345 lines
11 KiB
TypeScript
345 lines
11 KiB
TypeScript
import type { Tiktoken } from "@dqbd/tiktoken";
|
|
import type { Tokenizer } from "@mlc-ai/web-tokenizers";
|
|
import { DataBase, type character } from "./storage/database";
|
|
import { get } from "svelte/store";
|
|
import type { MultiModal, OpenAIChat } from "./process";
|
|
import { supportsInlayImage } from "./process/files/image";
|
|
import { risuChatParser } from "./parser";
|
|
import { tokenizeGGUFModel } from "./process/models/local";
|
|
import { globalFetch } from "./storage/globalApi";
|
|
|
|
|
|
export const tokenizerList = [
|
|
['tik', 'Tiktoken (OpenAI)'],
|
|
['mistral', 'Mistral'],
|
|
['novelai', 'NovelAI'],
|
|
['claude', 'Claude'],
|
|
['llama', 'Llama'],
|
|
['llama3', 'Llama3'],
|
|
['novellist', 'Novellist'],
|
|
['gemma', 'Gemma'],
|
|
] as const
|
|
|
|
export async function encode(data:string):Promise<(number[]|Uint32Array|Int32Array)>{
|
|
let db = get(DataBase)
|
|
if(db.aiModel === 'openrouter' || db.aiModel === 'reverse_proxy'){
|
|
switch(db.customTokenizer){
|
|
case 'mistral':
|
|
return await tokenizeWebTokenizers(data, 'mistral')
|
|
case 'llama':
|
|
return await tokenizeWebTokenizers(data, 'llama')
|
|
case 'novelai':
|
|
return await tokenizeWebTokenizers(data, 'novelai')
|
|
case 'claude':
|
|
return await tokenizeWebTokenizers(data, 'claude')
|
|
case 'novellist':
|
|
return await tokenizeWebTokenizers(data, 'novellist')
|
|
case 'llama3':
|
|
return await tokenizeWebTokenizers(data, 'llama')
|
|
case 'gemma':
|
|
return await tokenizeWebTokenizers(data, 'gemma')
|
|
default:
|
|
// Add exception for gpt-4o tokenizers on reverse_proxy
|
|
if(db.proxyRequestModel?.startsWith('gpt4o') ||
|
|
(db.proxyRequestModel === 'custom' && db.customProxyRequestModel.startsWith('gpt-4o'))) {
|
|
return await tikJS(data, 'o200k_base')
|
|
}
|
|
return await tikJS(data)
|
|
}
|
|
}
|
|
if(db.aiModel.startsWith('novellist')){
|
|
const nv= await tokenizeWebTokenizers(data, 'novellist')
|
|
return nv
|
|
}
|
|
if(db.aiModel.startsWith('claude')){
|
|
return await tokenizeWebTokenizers(data, 'claude')
|
|
}
|
|
if(db.aiModel.startsWith('novelai')){
|
|
return await tokenizeWebTokenizers(data, 'novelai')
|
|
}
|
|
if(db.aiModel.startsWith('mistral')){
|
|
return await tokenizeWebTokenizers(data, 'mistral')
|
|
}
|
|
if(db.aiModel === 'mancer' ||
|
|
db.aiModel === 'textgen_webui' ||
|
|
(db.aiModel === 'reverse_proxy' && db.reverseProxyOobaMode)){
|
|
return await tokenizeWebTokenizers(data, 'llama')
|
|
}
|
|
if(db.aiModel.startsWith('local_')){
|
|
return await tokenizeGGUFModel(data)
|
|
}
|
|
if(db.aiModel === 'ooba'){
|
|
if(db.reverseProxyOobaArgs.tokenizer === 'mixtral' || db.reverseProxyOobaArgs.tokenizer === 'mistral'){
|
|
return await tokenizeWebTokenizers(data, 'mistral')
|
|
}
|
|
else if(db.reverseProxyOobaArgs.tokenizer === 'llama'){
|
|
return await tokenizeWebTokenizers(data, 'llama')
|
|
}
|
|
else{
|
|
return await tokenizeWebTokenizers(data, 'llama')
|
|
}
|
|
}
|
|
if(db.aiModel.startsWith('gpt4o')){
|
|
return await tikJS(data, 'o200k_base')
|
|
}
|
|
if(db.aiModel.startsWith('gemini')){
|
|
return await tokenizeWebTokenizers(data, 'gemma')
|
|
}
|
|
if(db.aiModel.startsWith('cohere')){
|
|
return await tokenizeWebTokenizers(data, 'cohere')
|
|
}
|
|
|
|
return await tikJS(data)
|
|
}
|
|
|
|
type tokenizerType = 'novellist'|'claude'|'novelai'|'llama'|'mistral'|'llama3'|'gemma'|'cohere'
|
|
|
|
let tikParser:Tiktoken = null
|
|
let tokenizersTokenizer:Tokenizer = null
|
|
let tokenizersType:tokenizerType = null
|
|
let lastTikModel = 'cl100k_base'
|
|
|
|
async function tikJS(text:string, model='cl100k_base') {
|
|
if(!tikParser || lastTikModel !== model){
|
|
if(model === 'cl100k_base'){
|
|
const {Tiktoken} = await import('@dqbd/tiktoken')
|
|
const cl100k_base = await import("@dqbd/tiktoken/encoders/cl100k_base.json");
|
|
lastTikModel = model
|
|
|
|
tikParser = new Tiktoken(
|
|
cl100k_base.bpe_ranks,
|
|
cl100k_base.special_tokens,
|
|
cl100k_base.pat_str
|
|
);
|
|
}
|
|
if(model === 'o200k_base'){
|
|
const {Tiktoken} = await import('@dqbd/tiktoken')
|
|
const o200k_base = await import("src/etc/o200k_base.json");
|
|
lastTikModel = model
|
|
tikParser = new Tiktoken(
|
|
o200k_base.bpe_ranks,
|
|
o200k_base.special_tokens,
|
|
o200k_base.pat_str
|
|
);
|
|
}
|
|
}
|
|
return tikParser.encode(text)
|
|
}
|
|
|
|
async function geminiTokenizer(text:string) {
|
|
const db = get(DataBase)
|
|
const fetchResult = await globalFetch(`https://generativelanguage.googleapis.com/v1beta/${db.aiModel}:countTextTokens`, {
|
|
"headers": {
|
|
"content-type": "application/json",
|
|
"authorization": `Bearer ${db.google.accessToken}`
|
|
},
|
|
"body": JSON.stringify({
|
|
"prompt":{
|
|
text: text
|
|
}
|
|
}),
|
|
"method": "POST"
|
|
})
|
|
|
|
if(!fetchResult.ok){
|
|
//fallback to tiktoken
|
|
return await tikJS(text)
|
|
}
|
|
|
|
const result = fetchResult.data
|
|
|
|
return result.tokenCount ?? 0
|
|
}
|
|
|
|
async function tokenizeWebTokenizers(text:string, type:tokenizerType) {
|
|
if(type !== tokenizersType || !tokenizersTokenizer){
|
|
const webTokenizer = await import('@mlc-ai/web-tokenizers')
|
|
switch(type){
|
|
case "novellist":
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromSentencePiece(
|
|
await (await fetch("/token/trin/spiece.model")
|
|
).arrayBuffer())
|
|
break
|
|
case "claude":
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromJSON(
|
|
await (await fetch("/token/claude/claude.json")
|
|
).arrayBuffer())
|
|
break
|
|
case 'llama3':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromJSON(
|
|
await (await fetch("/token/llama/llama3.json")
|
|
).arrayBuffer())
|
|
break
|
|
case 'cohere':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromJSON(
|
|
await (await fetch("/token/cohere/tokenizer.json")
|
|
).arrayBuffer())
|
|
break
|
|
case 'novelai':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromSentencePiece(
|
|
await (await fetch("/token/nai/nerdstash_v2.model")
|
|
).arrayBuffer())
|
|
|
|
break
|
|
case 'llama':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromSentencePiece(
|
|
await (await fetch("/token/llama/llama.model")
|
|
).arrayBuffer())
|
|
break
|
|
case 'mistral':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromSentencePiece(
|
|
await (await fetch("/token/mistral/tokenizer.model")
|
|
).arrayBuffer())
|
|
break
|
|
case 'gemma':
|
|
tokenizersTokenizer = await webTokenizer.Tokenizer.fromSentencePiece(
|
|
await (await fetch("/token/gemma/tokenizer.model")
|
|
).arrayBuffer())
|
|
break
|
|
|
|
}
|
|
tokenizersType = type
|
|
}
|
|
return (tokenizersTokenizer.encode(text))
|
|
}
|
|
|
|
export async function tokenizerChar(char:character) {
|
|
const encoded = await encode(char.name + '\n' + char.firstMessage + '\n' + char.desc)
|
|
return encoded.length
|
|
}
|
|
|
|
export async function tokenize(data:string) {
|
|
const encoded = await encode(data)
|
|
return encoded.length
|
|
}
|
|
|
|
export async function tokenizeAccurate(data:string, consistantChar?:boolean) {
|
|
data = risuChatParser(data.replace('{{slot}}',''), {
|
|
tokenizeAccurate: true,
|
|
consistantChar: consistantChar,
|
|
})
|
|
const encoded = await encode(data)
|
|
return encoded.length
|
|
}
|
|
|
|
|
|
export class ChatTokenizer {
|
|
|
|
private chatAdditonalTokens:number
|
|
private useName:'name'|'noName'
|
|
|
|
constructor(chatAdditonalTokens:number, useName:'name'|'noName'){
|
|
this.chatAdditonalTokens = chatAdditonalTokens
|
|
this.useName = useName
|
|
}
|
|
async tokenizeChat(data:OpenAIChat) {
|
|
let encoded = (await encode(data.content)).length + this.chatAdditonalTokens
|
|
if(data.name && this.useName ==='name'){
|
|
encoded += (await encode(data.name)).length + 1
|
|
}
|
|
if(data.multimodals && data.multimodals.length > 0){
|
|
for(const multimodal of data.multimodals){
|
|
encoded += await this.tokenizeMultiModal(multimodal)
|
|
}
|
|
}
|
|
return encoded
|
|
}
|
|
|
|
async tokenizeMultiModal(data:MultiModal){
|
|
const db = get(DataBase)
|
|
if(!supportsInlayImage()){
|
|
return this.chatAdditonalTokens
|
|
}
|
|
if(db.gptVisionQuality === 'low'){
|
|
return 87
|
|
}
|
|
|
|
let encoded = this.chatAdditonalTokens
|
|
let height = data.height ?? 0
|
|
let width = data.width ?? 0
|
|
|
|
if(height === width){
|
|
if(height > 768){
|
|
height = 768
|
|
width = 768
|
|
}
|
|
}
|
|
else if(height > width){
|
|
if(width > 768){
|
|
width = 768
|
|
height = height * (768 / width)
|
|
}
|
|
}
|
|
else{
|
|
if(height > 768){
|
|
height = 768
|
|
width = width * (768 / height)
|
|
}
|
|
}
|
|
|
|
const chunkSize = Math.ceil(width / 512) * Math.ceil(height / 512)
|
|
encoded += chunkSize * 2
|
|
encoded += 85
|
|
|
|
return encoded
|
|
}
|
|
|
|
}
|
|
|
|
export async function tokenizeNum(data:string) {
|
|
const encoded = await encode(data)
|
|
return encoded
|
|
}
|
|
|
|
export async function strongBan(data:string, bias:{[key:number]:number}) {
|
|
|
|
if(localStorage.getItem('strongBan_' + data)){
|
|
return JSON.parse(localStorage.getItem('strongBan_' + data))
|
|
}
|
|
const performace = performance.now()
|
|
const length = Object.keys(bias).length
|
|
let charAlt = [
|
|
data,
|
|
data.trim(),
|
|
data.toLocaleUpperCase(),
|
|
data.toLocaleLowerCase(),
|
|
data[0].toLocaleUpperCase() + data.slice(1),
|
|
data[0].toLocaleLowerCase() + data.slice(1),
|
|
]
|
|
|
|
let banChars = " !\"#$%&'()*+,-./:;<=>?@[\\]^_`{|}~“”‘’«»「」…–―※"
|
|
let unbanChars:number[] = []
|
|
|
|
for(const char of banChars){
|
|
unbanChars.push((await tokenizeNum(char))[0])
|
|
}
|
|
|
|
|
|
|
|
for(const char of banChars){
|
|
const encoded = await tokenizeNum(char)
|
|
if(encoded.length > 0){
|
|
if(!unbanChars.includes(encoded[0])){
|
|
bias[encoded[0]] = -100
|
|
}
|
|
}
|
|
for(const alt of charAlt){
|
|
let fchar = char
|
|
|
|
const encoded = await tokenizeNum(alt + fchar)
|
|
if(encoded.length > 0){
|
|
if(!unbanChars.includes(encoded[0])){
|
|
bias[encoded[0]] = -100
|
|
}
|
|
}
|
|
const encoded2 = await tokenizeNum(fchar + alt)
|
|
if(encoded2.length > 0){
|
|
if(!unbanChars.includes(encoded2[0])){
|
|
bias[encoded2[0]] = -100
|
|
}
|
|
}
|
|
}
|
|
}
|
|
localStorage.setItem('strongBan_' + data, JSON.stringify(bias))
|
|
return bias
|
|
} |