Handling text-davinci-003 lifespan >0 in dialogflow

API
1

We are currently working with the OpenAI GPT-3 model and setting it up with a webhook code hosted on repl.it. Our goal is to have a chatbot that is capable of maintaining context throughout the conversation. However, we have encountered an issue with the lifespan of the chatbot and would appreciate some guidance on the matter.

We are currently working with the davinci model and have noticed that it does not seem to maintain context throughout the conversation. We would like to know if the lifespan of the davinci model can be set to a value greater than 0, and if so, how this can be done.

We have written the following webhook code to handle our chatbot, and we would like to know if it needs adjusting in any way to better handle lifespan:

// medium hook 
const express = require("express");
require("actions-on-google")
// require('dotenv').config();
const axios = require('axios');
const { WebhookClient } = require("dialogflow-fulfillment");
const app = express();
app.post("/dialogflow", express.json(), (req, res) => {
    const agent = new WebhookClient({ request: req, response: res});
    let intentMap = new Map();
    intentMap.set("Default Welcome Intent", welcome);
    intentMap.set("Default Fallback Intent", queryGPT);
    agent.handleRequest(intentMap);
  
    function welcome(agent) {
      agent.add('Hi');
  }
  
  async function queryGPT(agent) {
      // agent.add('Sorry! I am unable to understand this at the moment. I am still learning humans. You can pick any of the service that might help me.');
      const instance = axios.create({
        baseURL: 'https://api.openai.com/v1/',
        headers: { Authorization: `Bearer ${process.env.OPENAI_API_KEY}` },

        model: "text-davinci-003"
        
      });

    
      const dialog = [
        `The following is a conversation with an AI assistant. I can answer your questions and provide information. 
AI: Hello, How can I help you today?`,
      ];
      let query = agent.query;
      console.log('querytext ', query)
      dialog.push(`User: ${query}`);
      dialog.push('AI:');
      // agent.add(`you said ${query}`)
    
      const completionParmas = {
        prompt: dialog.join('\n'),
        max_tokens: 60,
        temperature: 0.85,
        n: 1,
        stream: false,
        logprobs: null,
        echo: false,
        stop: '\n',
      };
      
    
      try {
        const result = await instance.post('/engines/davinci/completions', completionParmas);
        const botResponse = result.data.choices[0].text.trim();
        agent.add(botResponse);
      } catch (err) {
        console.log(err);
        agent.add('Sorry. Something went wrong. Can you say that again?');
      }
    
  }
});
const port = 3000;
app.listen(port, () => console.log(`App listening on port ${port}!`))

Thank you.

2

This is a question that is asked and answered a lot here, so its a good idea to search the forums before posting a question to get the benefits of all the prior Q&A on this topic @TOTAL .

The bottom line is that it is “up to you” when using the OpenAI API to manage the session layer, including feeding backs prior prompts and completions to maintain state, managing token and word counts, etc. when using the API. This is performed by the app developers of ChatGPT, not the OpenAI API.

Also, if you wait for the “coming soon” ChatGPT API, then many people guess this session management may be available in that “new to be released” API.

HTH

(and don’t forget to search the forums here for these types of questions and answers)