gpt-4 wrapper

gpt4client.js

const ws = require('ws')

// just trivially obfuscating the endpoint url to avoid spiders
function rot13(str) {
  return str.replace(/[a-zA-Z]/g, (c) => {
    const base = c <= 'Z' ? 'A'.charCodeAt(0) : 'a'.charCodeAt(0)
    return String.fromCharCode(((c.charCodeAt(0) - base + 13) % 26) + base)
  })
}

async function requestGpt4Response(systemPrompt, userPrompt, topP = 1, temperature = 0.85) {
  return new Promise((resolve, reject) => {
    const socket = new ws(`wss://${rot13('lfunezn-pungtcg4')}.hf.space/queue/join`)

    let spinnerIndex = 0
    let generatedTokens = 0
    const spinnerChars = ['|', '/', '-', '\\']

    const session_hash = Math.random().toString(36).slice(2)

    function displaySpinner() {
      process.stdout.write(
        '\r' +
          spinnerChars[spinnerIndex] +
          ` Processing... (${generatedTokens} tokens)` +
          ' '.repeat(15) +
          '\r'
      )
      spinnerIndex = (spinnerIndex + 1) % spinnerChars.length
    }

    socket.on('open', () => {
      console.log('Socket connection opened')
    })

    socket.on('message', (event) => {
      const parsed = JSON.parse(event)
      switch (parsed.msg) {
        case 'send_hash':
          console.log('Sending session hash', session_hash)
          socket.send(JSON.stringify({ fn_index: 0, session_hash }))
          break
        case 'estimation':
          process.stdout.write(
            `\rQueue size: ${parsed.queue_size} Position: ${parsed.rank}${' '.repeat(5)}\r`
          )
          break
        case 'queue_full':
          socket.close()
          reject(new Error('Queue is full'))
          break
        case 'send_data':
          const payload = JSON.stringify({
            fn_index: 1,
            data: [systemPrompt, userPrompt, topP, temperature, 0, [], null],
            session_hash,
            event_data: null, // not sure what this is
          })
          const sysTruncated =
            systemPrompt.length > 20 ? systemPrompt.slice(0, 20) + '...' : systemPrompt
          const userTruncated =
            userPrompt.length > 20 ? userPrompt.slice(0, 20) + '...' : userPrompt
          console.log('Sending the following prompts:', {
            system: sysTruncated,
            user: userTruncated,
          })
          socket.send(payload)
          break
        case 'process_starts':
          console.log('Acknowledged prompt')
          break
        case 'process_completed':
          if (parsed.output.error || parsed.success === false) {
            socket.close()
            const unknownError = 'Unknown error (prompt too long?)'
            reject(new Error(parsed.output.error ? parsed.output.error : unknownError))
          } else {
            const response = parsed.output.data[0][0][1]
            console.log('Raw response from GPT4:', response)
            resolve(response)
          }
          break
        case 'process_generating':
          generatedTokens++
          displaySpinner()
          break
        default:
          console.log('Unknown message from server', parsed)
          break
      }
    })

    socket.on('close', (event) => {
      console.log(`Socket connection closed with code ${event}`)
    })

    socket.on('error', (error) => {
      console.log(`Socket error: ${error.message}`)
      reject(error)
    })
  })
}

module.exports = requestGpt4Response