Text-davinci-003 deprecated

API Deprecations
1

Hello, I need help with changing the model because when I run the website it give me the error openai.error.InvalidRequestError: The model text-davinci-003 has been deprecated, learn more here: https://platform.openai.com/docs/deprecations
I tried implementing the gpt-3.5-turbo-instruct but it doesn’t work, my code is quite long but I’ll show the snippet where I assume the error is coming from and my imports:

from flask import Flask, render_template, request, redirect, url_for, session, Response
import os
from PyPDF2 import PdfReader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import FAISS
from langchain.llms import OpenAI
from langchain.chains.question_answering import load_qa_chain
from langchain.callbacks import get_openai_callback
import io
import boto3
import botocore
import uuid
from werkzeug.security import check_password_hash, generate_password_hash

@app.route(‘/continuous_questioning’, methods=[‘GET’, ‘POST’])
def continuous_questioning():
# Retrieve the current user ID using session cookie
user_id = session.get(‘user_id’, None)
if user_id is None:
return redirect(url_for(‘index’))

# Retrieve user-specific data from session
pdf_key = session.get('pdf_key', None)

# Retrieve user-specific responses from DynamoDB
responses = get_user_responses(user_id)

# Initialize a flag to indicate if the answer is processing
answer_processing = False

if request.method == 'POST':
    query = request.form['query']

    if query and pdf_key:
        # Set the flag to indicate that the answer is processing
        answer_processing = True

        # Download PDF file from S3
        pdf_content = download_file_from_s3(pdf_key)

        if pdf_content:
            # Read the PDF file
            with io.BytesIO(pdf_content) as pdf_file:
                pdf_reader = PdfReader(pdf_file)
                text = ""
                for page in pdf_reader.pages:
                    text += page.extract_text()

                text_splitter = RecursiveCharacterTextSplitter(
                    chunk_size=1000,
                    chunk_overlap=200,
                    length_function=len
                )
                chunks = text_splitter.split_text(text=text)

                # Replace VectorStore with the appropriate vector store or embedding mechanism
                # based on your specific use case
                embeddings = OpenAIEmbeddings(model="gpt-3.5-turbo")
                VectorStore = FAISS.from_texts(chunks, embedding=embeddings)

                docs = VectorStore.similarity_search(query=query, k=3)

                llm = OpenAI()
                chain = load_qa_chain(llm=llm, chain_type="stuff")
                with get_openai_callback() as cb:
                    response = chain.run(input_documents=docs, question=query)

                    # Append the question and its answer as a pair to user-specific responses
                    responses.append({'question': query, 'answer': response})

                    # Update user-specific responses in DynamoDB
                    set_user_responses(user_id, responses)

        return render_template('continuous_questioning.html', responses=responses, answer_processing=answer_processing)

return render_template('continuous_questioning.html', responses=responses, answer_processing=answer_processing)

Please update my code with the new model because this project is for my thesis that has to be submitted in 3 days.

2

I face similar challenges with my code. Any assistance to upgrade.

3

Requesting assistance too. We don’t have a clear alternative now that text-davinci-003 is deprecated. We’ve tried a few alternatives but none have worked as good.

4

To drop into your existing code that was using text-davinci-002 or text-davinci-003 (which can follow instructions), you can just specify the new model gpt-3.5-turbo-instruct.

Your prompt and application may need adjustment, because the model has different skills. However, no code change is necessary.

OpenAI python libraries have recently been significantly changed in November, and especially with langchain, you need a fully-compatible stack and calling code. You can revert to openai < 1.0.0 and langchain version that was previously working before half-hearted attempts to update.

A major error was made in the code shown.

OpenAIEmbeddings(model="gpt-3.5-turbo")

The OpenAI model for embeddings is text-embedding-ada-002, as has been available for a year. 1536 dimensions.

For advanced applications, refactoring using GPT-4 on the chat completion endpoint and its messages format should be the way forward. If not chatting with a user, the purpose and role AI serves should be clearly defined in system message.

5 Likes
5

This suggestion solved my problem, thank you

1 Like
6

llm = OpenAI(model_name=“gpt-3.5-turbo-instruct”)
use this we have to specify model name

7

llm = OpenAI(model_name=“gpt-3.5-turbo-instruct”)
add this in code . if you dont no where and how to add simply copy your code put into chatgpt and tell them to add this code in my code llm = OpenAI(model_name=“gpt-3.5-turbo-instruct”) it will adjust your code

8

As other folks have mentioned, gpt-3.5-turbo-instruct is the replacement for the now shut down completion models. Let me know if you run into any issues getting started with the new model.

2 Likes
9

The emotional intelligence of DaVinci-003 is way higher than gpt-3.5-turbo. Is there any way of still using it?
See paper: Emotional intelligence of Large Language Models Xuena Wang, Xueting Li, Zi Yin, Yue Wu, Jia Liu

10

New deployments of text-davinci-003 on Microsoft Azure were disabled in July 2023, however, those that are in operation can continue through July 5 2024. So there may be some still with model access.

Unless you find someone who still has the model as a customer-facing application, there is no way of still using it.

1 Like