kb.py

from llama_index import SimpleDirectoryReader, LLMPredictor, PromptHelper, StorageContext, ServiceContext, GPTVectorStoreIndex, load_index_from_storage
from langchain.chat_models import ChatOpenAI
import gradio as gr
import sys
import os
import openai
from ratelimit import limits, sleep_and_retry
from langchain import HuggingFaceHub


# fixing bugs
# 1. open ai key: https://stackoverflow.com/questions/76425556/tenacity-retryerror-retryerrorfuture-at-0x7f89bc35eb90-state-finished-raised
# 2. rate limit error in lang_chain default version - install langchain==0.0.188. https://github.com/jerryjliu/llama_index/issues/924
# 3. added true Config variable in langchain: https://github.com/pydantic/pydantic/issues/3320
# 4. deploy on huggingfaces https://huggingface.co/welcome
#   create huggingfaces token https://huggingface.co/settings/tokens
#   login: huggingface-cli login
#   add requirements.txt file  https://huggingface.co/docs/hub/spaces-dependencies

os.environ["OPENAI_API_KEY"] = os.environ.get("openai_key")
openai.api_key = os.environ["OPENAI_API_KEY"]

# Define the rate limit for API calls (requests per second)
RATE_LIMIT = 3

# Implement the rate limiting decorator
@sleep_and_retry
@limits(calls=RATE_LIMIT, period=1)
def create_service_context():

    #constraint parameters
    max_input_size = 4096
    num_outputs = 512
    max_chunk_overlap = 20
    chunk_size_limit = 600

    #allows the user to explicitly set certain constraint parameters
    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)

    #LLMPredictor is a wrapper class around LangChain's LLMChain that allows easy integration into LlamaIndex
    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.5, model_name="gpt-3.5-turbo", max_tokens=num_outputs))

    #constructs service_context
    service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
    return service_context


# Implement the rate limiting decorator
@sleep_and_retry
@limits(calls=RATE_LIMIT, period=1)
def data_ingestion_indexing(directory_path):

    #loads data from the specified directory path
    documents = SimpleDirectoryReader(directory_path).load_data()

    #when first building the index
    index = GPTVectorStoreIndex.from_documents(
        documents, service_context=create_service_context()
    )

    #persist index to disk, default "storage" folder
    index.storage_context.persist()

    return index

def data_querying(input_text):

    #rebuild storage context
    storage_context = StorageContext.from_defaults(persist_dir="./storage")

    #loads index from storage
    index = load_index_from_storage(storage_context, service_context=create_service_context())

    #queries the index with the input text
    response = index.as_query_engine().query(input_text)

    return response.response
    
with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    msg = gr.Textbox()

    def respond(message, chat_history):
        original_message = message
        message = "As a therapy chatbot designed specifically to assist teenagers and young adults\
please provide a thorough and detailed response by explaining your capabilities\
features, and methods for helping individuals in this age group\
In order to receive the most precise, comprehensive and high-quality response\
please provide your answer while keeping in mind the following guidelines: \n1. \
Make sure your response is prompt, without unnecessary delays.\n2. \
Aim for perfection by providing accurate and well-thought-out information. \n3. \
Maintain a supportive and social tone, similar to that of a human conversation\
keep all responses at maximum 30 words when users tell you how they feel give them advice\n4. \
Never say \"Based on the given context information\"\n \
The patient asks you the following question: " + message
        bot_message = data_querying(message)
        chat_history.append((original_message, bot_message))
#        time.sleep(1)
        return "", chat_history

    msg.submit(respond, [msg, chatbot], [msg, chatbot])

#iface = gr.Interface(fn=data_querying,
#                     inputs=gr.components.Textbox(lines=7, label="Enter your question"),
#                     outputs="text",
#                     title="Therapy GPT 0.1 pre alpha")

#passes in data directory
index = data_ingestion_indexing("therapy")

#iface.launch(inline=True)
demo.launch()