deployment_app/conversation.py

# from langchain.document_loaders import TextLoader
# import pinecone
# from langchain.vectorstores import Pinecone
# import os
# from transformers import AutoTokenizer, AutoModel
# from langchain.agents.agent_toolkits import create_conversational_retrieval_agent
# from langchain.agents.agent_toolkits import create_retriever_tool
# from langchain.chat_models import ChatOpenAI
# import torch
# from langchain.agents.openai_functions_agent.agent_token_buffer_memory import (AgentTokenBufferMemory)
# from langchain.agents.openai_functions_agent.base import OpenAIFunctionsAgent
# from langchain.schema.messages import SystemMessage
# from langchain.prompts import MessagesPlaceholder
# import gradio as gr
# import time
# from db_func import insert_one
# from global_variable_module import gobal_input, global_output
# import random   


# def get_bert_embeddings(sentence):
#     embeddings = []
#     input_ids = tokenizer.encode(sentence, return_tensors="pt")
#     with torch.no_grad():
#         output = model(input_ids)
#     embedding = output.last_hidden_state[:,0,:].numpy().tolist()
#     return embedding

# model_name = "BAAI/bge-base-en-v1.5"
# model = AutoModel.from_pretrained("/Users/aakashbhatnagar/Documents/masters/ophthal_llm/models/models--BAAI--bge-base-en-v1.5/snapshots/617ca489d9e86b49b8167676d8220688b99db36e")
# tokenizer = AutoTokenizer.from_pretrained("/Users/aakashbhatnagar/Documents/masters/ophthal_llm/models/models--BAAI--bge-base-en-v1.5/snapshots/617ca489d9e86b49b8167676d8220688b99db36e")
# prompt_file = open("prompts/version_2.txt", "r").read()

# pinecone.init(
#     api_key=os.getenv("PINECONE_API_KEY"),  # find at app.pinecone.io
#     environment=os.getenv("PINECONE_ENV"),  # next to api key in console
# )

# index_name = "ophtal-knowledge-base"
# index = pinecone.Index(index_name)
# vectorstore = Pinecone(index, get_bert_embeddings, "text")
# retriever = vectorstore.as_retriever()
# tool = create_retriever_tool(
#     retriever,
#     "search_ophtal-knowledge-base",
#     "Searches and returns documents regarding the ophtal-knowledge-base.",
# )
# tools = [tool]
# system_message = SystemMessage(content=prompt_file)
# memory_key='history'

# llm = ChatOpenAI(openai_api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4", temperature=0.2)
# prompt = OpenAIFunctionsAgent.create_prompt(
#     system_message=system_message,
#     extra_prompt_messages=[MessagesPlaceholder(variable_name=memory_key)],
# )

# agent_executor = create_conversational_retrieval_agent(llm, tools, verbose=False, prompt=prompt)
# user_name = None

# def run(input_):
#     output = agent_executor({"input": input_})
#     output_text = output["output"]
#     source_text = ""
#     doc_text = ""
#     global_input = input_
#     global_output = output_text
        

#     if len(output["intermediate_steps"])>0:
#         documents = output["intermediate_steps"][0][1]
#         sources = []
#         docs = []

#         for doc in documents:
#             if doc.metadata["source"] not in sources:
#                 sources.append(doc.metadata["source"])
#                 docs.append(doc.page_content)

#         for i in range(len(sources)):
#             temp = sources[i].replace('.pdf',  '').replace('.txt', '').replace("AAO", "").replace("2022-2023", "").replace("data/book", "").replace("text", "").replace("  ", " ")
#             source_text += f"{i+1}. {temp}\n"
#             doc_text += f"{i+1}. {docs[i]}\n"

#         # output_text =  f"{output_text} \n\nSources: \n{source_text}\n\nDocuments: \n{doc_text}"
#         # output_text =  f"{output_text}"
    
#     doc_to_insert = {
#         "user": user_name,
#         "input": input_,
#         "output": output_text,
#         "source": source_text,
#         "documents": doc_text
#     }

#     insert_one(doc_to_insert)

#     return output_text

# def make_conversation(message, history):
#     text_ = run(message)
#     for i in range(len(text_)):
#         time.sleep(0.001)
#         yield text_[: i+1]
    
# def auth_function(username, password):
#     user_name = username
#     return username == password

# def random_response(message, accuracy, history):
#     print(type(message))
#     print(message)
#     print(accuracy)
#     out = random.choice(["Yes", "No"])
#     gobal_input = out
#     # open a txt file
#     with open("function hit", "a+") as f:
#         f.write(message)
#     return out

from langchain.document_loaders import TextLoader
import pinecone
from langchain.vectorstores import Pinecone
import os
from transformers import AutoTokenizer, AutoModel
from langchain.agents.agent_toolkits import create_conversational_retrieval_agent
from langchain.agents.agent_toolkits import create_retriever_tool
from langchain.chat_models import ChatOpenAI
import torch
from langchain.agents.openai_functions_agent.agent_token_buffer_memory import (AgentTokenBufferMemory)
from langchain.agents.openai_functions_agent.base import OpenAIFunctionsAgent
from langchain.schema.messages import SystemMessage
from langchain.prompts import MessagesPlaceholder
import gradio as gr
import time
from db_func import insert_one
from langchain.agents import AgentExecutor
import wordninja

def clean_text(text):

    text = text.strip().lower()
    output_paragraph = ' '.join(''.join(text.split()).split(' '))
    words = wordninja.split(output_paragraph)

    return ' '.join(words)


def get_bert_embeddings(sentence):
    embeddings = []
    input_ids = tokenizer.encode(sentence, return_tensors="pt")
    with torch.no_grad():
        output = model(input_ids)
    embedding = output.last_hidden_state[:,0,:].numpy().tolist()
    return embedding

model_name = "BAAI/bge-base-en-v1.5"
model = AutoModel.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
# prompt_file = open("prompts/version_2.txt", "r").read()

pinecone.init(
    api_key=os.getenv("PINECONE_API_KEY"),  # find at app.pinecone.io
    environment=os.getenv("PINECONE_ENV"),  # next to api key in console
)

index_name = "ophtal-knowledge-base"
index = pinecone.Index(index_name)
vectorstore = Pinecone(index, get_bert_embeddings, "text")
retriever = vectorstore.as_retriever()
tool = create_retriever_tool(
    retriever,
    "search_ophtal-knowledge-base",
    "Searches and returns documents regarding the ophtamology from ophtal-knowledge-base.",
)
tools = [tool]
system_message = SystemMessage(content="Your name is Dr. V AI. You are an assistant to Ophthamologists. Your task is to assist them with their questions. Communicate in mecial terms and be professional. You are working with a doctor so do not give the warning that you are an AI agent be confident.")
memory_key='history'

llm = ChatOpenAI(openai_api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4", temperature=0.2)
# llm = ChatOpenAI(openai_api_key="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", model="gpt-4", temperature=0.2)
prompt = OpenAIFunctionsAgent.create_prompt(
    system_message=system_message,
    extra_prompt_messages=[MessagesPlaceholder(variable_name=memory_key)],
)

memory = AgentTokenBufferMemory(memory_key=memory_key, llm=llm, max_token_limit=4000)



# agent_executor = create_conversational_retrieval_agent(llm, tools, verbose=False, prompt=prompt, )

agent = OpenAIFunctionsAgent(llm=llm, tools=tools, prompt=prompt)

agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
    memory=memory,
    verbose=True,
    return_intermediate_steps=True,
    max_iterations = 2
)

# agent_executor.s

user_name = None

def run(input_):
    output = agent_executor({"input": input_})
    output_text = output["output"]
    print(output_text)
    source_text = ""
    doc_text = ""    

    if len(output["intermediate_steps"])>0:
        documents = output["intermediate_steps"][0][1]
        sources = []
        docs = []

        for doc in documents:
            if doc.metadata["source"] not in sources:
                sources.append(doc.metadata["source"])
                docs.append(doc.page_content)

        for i in range(len(sources)):
            temp = sources[i].replace('.pdf',  '').replace('.txt', '').replace("AAO", "").replace("2022-2023", "").replace("data/book", "").replace("text", "").replace("  ", " ")
            source_text += f"{i+1}. {temp}\n"
            doc_text += f"{i+1}. {docs[i]}\n"

        output_text =  f"{output_text} \n\nSources: \n{source_text}\n\nDocuments: \n{doc_text}"
        # output_text =  f"{output_text} \n\nSources: \n{source_text}"
        # output_text =  f"{output_text}"
    
    doc_to_insert = {
        "user": user_name,
        "input": input_,
        "output": output_text,
        "source": source_text,
        "documents": doc_text
    }

    insert_one(doc_to_insert)

    return output_text

def make_conversation(message, history = {}):
    text_ = run(message)
    for i in range(len(text_)):
        time.sleep(0.001)
        yield text_[: i+1]
    
# def auth_function(username, password):
#     user_name = username
#     return username == password