cohere_api']

current_dir = os.getcwd()



prompt_templates = {"All Needs Experts": "Respond as if you are combination of all needs assessment experts."}
actor_description = {"All Needs Experts": "<div style='float: left;margin: 0px 5px 0px 5px;'><img src='https://na.weshareresearch.com/wp-content/uploads/2023/04/experts2.jpg' alt='needs expert image' style='width:70px;align:top;'></div>A combiation of all needs assessment experts."}

def get_empty_state():
    return { "messages": []}


def download_prompt_templates():
    url = "https://huggingface.co/spaces/ryanrwatkins/needs/raw/main/gurus.txt"
    try:
        response = requests.get(url)
        reader = csv.reader(response.text.splitlines())
        next(reader)  # skip the header row
        for row in reader:
            if len(row) >= 2:
                act = row[0].strip('"')
                prompt = row[1].strip('"')
                description = row[2].strip('"')
                prompt_templates[act] = prompt
                actor_description[act] = description

    except requests.exceptions.RequestException as e:
        print(f"An error occurred while downloading prompt templates: {e}")
        return

    choices = list(prompt_templates.keys())
    choices = choices[:1] + sorted(choices[1:])
    return gr.update(value=choices[0], choices=choices)

def on_prompt_template_change(prompt_template):
    if not isinstance(prompt_template, str): return
    return prompt_templates[prompt_template]

def on_prompt_template_change_description(prompt_template):
    if not isinstance(prompt_template, str): return
    return actor_description[prompt_template]
















def langchain_document_loader():
    """
    Load documents from the temporary directory (TMP_DIR). 
    Files can be in txt, pdf, CSV or docx format.
    """
    #current_dir = os.getcwd()
    #TMP_DIR = current_dir
    global documents
    documents = []

    """
    txt_loader = DirectoryLoader(
        TMP_DIR.as_posix(), glob="**/*.txt", loader_cls=TextLoader, show_progress=True
    )
    documents.extend(txt_loader.load())
    """
    pdf_loader = DirectoryLoader(
        current_dir, glob="*.pdf", loader_cls=PyPDFLoader, show_progress=True
    )
    documents.extend(pdf_loader.load())
    """
    csv_loader = DirectoryLoader(
        TMP_DIR.as_posix(), glob="**/*.csv", loader_cls=CSVLoader, show_progress=True,
        loader_kwargs={"encoding":"utf8"}
    )
    documents.extend(csv_loader.load())
    
    doc_loader = DirectoryLoader(
        #TMP_DIR.as_posix(),
        current_dir,
        glob="**/*.docx",
        loader_cls=Docx2txtLoader,
        show_progress=True,
    )
    documents.extend(doc_loader.load())
    """
    return documents

langchain_document_loader()

text_splitter = RecursiveCharacterTextSplitter(
    separators = ["\n\n", "\n", " ", ""],    
    chunk_size = 1600,
    chunk_overlap= 200
)

# Text splitting
chunks = text_splitter.split_documents(documents=documents)




def tiktoken_tokens(documents,model="gpt-3.5-turbo"):
    """Use tiktoken (tokeniser for OpenAI models) to return a list of token lengths per document."""    
    encoding = tiktoken.encoding_for_model(model) # returns the encoding used by the model.
    
    tokens_length = [len(encoding.encode(documents[i].page_content)) for i in range(len(documents))]

    return tokens_length


chunks_length = tiktoken_tokens(chunks,model="gpt-3.5-turbo")

print(f"Number of tokens - Average : {int(np.mean(chunks_length))}")
print(f"Number of tokens - 25% percentile : {int(np.quantile(chunks_length,0.25))}")
print(f"Number of tokens - 50% percentile : {int(np.quantile(chunks_length,0.5))}")
print(f"Number of tokens - 75% percentile : {int(np.quantile(chunks_length,0.75))}")



def select_embeddings_model(LLM_service="HuggingFace"):
    """Connect to the embeddings API endpoint by specifying 
    the name of the embedding model.
    if LLM_service == "OpenAI":
        embeddings = OpenAIEmbeddings(
            model='text-embedding-ada-002',
            api_key=openai_api_key)
    """

    """
    if LLM_service == "Google":
        embeddings = GoogleGenerativeAIEmbeddings(
            model="models/embedding-001",
            google_api_key=google_api_key,
        )
    """
        
    if LLM_service == "HuggingFace":
        embeddings = HuggingFaceInferenceAPIEmbeddings(    
            api_key=HF_key, 
            model_name="thenlper/gte-large"
        )
         
    return embeddings

#embeddings_OpenAI = select_embeddings_model(LLM_service="OpenAI")
#embeddings_google = select_embeddings_model(LLM_service="Google")
embeddings_HuggingFace = select_embeddings_model(LLM_service="HuggingFace")




def create_vectorstore(embeddings,documents,vectorstore_name):
    """Create a Chroma vector database."""
    persist_directory = (current_dir + "/" + vectorstore_name)
    embedding_function=embeddings
    vector_store = Chroma.from_documents(
        documents=documents,
        embedding=embeddings,
        persist_directory=persist_directory
    )
    return vector_store



create_vectorstores = True # change to True to create vectorstores

if create_vectorstores:
    """
    vector_store_OpenAI,_ = create_vectorstore(
        embeddings=embeddings_OpenAI,
        documents = chunks,
        vectorstore_name="Vit_All_OpenAI_Embeddings",
    )
    print("vector_store_OpenAI:",vector_store_OpenAI._collection.count(),"chunks.")
    """
    """
    vector_store_google,new_vectorstore_name = create_vectorstore(
        embeddings=embeddings_google,
        documents = chunks,
        vectorstore_name="Vit_All_Google_Embeddings"
    )
    print("vector_store_google:",vector_store_google._collection.count(),"chunks.")
    """
    
    vector_store_HF = create_vectorstore(
        embeddings=embeddings_HuggingFace,
        documents = chunks,
        vectorstore_name="Vit_All_HF_Embeddings"
    )
    print("vector_store_HF:",vector_store_HF._collection.count(),"chunks.")
    print("")

"""
vector_store_OpenAI = Chroma(
    persist_directory = LOCAL_VECTOR_STORE_DIR.as_posix() + "/Vit_All_OpenAI_Embeddings",
    embedding_function=embeddings_OpenAI)
print("vector_store_OpenAI:",vector_store_OpenAI._collection.count(),"chunks.")
"""

"""
vector_store_google = Chroma(
    persist_directory = current_dir + "/Vit_All_Google_Embeddings",
    embedding_function=embeddings_google)
print("vector_store_google:",vector_store_google._collection.count(),"chunks.")
"""

vector_store_HF = Chroma(
    persist_directory = current_dir + "/Vit_All_HF_Embeddings",
    embedding_function=embeddings_HuggingFace)
print("vector_store_HF:",vector_store_HF._collection.count(),"chunks.")


def Vectorstore_backed_retriever(
vectorstore,search_type="similarity",k=4,score_threshold=None
):
    """create a vectorsore-backed retriever
    Parameters: 
        search_type: Defines the type of search that the Retriever should perform.
            Can be "similarity" (default), "mmr", or "similarity_score_threshold"
        k: number of documents to return (Default: 4) 
        score_threshold: Minimum relevance threshold for similarity_score_threshold (default=None)
    """
    search_kwargs={}
    if k is not None:
        search_kwargs['k'] = k
    if score_threshold is not None:
        search_kwargs['score_threshold'] = score_threshold
    global retriever
    retriever = vectorstore.as_retriever(
        search_type=search_type,
        search_kwargs=search_kwargs
    )
    return retriever

# similarity search
#base_retriever_OpenAI = Vectorstore_backed_retriever(vector_store_OpenAI,"similarity",k=10)
#base_retriever_google = Vectorstore_backed_retriever(vector_store_google,"similarity",k=10)
base_retriever_HF = Vectorstore_backed_retriever(vector_store_HF,"similarity",k=10)



def create_compression_retriever(embeddings, base_retriever, chunk_size=500, k=16, similarity_threshold=None):
    """Build a ContextualCompressionRetriever.
    We wrap the the base_retriever (a vectorstore-backed retriever) into a ContextualCompressionRetriever.
    The compressor here is a Document Compressor Pipeline, which splits documents
    into smaller chunks, removes redundant documents, filters out the most relevant documents,
    and reorder the documents so that the most relevant are at the top and bottom of the list.
    
    Parameters:
        embeddings: OpenAIEmbeddings, GoogleGenerativeAIEmbeddings or HuggingFaceInferenceAPIEmbeddings.
        base_retriever: a vectorstore-backed retriever.
        chunk_size (int): Documents will be splitted into smaller chunks using a CharacterTextSplitter with a default chunk_size of 500. 
        k (int): top k relevant chunks to the query are filtered using the EmbeddingsFilter. default =16.
        similarity_threshold : minimum relevance threshold used by the EmbeddingsFilter. default =None.
    """
    
    # 1. splitting documents into smaller chunks
    splitter = CharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=0, separator=". ")
    
    # 2. removing redundant documents
    redundant_filter = EmbeddingsRedundantFilter(embeddings=embeddings)

    # 3. filtering based on relevance to the query    
    relevant_filter = EmbeddingsFilter(embeddings=embeddings, k=k, similarity_threshold=similarity_threshold) # similarity_threshold and top K

    # 4. Reorder the documents 
    
    # Less relevant document will be at the middle of the list and more relevant elements at the beginning or end of the list.
    # Reference: https://python.langchain.com/docs/modules/data_connection/retrievers/long_context_reorder
    reordering = LongContextReorder()

    # 5. Create compressor pipeline and retriever
    
    pipeline_compressor = DocumentCompressorPipeline(
        transformers=[splitter, redundant_filter, relevant_filter, reordering]  
    )
    compression_retriever = ContextualCompressionRetriever(
        base_compressor=pipeline_compressor, 
        base_retriever=base_retriever
    )

    return compression_retriever

def CohereRerank_retriever(
    base_retriever, 
    cohere_api_key,cohere_model="rerank-multilingual-v2.0", top_n=8
):
    """Build a ContextualCompressionRetriever using Cohere Rerank endpoint to reorder the results based on relevance.
    Parameters:
       base_retriever: a Vectorstore-backed retriever
       cohere_api_key: the Cohere API key
       cohere_model: The Cohere model can be either 'rerank-english-v2.0' or 'rerank-multilingual-v2.0', with the latter being the default.
       top_n: top n results returned by Cohere rerank, default = 8.
    """
    
    compressor = CohereRerank(
        cohere_api_key=cohere_api_key, 
        model=cohere_model, 
        top_n=top_n
    )

    retriever_Cohere = ContextualCompressionRetriever(
        base_compressor=compressor,
        base_retriever=base_retriever
    )
    return retriever_Cohere




def retrieval_blocks(
    create_vectorstore=True,# if True a Chroma vectorstore is created, else the Chroma vectorstore will be loaded
    LLM_service="HuggingFace",
    vectorstore_name="Vit_All_HF_Embeddings",
    chunk_size = 1600, chunk_overlap=200, # parameters of the RecursiveCharacterTextSplitter
    retriever_type="Vectorstore_backed_retriever",
    base_retriever_search_type="similarity", base_retriever_k=10, base_retriever_score_threshold=None,
    compression_retriever_k=16,
    cohere_api_key="***", cohere_model="rerank-multilingual-v2.0", cohere_top_n=8,
):
    """
    Rertieval includes: document loaders, text splitter, vectorstore and retriever. 
    
    Parameters: 
        create_vectorstore (boolean): If True, a new Chroma vectorstore will be created. Otherwise, an existing vectorstore will be loaded.
        LLM_service: OpenAI, Google or HuggingFace.
        vectorstore_name (str): the name of the vectorstore.
        chunk_size and chunk_overlap: parameters of the RecursiveCharacterTextSplitter, default = (1600,200).
        
        retriever_type (str): in [Vectorstore_backed_retriever,Contextual_compression,Cohere_reranker]
        
        base_retriever_search_type: search_type in ["similarity", "mmr", "similarity_score_threshold"], default = similarity.
        base_retriever_k: The most similar vectors to retrieve (default k = 10).  
        base_retriever_score_threshold: score_threshold used by the base retriever, default = None.

        compression_retriever_k: top k documents returned by the compression retriever, default=16
        
        cohere_api_key: Cohere API key
        cohere_model (str): The Cohere model can be either 'rerank-english-v2.0' or 'rerank-multilingual-v2.0', with the latter being the default.
        cohere_top_n: top n results returned by Cohere rerank, default = 8.
   
    Output:
        retriever.
    """
    try:
        # Create new Vectorstore (Chroma index)
        if create_vectorstore: 
            # 1. load documents
            documents = langchain_document_loader(current_dir)
            
            # 2. Text Splitter: split documents to chunks
            text_splitter = RecursiveCharacterTextSplitter(
                separators = ["\n\n", "\n", " ", ""],    
                chunk_size = chunk_size,
                chunk_overlap= chunk_overlap
            )
            chunks = text_splitter.split_documents(documents=documents)
            
            # 3. Embeddings
            embeddings = select_embeddings_model(LLM_service=LLM_service)
        
            # 4. Vectorsore: create Chroma index
            vector_store = create_vectorstore(
                embeddings=embeddings,
                documents = chunks,
                vectorstore_name=vectorstore_name,
            )
    
        # 5. Load a Vectorstore (Chroma index)
        else: 
            embeddings = select_embeddings_model(LLM_service=LLM_service)        
            vector_store = Chroma(
                persist_directory = current_dir + "/" + vectorstore_name,
                embedding_function=embeddings
            )
            
        # 6. base retriever: Vector store-backed retriever 
        base_retriever = Vectorstore_backed_retriever(
            vector_store,
            search_type=base_retriever_search_type,
            k=base_retriever_k,
            score_threshold=base_retriever_score_threshold
        )
        retriever = None
        if retriever_type=="Vectorstore_backed_retriever": 
            retriever = base_retriever
    
        # 7. Contextual Compression Retriever
        if retriever_type=="Contextual_compression":    
            retriever = create_compression_retriever(
                embeddings=embeddings,
                base_retriever=base_retriever,
                k=compression_retriever_k,
            )
    
        # 8. CohereRerank retriever
        if retriever_type=="Cohere_reranker":
            retriever = CohereRerank_retriever(
                base_retriever=base_retriever, 
                cohere_api_key=cohere_api_key, 
                cohere_model=cohere_model, 
                top_n=cohere_top_n
            )
    
        print(f"\n{retriever_type} is created successfully!")
        print(f"Relevant documents will be retrieved from vectorstore ({vectorstore_name}) which uses {LLM_service} embeddings \
and has {vector_store._collection.count()} chunks.")
        
        return retriever
    except Exception as e:
        print(e)









def instantiate_LLM(LLM_provider,api_key,temperature=0.5,top_p=0.95,model_name=None):
    """Instantiate LLM in Langchain.
    Parameters:
        LLM_provider (str): the LLM provider; in ["OpenAI","Google","HuggingFace"]
        model_name (str): in ["gpt-3.5-turbo", "gpt-3.5-turbo-0125", "gpt-4-turbo-preview", 
            "gemini-pro", "mistralai/Mistral-7B-Instruct-v0.2"].            
        api_key (str): google_api_key or openai_api_key or huggingfacehub_api_token 
        temperature (float): Range: 0.0 - 1.0; default = 0.5
        top_p (float): : Range: 0.0 - 1.0; default = 1.
    """
    if LLM_provider == "OpenAI":
        llm = ChatOpenAI(
            api_key=api_key,
            model="gpt-3.5-turbo", # in ["gpt-3.5-turbo", "gpt-3.5-turbo-0125", "gpt-4-turbo-preview"]
            temperature=temperature,
            model_kwargs={
                "top_p": top_p
            }
        )
    if LLM_provider == "Google":
        llm = ChatGoogleGenerativeAI(
            google_api_key=api_key,
            model="gemini-pro", # "gemini-pro"
            temperature=temperature,
            top_p=top_p,
            convert_system_message_to_human=True,
            safety_settings={
                HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_NONE,
                HarmCategory.HARM_CATEGORY_HARASSMENT: HarmBlockThreshold.BLOCK_NONE,
                HarmCategory.HARM_CATEGORY_HATE_SPEECH: HarmBlockThreshold.BLOCK_NONE,
                HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: HarmBlockThreshold.BLOCK_NONE,
                HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: HarmBlockThreshold.BLOCK_NONE},
    
        )
    if LLM_provider == "HuggingFace":
        llm = HuggingFaceHub(
            repo_id="mistralai/Mistral-7B-Instruct-v0.2", # "mistralai/Mistral-7B-Instruct-v0.2"
            huggingfacehub_api_token=api_key,
            model_kwargs={
                "temperature":temperature,
                "top_p": top_p,
                "do_sample": True,
                "max_new_tokens":1024
            },
        )
    return llm


def get_environment_variable(key):
    if key in os.environ:
        value = os.environ.get(key)
        print(f"\n[INFO]: {key} retrieved successfully.")
    else :
        print(f"\n[ERROR]: {key} is not found in your environment variables.") 
        value = getpass(f"Insert your {key}")
    return value






def create_memory(model_name='gpt-3.5-turbo',memory_max_token=None):
    """Creates a ConversationSummaryBufferMemory for gpt-3.5-turbo.
    Creates a ConversationBufferMemory for the other models."""
    
    if model_name=="gpt-3.5-turbo":
        if memory_max_token is None:
            memory_max_token = 1024 # max_tokens for 'gpt-3.5-turbo' = 4096
        memory = ConversationSummaryBufferMemory(
            max_token_limit=memory_max_token,
            llm=ChatOpenAI(model_name="gpt-3.5-turbo",openai_api_key=openai_api_key,temperature=0.1),
            return_messages=True,
            memory_key='chat_history',
            output_key="answer",
            input_key="question"
        )
    else:
        memory = ConversationBufferMemory(
            return_messages=True,
            memory_key='chat_history',
            output_key="answer",
            input_key="question",
        )  
    return memory

# Set a small memory_max_token, just to show how older messages are summarized if max_token_limit is exceeded.

memory = create_memory(model_name='gpt-3.5-turbo',memory_max_token=20)

# save context
memory.save_context(
    inputs={"question":"what does DTC stand for?"},
    outputs={"answer":"""Diffuse to Choose (DTC) is a novel diffusion inpainting approach designed for the Vit-All application, 
    which allows users to virtually place any e-commerce item in any setting, ensuring detailed, semantically coherent blending with realistic 
    lighting and shadows. It effectively incorporates fine-grained cues from the reference image into the main U-Net decoder 
    using a secondary U-Net encoder.
    DTC can handle a variety of e-commerce products and can generate images using in-the-wild images & references. 
    It is superior to existing zero-shot personalization methods, especially in preserving the fine-grained details of items."""}
)
memory.save_context(
    inputs={"question":"what does Vit-all stand for?"},
    outputs={"answer":"Virtual Try-All"}
)

memory.load_memory_variables({})






standalone_question_template = """Given the following conversation and a follow up question, 
rephrase the follow up question to be a standalone question, in its original language.\n\n
Chat History:\n{chat_history}\n
Follow Up Input: {question}\n
Standalone question:"""

standalone_question_prompt = PromptTemplate(
    input_variables=['chat_history', 'question'], 
    template=standalone_question_template
)


def answer_template(language="english"):
    """Pass the standalone question along with the chat history and context 
    to the `LLM` wihch will answer"""
    
    template = f"""Answer the question at the end, using only the following context (delimited by <context></context>).
    Your answer must be in the language at the end. 
    
    <context>
    {{chat_history}}
    
    {{context}} 
    </context>
    
    Question: {{question}}
    Language: {language}.
    
    """
    return template

answer_prompt = ChatPromptTemplate.from_template(answer_template())

# invoke the ChatPromptTemplate
answer_prompt.invoke(
    {"question":"plaese give more details about DTC, including its use cases and implementation.",
     "context":[Document(page_content="DTC use cases include...")], # the context is a list of retrieved documents.
     "chat_history":memory.chat_memory}
)



"""
# Instantiate the retriever and the ConversationalRetrievalChain :

retriever_Google = retrieval_blocks(
    create_vectorstore=False,
    LLM_service="Google",
    vectorstore_name="Vit_All_Google_Embeddings",
    retriever_type="Cohere_reranker",
    base_retriever_search_type="similarity", base_retriever_k=12,
    compression_retriever_k=16,
    cohere_api_key=cohere_api_key,cohere_top_n=10,
)


chain_gemini,memory_gemini = custom_ConversationalRetrievalChain(
    llm = instantiate_LLM(
        LLM_provider="Google",api_key=google_api_key,temperature=0.5,model_name="gemini-pro"
    ),
    condense_question_llm = instantiate_LLM(
        LLM_provider="Google",api_key=google_api_key,temperature=0.1,model_name="gemini-pro"),
    retriever=retriever_Google,
    language="english",
    llm_provider="Google",
    model_name="gemini-pro"
)


memory_gemini.clear()
"""


chain = ConversationalRetrievalChain.from_llm(
    condense_question_prompt=standalone_question_prompt,
    combine_docs_chain_kwargs={'prompt': answer_prompt},
    condense_question_llm=instantiate_LLM(
        LLM_provider="Google",api_key=google_api_key,temperature=0.1,
        model_name="gemini-pro"),
    memory=create_memory("gemini-pro"),
    retriever = base_retriever_HF,  #base_retriever_HF
    llm=instantiate_LLM(
        LLM_provider="Google",api_key=google_api_key,temperature=0.5,
        model_name="gemini-pro"),
    chain_type= "stuff",
    verbose= False,
    return_source_documents=True   
)



# let's invoke the chain
response = chain.invoke({"question":"what does Google stand for?"})
print(response)


chain.memory.load_memory_variables({})

follow_up_question = "plaese give more details about it, including its use cases and implementation."

chain.invoke({"question":follow_up_question})['answer']


from langchain.chains import ConversationalRetrievalChain

def create_ConversationalRetrievalChain(
    llm,condense_question_llm,
    retriever,
    chain_type= 'stuff',
    language="english",
    model_name='gpt-3.5-turbo'
):
    """Create a ConversationalRetrievalChain.
    First, it passes the follow-up question along with the chat history to an LLM which rephrases 
    the question and generates a standalone query. 
    This query is then sent to the retriever, which fetches relevant documents (context) 
    and passes them along with the standalone question and chat history to an LLM to answer.
    """
    
    # 1. Define the standalone_question prompt. 
    # Pass the follow-up question along with the chat history to the `condense_question_llm`
    # which rephrases the question and generates a standalone question.

    standalone_question_prompt = PromptTemplate(
        input_variables=['chat_history', 'question'], 
        template="""Given the following conversation and a follow up question, 
rephrase the follow up question to be a standalone question, in its original language.\n\n
Chat History:\n{chat_history}\n
Follow Up Input: {question}\n
Standalone question:""")

    # 2. Define the answer_prompt
    # Pass the standalone question + the chat history + the context (retrieved documents) to the `LLM` wihch will answer
    
    answer_prompt = ChatPromptTemplate.from_template(answer_template(language=language))

    # 3. Add ConversationSummaryBufferMemory for gpt-3.5, and ConversationBufferMemory for the other models
    
    memory = create_memory(model_name)

    # 4. Create the ConversationalRetrievalChain

    chain = ConversationalRetrievalChain.from_llm(
        condense_question_prompt=standalone_question_prompt,
        combine_docs_chain_kwargs={'prompt': answer_prompt},
        condense_question_llm=condense_question_llm,

        memory=memory,
        retriever = retriever,
        llm=llm,

        chain_type= chain_type,
        verbose= False,
        return_source_documents=True    
    )

    print("Conversational retriever chain created successfully!")
    
    return chain,memory





def submit_message(prompt, prompt_template, temperature, max_tokens, context_length, state):
    
    
    history = state['messages']

    if not prompt:
        return gr.update(value=''), [(history[i]['content'], history[i+1]['content']) for i in range(0, len(history)-1, 2)], state
    
    prompt_template = prompt_templates[prompt_template]


    #chain = load_qa_chain(ChatOpenAI(temperature=temperature, max_tokens=max_tokens, model_name="gpt-3.5-turbo"), chain_type="stuff")
    #completion = chain.run(input_documents=docs, question=query)
    

    
    get_empty_state()
    state['content'] = completion
    #state.append(completion.copy())
    
    completion = { "content": completion }
    



    chat_messages = [(prompt_msg['content'], completion['content'])]    
    return '', chat_messages,  state    # total_tokens_used_msg,


def clear_conversation():
    return gr.update(value=None, visible=True), None, "", get_empty_state()







css = """
      #col-container {max-width: 80%; margin-left: auto; margin-right: auto;}
      #chatbox {min-height: 400px;}
      #header {text-align: center;}
      #prompt_template_preview {padding: 1em; border-width: 1px; border-style: solid; border-color: #e0e0e0; border-radius: 4px; min-height: 150px;}
      #total_tokens_str {text-align: right; font-size: 0.8em; color: #666;}
      #label {font-size: 0.8em; padding: 0.5em; margin: 0;}
      .message { font-size: 1.2em; }
      """

with gr.Blocks(css=css) as demo:
    
    #state = gr.State(get_empty_state())


    with gr.Column(elem_id="col-container"):

           
        gr.Markdown("""## Ask questions of *needs assessment* experts,  
                    ## get responses from a *needs assessment experts* version of ChatGPT.  
                    Ask questions of all of them, or pick your expert below.
                    This is a free resource but it does cost us money to run. Unfortunately someone has been abusing this approach.
                    In response, we have had to temporarily turn it off until we can put improve the monitoring. Sorry for the inconvenience.""" ,
                    elem_id="header")
        
        
        with gr.Row():
            with gr.Column():
                chatbot = gr.Chatbot(elem_id="chatbox")
                input_message = gr.Textbox(show_label=False, placeholder="Enter your needs assessment question", visible=True).style(container=False)

                btn_submit = gr.Button("Submit")
                #total_tokens_str = gr.Markdown(elem_id="total_tokens_str")
                btn_clear_conversation = gr.Button("Start New Conversation")
            with gr.Column():
                prompt_template = gr.Dropdown(label="Choose an Expert:", choices=list(prompt_templates.keys()))
                prompt_template_preview = gr.Markdown(elem_id="prompt_template_preview")
                with gr.Accordion("Advanced parameters", open=False):
                    temperature = gr.Slider(minimum=0, maximum=2.0, value=0.7, step=0.1, label="Flexibility", info="Higher = More AI, Lower = More Expert")
                    max_tokens = gr.Slider(minimum=100, maximum=400, value=200, step=1, label="Length of Response.")
                    context_length = gr.Slider(minimum=1, maximum=5, value=2, step=1, label="Context Length", info="Number of previous questions you have asked.")

   
    btn_submit.click(submit_message, [ input_message, prompt_template, temperature, max_tokens, context_length, state], [input_message, chatbot,  state])
    input_message.submit(submit_message, [ input_message, prompt_template, temperature, max_tokens, context_length, state], [input_message, chatbot,  state])
    btn_clear_conversation.click(clear_conversation, [], [input_message, chatbot,  state])
    prompt_template.change(on_prompt_template_change_description, inputs=[prompt_template], outputs=[prompt_template_preview])

    
    demo.load(download_prompt_templates, inputs=None, outputs=[prompt_template], queur=False)


demo.queue(concurrency_count=10)
demo.launch(height='800px')