Andreas/LibreChat
1
0
Fork 0
mirror of https://github.com/danny-avila/LibreChat.git synced 2025-12-20 10:20:15 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Minor fixes: tokenizer, default Bing toneStyle, SiblingSwitch (#348)

Browse source 
* fix: tokenizer will count completion tokens correctly, remove global var, will allow unofficial models for alternative endpoints

* refactor(askBingAI.js, Settings.jsx, types.ts, cleanupPreset.js, getDefaultConversation.js, handleSubmit.js): change default toneStyle to 'creative' instead of 'fast' for Bing AI endpoint.

* fix(SiblingSwitch): correctly appears now
style(HoverButtons.jsx): add 'active' class to hover buttons
This commit is contained in:
Danny Avila 2023-05-21 12:43:06 -04:00 • committed by GitHub
parent 791b515937
commit 4beb06aa4b
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
10 changed files with 125 additions and 78 deletions
Download patch file Download diff file Expand all files Collapse all files

33
api/app/clients/chatgpt-client.js
View file

@ -2,6 +2,7 @@ require('dotenv').config();
const { KeyvFile } = require('keyv-file');
const { genAzureEndpoint } = require('../../utils/genAzureEndpoints');
const tiktoken = require('@dqbd/tiktoken');
const tiktokenModels = require('../../utils/tiktokenModels');
const encoding_for_model = tiktoken.encoding_for_model;
const askClient = async ({
@ -26,9 +27,8 @@ const askClient = async ({
};
const azure = process.env.AZURE_OPENAI_API_KEY ? true : false;
if (promptPrefix == null) {
promptText = 'You are ChatGPT, a large language model trained by OpenAI.';
} else {
let promptText = 'You are ChatGPT, a large language model trained by OpenAI.';
if (promptPrefix) {
promptText = promptPrefix;
}
const maxContextTokens = model === 'gpt-4' ? 8191 : model === 'gpt-4-32k' ? 32767 : 4095; // 1 less than maximum
@ -68,25 +68,18 @@ const askClient = async ({
...(parentMessageId && conversationId ? { parentMessageId, conversationId } : {})
};
const enc = encoding_for_model(model);
const text_tokens = enc.encode(text);
const prompt_tokens = enc.encode(promptText);
// console.log("Prompt tokens = ", prompt_tokens.length);
// console.log("Message Tokens = ", text_tokens.length);
const enc = encoding_for_model(tiktokenModels.has(model) ? model : 'gpt-3.5-turbo');
const usage = {
prompt_tokens: (enc.encode(promptText)).length + (enc.encode(text)).length,
}
const res = await client.sendMessage(text, { ...options, userId });
// return res;
// create a new response object that includes the token counts
const newRes = {
usage.completion_tokens = (enc.encode(res.response)).length;
usage.total_tokens = usage.prompt_tokens + usage.completion_tokens;
return {
...res,
usage: {
prompt_tokens: prompt_tokens.length,
completion_tokens: text_tokens.length,
total_tokens: prompt_tokens.length + text_tokens.length
}
};
return newRes;
usage,
}
};
module.exports = { askClient };