validation.py

from llama_index import LLMPredictor, PromptHelper, StorageContext, ServiceContext,  load_index_from_storage,  SimpleDirectoryReader, GPTVectorStoreIndex
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():

#    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
    max_input_size = 4096
    num_outputs = 512
    max_chunk_overlap = 20
    chunk_size_limit = 600 
    prompt_helper = PromptHelper(max_input_size, num_outputs, chunk_overlap_ratio= 0.1, chunk_size_limit=chunk_size_limit)
#    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.7, model_name="gpt-4", max_tokens=num_outputs))


    #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

iface = gr.Interface(fn=data_querying,
                     inputs=gr.components.Textbox(lines=20, label="Enter your question"),
                     outputs=gr.components.Textbox(lines=25, label="Response", style="height: 400px; overflow-y: scroll;"),  
                     title="Therapy Validation GPT 0.1 pre alpha")

#passes in data directory
index = data_ingestion_indexing("book-validation")
iface.launch(inline=True)