`
- if (replyMsgs.length > 0) {
- footer += `You replied to a message, so you are in a conversation with ${messages.length} messages.`
- sendFooter = true
- }
-
- if (tokenLimit > 4096) {
- footer += ` 🚀 This request used the ${tokenLimit} token limit.`
- sendFooter = true
- }
-
- const sensible = tokenLimit < 4097 ? 2400 : 14000
- if (
- data.usage.total_tokens > sensible &&
- data.usage.total_tokens < tokenLimit
- ) {
- footer += ` 🟠 Warning: ${
- replyMsgs.length > 0 ? 'You have ' : 'In this request, you '
- }used ${
- data.usage.total_tokens
- } tokens of the ${tokenLimit} token limit.`
- sendFooter = true
- }
-
- if (data.usage.total_tokens > tokenLimit - 1) {
- footer += ` 🚨 Error: In this request, you have gone over the ${tokenLimit} token limit, and you may have been cut off. ${
- replyMsgs.length > 0
- ? 'You may want to try again without replying to a message.'
- : 'You may want to shorten your prompt.'
- }`
- sendFooter = true
- }
-
- if (sendFooter) footer += '
'
-
- let filter = new Filter()
-
- //between the text and footer as the acts as new line on ios renderer
- return textData(
- `
${await renderMath(filter.clean(text))}
${
- sendFooter ? footer : ''
- }`
- )
- } catch (error) {
- if (error.data && error.data.error) {
- if (
- error.data.error.code &&
- error.data.error.code === 'context_length_exceeded'
- ) {
- return textData(
- `
⛔️ ErrorYou have gone over the ${tokenLimit} token limit. ${
- replyMsgs.length > 0
- ? 'You may want to try again without replying to a message.'
- : 'You may want to shorten your prompt.'
- }`
- )
- }
- if (error.data.error.message) {
- throw new Error(error.data.error.message)
- }
- }
- throw new Error(error.message)
- }
-}
From d3971a48b3206d3da065af99b6d1001e5f53853e Mon Sep 17 00:00:00 2001
From: M0HID <91326009+M0HID@users.noreply.github.com>
Date: Sat, 22 Jul 2023 15:28:14 +0100
Subject: [PATCH 2/2] Add files via upload
---
src/commands/ai/gpt.js | 222 +++++++++++++++++++++++++++++++++++++++++
1 file changed, 222 insertions(+)
create mode 100644 src/commands/ai/gpt.js
diff --git a/src/commands/ai/gpt.js b/src/commands/ai/gpt.js
new file mode 100644
index 0000000..6adea02
--- /dev/null
+++ b/src/commands/ai/gpt.js
@@ -0,0 +1,222 @@
+import dayjs from 'dayjs'
+import relativeTime from 'dayjs/plugin/relativeTime'
+import { textData } from '../../util/botData'
+import { renderMath } from '../../util/renderMath'
+import { getGPTMessagesFromReply } from '../../util/replyParser'
+import Filter from 'bad-words'
+import { getUserLastUsed, setUserLastUsed } from '../../util/user_ratelimit'
+import { logTokenEvent } from '../../util/user_tokenlog'
+
+export const gpt = async ({
+ msg,
+ args,
+ mArgs,
+ user,
+ config,
+ replyMessage,
+ cmdData,
+ event,
+}) => {
+ const { replyMsgs, userId, userName } = getGPTMessagesFromReply(
+ replyMessage
+ )
+
+ let use16k =
+ mArgs['16k'] ||
+ mArgs['16'] ||
+ mArgs['large'] ||
+ mArgs['bigtokenlimit'] ||
+ (mArgs['1'] && mArgs['6'])
+ let useGPT4 =
+ mArgs['4'] ||
+ mArgs['GPT4'] ||
+ mArgs['gpt4'] ||
+ mArgs['gpt-4'] ||
+ (mArgs['1'] && mArgs['4'] )
+
+ //const parsedQuery = [mArgs['6'] || use16k || [], mArgs._].flat().join(' ')
+ const parsedQuery = [mArgs['6'] || use16k || useGPT4 || [], mArgs._].flat().join(' ')
+
+ // lets just make it use 16k if its longer than 4096, but enforce the ratelimits
+ if (parsedQuery.length > 3800) use16k = true
+ let tokenLimit = 4096
+ if (use16k) {
+ if (cmdData['16k_no_ratelimit'].includes(user.id)) tokenLimit = 16384
+ else {
+ const last_used = await getUserLastUsed(user.id, '16k')
+ if (!last_used) tokenLimit = 16384
+ //6 hours
+ if (Date.now() - last_used > 1000 * 60 * 60 * 6) {
+ tokenLimit = 16384
+ event.waitUntil(setUserLastUsed(user.id, '16k'))
+ } else {
+ dayjs.extend(relativeTime)
+ const can_use = dayjs(last_used)
+ .add(6, 'hour')
+ .fromNow()
+ return textData(
+ `
🚦 Error - Usage limit
You can make a request with the 16,384 token limit again ${can_use}. Bypass this limit by paying £2/month
`
+ )
+ }
+ }
+ }
+
+ let messages = [
+ ...replyMsgs,
+ {
+ role: 'user',
+ content: parsedQuery,
+ },
+ ]
+ if (replyMsgs.length == 0) {
+ messages.push({
+ role: 'system',
+ // "content": "You have some special formatting options. If you want to output any maths or mathematics or math symbols, you should use a LaTeX format, inside 3 curly brackets like this {{{
}}}, for example: you would write x squared or x^2 as: {{{x^2}}}. Please do this for any maths conversations, in place of using normal quotes around letters, even if it is a small term such as x^2."
+ content: `You are a helpful assistant that accurately answers the user's queries. You have some special formatting options - you can use HTML lists/indented lists, and you can have bold text with bold and italics with italics. You can also define large titles using title text or you can also use px values in place of x-large. Please use the relevant formatting when you are writing your responses. The current date and time is ${new Date().toLocaleString()}.`,
+ })
+ }
+ if (replyMsgs.length > 0 && userId && user.id && userId !== user.id) {
+ return textData(
+ `⛔️ Warning
It looks like you are trying to reply to ${
+ userName ? userName.split(' ')[0] : 'somebody else'
+ }'s conversation. If you are trying to start a new conversation, please do not reply to any messages.`
+ )
+ }
+ // only new lines or empty strings
+ if (
+ args.length == 0 ||
+ args.join(' ').length == 0 ||
+ args.join(' ').trim().length == 0
+ ) {
+ return textData(`Text generation command
Generate text using an AI model. Run /gpt [what you want to generate - a prompt] to generate something.
+ Advanced options
+
+ -t or --temp or --temperature: set the temperature of the model-p or --top_p: set the top_p of the model-p or --presence_penalty: set the presence_penalty of the model-f or --frequency_penalty: set the frequency_penalty of the model
`)
+ }
+
+ const temperature = mArgs.temperature || mArgs.temp || mArgs.t || 0.65
+ const top_p = mArgs.top_p || 1
+ const presence_penalty = mArgs.presence_penalty || mArgs.presence || 0
+ const frequency_penalty = mArgs.frequency_penalty || mArgs.frequency || 0
+ console.log(
+ `[gpt openai] ${
+ messages.map(i => i.content).join(' ').length
+ } chars - t${temperature} p${top_p} pp${presence_penalty} fp${frequency_penalty} tl4096`
+ )
+
+ const res = await fetch('https://api.openai.com/v1/chat/completions', {
+ method: 'POST',
+ headers: {
+ 'Content-Type': 'application/json',
+ Authorization: useGPT4 ? `Bearer ${GPT4_KEY}` : `Bearer ${OPENAI_KEY}`,
+ },
+ body: JSON.stringify({
+ model: useGPT4 ? 'gpt-4' : (use16k && tokenLimit < 4097 ? 'gpt-3.5-turbo-16k' : 'gpt-3.5-turbo'),
+ messages,
+ temperature,
+ top_p,
+ presence_penalty,
+ frequency_penalty,
+ }),
+ })
+
+ try {
+ const data = await res.json()
+ if (!res.ok) {
+ let err = new Error(res.statusText)
+ err.data = data
+ throw err
+ }
+ event.waitUntil(
+ logTokenEvent(
+ user.id,
+ user.displayName,
+ data.usage.prompt_tokens,
+ data.usage.completion_tokens,
+ ['gpt', tokenLimit < 4097 ? '4k' : '16k']
+ )
+ )
+ const text = data.choices[0].message.content.trim()
+ messages.push({
+ role: 'assistant',
+ content: text,
+ })
+
+ //You replied to a message, so you are in a conversation with 7 messages. You used: 372 tokens.
+ let sendFooter = false
+ let footer = `
`
+ if (replyMsgs.length > 0) {
+ footer += `You replied to a message, so you are in a conversation with ${messages.length} messages.`
+ sendFooter = true
+ }
+
+ if (tokenLimit > 4096) {
+ footer += ` 🚀 This request used the ${tokenLimit} token limit.`
+ sendFooter = true
+ }
+
+ const sensible = tokenLimit < 4097 ? 2400 : 14000
+ if (
+ data.usage.total_tokens > sensible &&
+ data.usage.total_tokens < tokenLimit
+ ) {
+ footer += ` 🟠 Warning: ${
+ replyMsgs.length > 0 ? 'You have ' : 'In this request, you '
+ }used ${
+ data.usage.total_tokens
+ } tokens of the ${tokenLimit} token limit.`
+ sendFooter = true
+ }
+
+ if (data.usage.total_tokens > tokenLimit - 1) {
+ footer += ` 🚨 Error: In this request, you have gone over the ${tokenLimit} token limit, and you may have been cut off. ${
+ replyMsgs.length > 0
+ ? 'You may want to try again without replying to a message.'
+ : 'You may want to shorten your prompt.'
+ }`
+ sendFooter = true
+ }
+
+ if (sendFooter) footer += '
'
+
+ let filter = new Filter()
+
+ //between the text and footer as the acts as new line on ios renderer
+ return textData(
+ `
${await renderMath(filter.clean(text))}
${
+ sendFooter ? footer : ''
+ }`
+ )
+ } catch (error) {
+ if (error.data && error.data.error) {
+ if (
+ error.data.error.code &&
+ error.data.error.code === 'context_length_exceeded'
+ ) {
+ return textData(
+ `
⛔️ ErrorYou have gone over the ${tokenLimit} token limit. ${
+ replyMsgs.length > 0
+ ? 'You may want to try again without replying to a message.'
+ : 'You may want to shorten your prompt.'
+ }`
+ )
+ }
+ if (error.data.error.message) {
+ throw new Error(error.data.error.message)
+ }
+ }
+ throw new Error(error.message)
+ }
+}