app_module/configuration.py

from llama_index import LLMPredictor, PromptHelper, ServiceContext, StorageContext, OpenAIEmbedding, set_global_service_context, load_index_from_storage
from llama_index.langchain_helpers.text_splitter import TokenTextSplitter
from llama_index.node_parser import SimpleNodeParser
from llama_index.callbacks import CallbackManager, LlamaDebugHandler, TokenCountingHandler
from llama_index.vector_stores import PineconeVectorStore
from llama_index.storage.docstore import MongoDocumentStore
from llama_index.storage.index_store.mongo_index_store import MongoIndexStore
from langchain.chat_models import ChatOpenAI
import tiktoken
import openai
import pinecone
import sys
import os
import logging

_pinecode_index_name = os.environ['PINECONE_INDEX_NAME']
_pinecone_api_key = os.environ['PINECONE_API_KEY']
_pinecone_environment = os.environ['PINECONE_ENVIRONMENT']
_openai_api_key = os.environ['OPENAI_API_KEY']
_mongo_uri = os.environ['MONGO_URI']
_model = 'gpt-3.5-turbo-16k'

# configure logging
logging.basicConfig(
    stream=sys.stdout,
    level=logging.DEBUG,
    format="%(asctime)s [%(levelname)s] [%(filename)s:%(lineno)d] %(message)s",
)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

# initialize pinecone and openai
openai.api_key = _openai_api_key
pinecone.init(api_key=_pinecone_api_key, environment=_pinecone_environment)

# Creating a Pinecone index
if _pinecode_index_name not in pinecone.list_indexes():
    pinecone.create_index(
        _pinecode_index_name,
        dimension=1536,
        metric="euclidean",
        pod_type="p1"
    )
pinecode_index = pinecone.Index(_pinecode_index_name)

# construct vector store
vector_store = PineconeVectorStore(pinecone_index=pinecode_index)

# construct document store
docstore = MongoDocumentStore.from_uri(uri=_mongo_uri)

# construct index store
index_store = MongoIndexStore.from_uri(uri=_mongo_uri)

# construct callback manager
token_counter = TokenCountingHandler(
    tokenizer=tiktoken.encoding_for_model(_model).encode,
    verbose=False  # set to true to see usage printed to the console
)
llama_debug = LlamaDebugHandler(print_trace_on_end=True)
callback_manager = CallbackManager([
    llama_debug,
    token_counter,
])

# construct embed model
embed_model = OpenAIEmbedding()

# construct prompt helper
context_window = 4096
num_outputs = 512
chunk_overlap_ratio = 0.1
chunk_size_limit = 600
prompt_helper = PromptHelper(
    context_window, num_outputs, chunk_overlap_ratio, chunk_size_limit)

# construct llm predictor
llm = ChatOpenAI(temperature=0.7, model=_model,
                 max_tokens=num_outputs)
llm_predictor = LLMPredictor(llm=llm)

# construct node parser
node_parser = SimpleNodeParser(
    text_splitter=TokenTextSplitter(chunk_size=1024, chunk_overlap=20))

# create storage context
logging.info("Constructing service context")
service_context = ServiceContext.from_defaults(
    llm_predictor=llm_predictor,
    embed_model=embed_model,
    node_parser=node_parser,
    prompt_helper=prompt_helper,
    callback_manager=callback_manager,

)


logging.info("Constructing storage context")
storage_context = StorageContext.from_defaults(
    vector_store=vector_store,
    docstore=docstore,
    index_store=index_store,
)

logging.info("Setting global service context")
set_global_service_context(service_context)

def get_index():
    logging.info("Loading index")
    index = load_index_from_storage(storage_context)
    logging.info("Loaded index")

    return index