feat: add translation feature and ban character set functionality

This commit is contained in:
Kwaroran
2024-12-29 04:03:21 +09:00
parent ebdcd5ffcd
commit cd092a5a01
12 changed files with 221 additions and 56 deletions

View File

@@ -837,4 +837,5 @@ export const languageEnglish = {
showSavingIcon: "Show Saving Icon",
pluginVersionWarn: "This is {{plugin_version}} version of the plugin. which is not compatible with this version of RisuAI. please update the plugin to {{required_version}} version.",
imageTranslation: "Image Translation",
banCharacterset: 'Auto Regenerate On Characterset'
}

View File

@@ -16,6 +16,7 @@
import { joinMultiuserRoom } from "src/ts/sync/multiuser";
import PlaygroundSubtitle from "./PlaygroundSubtitle.svelte";
import PlaygroundImageTrans from "./PlaygroundImageTrans.svelte";
import PlaygroundTranslation from "./PlaygroundTranslation.svelte";
let easterEggTouch = $state(0)
@@ -95,6 +96,11 @@
}}>
<h1 class="text-2xl font-bold text-start">{language.imageTranslation}</h1>
</button>
<button class="bg-darkbg rounded-md p-6 flex flex-col transition-shadow hover:ring-1" onclick={() => {
PlaygroundStore.set(11)
}}>
<h1 class="text-2xl font-bold text-start">{language.translator}</h1>
</button>
<button class="bg-darkbg rounded-md p-6 flex flex-col transition-shadow hover:ring-1" onclick={() => {
PlaygroundStore.set(101)
}}>
@@ -157,6 +163,9 @@
{#if $PlaygroundStore === 10}
<PlaygroundImageTrans/>
{/if}
{#if $PlaygroundStore === 11}
<PlaygroundTranslation/>
{/if}
{#if $PlaygroundStore === 101}
<ToolConvertion/>
{/if}

View File

@@ -10,7 +10,7 @@
import { selectFileByDom, selectSingleFile, sleep } from "src/ts/util";
import { alertError, alertSelect } from "src/ts/alert";
import { risuChatParser } from "src/ts/parser.svelte";
import { AppendableBuffer, downloadFile, globalFetch } from "src/ts/globalApi.svelte";
import { AppendableBuffer, downloadFile, getLanguageCodes, globalFetch } from "src/ts/globalApi.svelte";
import SliderInput from "../UI/GUI/SliderInput.svelte";
import SelectInput from "../UI/GUI/SelectInput.svelte";
import OptionInput from "../UI/GUI/OptionInput.svelte";
@@ -31,40 +31,6 @@
let mode = $state('llm')
let sourceLang:string|null = $state(null)
function getLanguageCodes(){
let languageCodes:{
code: string
name: string
}[] = []
for(let i=0x41;i<=0x5A;i++){
for(let j=0x41;j<=0x5A;j++){
languageCodes.push({
code: String.fromCharCode(i) + String.fromCharCode(j),
name: ''
})
}
}
languageCodes = languageCodes.map(v => {
return {
code: v.code,
name: new Intl.DisplayNames([
DBState.db.language === 'cn' ? 'zh' : DBState.db.language
], {
type: 'language',
fallback: 'none'
}).of(v.code)
}
}).filter((a) => {
return a.name
}).sort((a, b) => a.name.localeCompare(b.name))
return languageCodes
}
async function runLLMMode() {
outputText = 'Loading...\n\n'

View File

@@ -0,0 +1,62 @@
<script lang="ts">
import { language } from "src/lang";
import TextAreaInput from "../UI/GUI/TextAreaInput.svelte";
import { LLMCacheStorage, runTranslator } from "src/ts/translator/translator";
import Button from "../UI/GUI/Button.svelte";
import SelectInput from "../UI/GUI/SelectInput.svelte";
import { getLanguageCodes } from "src/ts/globalApi.svelte";
import OptionInput from "../UI/GUI/OptionInput.svelte";
const userPreferedLang = navigator.language.split('-')[0]
let r = $state('')
let sourceLang = $state('en')
let output = $state('')
let outputLang = $state(userPreferedLang)
let loading = $state(false)
</script>
<span>{language.sourceLanguage}</span>
<SelectInput bind:value={sourceLang}>
{#each getLanguageCodes() as lang}
<OptionInput value={lang.code}>{lang.name}</OptionInput>
{/each}
</SelectInput>
<TextAreaInput bind:value={r} />
<span>{language.translatorLanguage}</span>
<SelectInput bind:value={outputLang}>
{#each getLanguageCodes() as lang}
<OptionInput value={lang.code}>{lang.name}</OptionInput>
{/each}
</SelectInput>
<TextAreaInput value={output} />
<Button className="mt-4" onclick={async () => {
try {
if(loading){
return
}
loading = true
output = await runTranslator(r, false, sourceLang, outputLang)
loading = false
} catch (error) {
console.error(error)
loading = false
}
}}>
{#if loading}
Loading...
{:else}
Translate
{/if}
</Button>
<Button className="mt-4" onclick={() => {
LLMCacheStorage.clear()
}}>
Clear Cache
</Button>

View File

@@ -13,12 +13,53 @@
import { installPython } from "src/ts/process/models/local";
import { Capacitor } from "@capacitor/core";
import { capStorageInvestigation } from "src/ts/storage/mobileStorage";
import Arcodion from "src/lib/UI/Arcodion.svelte";
let estaStorage:{
key:string,
size:string,
}[] = $state([])
const characterSets = [
'Latn',
'Hani',
'Arab',
'Deva',
'Cyrl',
'Beng',
'Hrkt',
'Telu',
'Hang',
'Taml',
'Thai',
'Gujr',
'Knda',
'Ethi',
'Khmr',
'Grek',
'Hebr',
]
const characterSetsPreview = {
'Latn': "ABC",
'Hani': "汉漢",
'Arab': "اعب",
'Deva': "अआइ",
'Cyrl': "АБВ",
'Beng': "অআই",
'Hrkt': "あア",
'Telu': "అఆఇ",
'Hang': "가나다",
'Taml': "அஆஇ",
'Thai': "กขค",
'Gujr': "અઆઇ",
'Knda': "ಅಆಇ",
'Ethi': "ሀሁሂ",
'Khmr': "កខគ",
'Grek': "ΑΒΓ",
'Hebr': "אבג",
}
</script>
<h2 class="text-2xl font-bold mt-2">{language.advancedSettings}</h2>
<span class="text-draculared text-xs mb-2">{language.advancedSettingsWarn}</span>
@@ -168,6 +209,21 @@
<Check bind:check={DBState.db.usePlainFetch} name={language.forcePlainFetch}> <Help key="forcePlainFetch" unrecommended/></Check>
</div>
{/if}
<Arcodion styled name={language.banCharacterset}>
{#each characterSets as set}
<Button styled={DBState.db.banCharacterset.includes(set) ? 'primary' : "outlined"} onclick={(e) => {
if (DBState.db.banCharacterset.includes(set)) {
DBState.db.banCharacterset = DBState.db.banCharacterset.filter((item) => item !== set)
} else {
DBState.db.banCharacterset.push(set)
}
}}>
{new Intl.DisplayNames([navigator.language,'en'], { type: 'script' }).of(set)} ({characterSetsPreview[set]})
</Button>
{/each}
</Arcodion>
<Button
className="mt-4"
onclick={async () => {

View File

@@ -7,7 +7,7 @@ import { v4 as uuidv4, v4 } from 'uuid';
import { characterFormatUpdate } from "./characters"
import { AppendableBuffer, BlankWriter, checkCharOrder, downloadFile, isNodeServer, isTauri, loadAsset, LocalWriter, openURL, readImage, saveAsset, VirtualWriter } from "./globalApi.svelte"
import { SettingsMenuIndex, ShowRealmFrameStore, selectedCharID, settingsOpen } from "./stores.svelte"
import { convertImage, hasher } from "./parser.svelte"
import { checkImageType, convertImage, hasher } from "./parser.svelte"
import { CCardLib, type CharacterCardV3, type LorebookEntry } from '@risuai/ccardlib'
import { reencodeImage } from "./process/files/inlays"
import { PngChunk } from "./pngChunk"
@@ -1316,6 +1316,21 @@ export async function exportCharacterCard(char:character, type:'png'|'json'|'cha
path = `assets/${type}/${itype}/${name}.${card.data.assets[i].ext}`
}
card.data.assets[i].uri = 'embeded://' + path
const imageType = checkImageType(rData)
if(imageType === 'PNG' && writer instanceof CharXWriter){
const metadatas:Record<string,string> = {}
const gen = PngChunk.readGenerator(rData)
for await (const chunk of gen){
if(!chunk || chunk instanceof AppendableBuffer){
continue
}
metadatas[chunk.key] = chunk.value
}
if(Object.keys(metadatas).length > 0){
const metaPath = `x_meta/${name}.json`
await writer.write(metaPath, Buffer.from(JSON.stringify(metadatas, null, 4)), 6)
}
}
await writer.write(path, Buffer.from(await convertImage(rData)))
}
}

View File

@@ -2261,4 +2261,36 @@ export class PerformanceDebugger{
this.kv[key].push(...other.kv[key])
}
}
}
export function getLanguageCodes(){
let languageCodes:{
code: string
name: string
}[] = []
for(let i=0x41;i<=0x5A;i++){
for(let j=0x41;j<=0x5A;j++){
languageCodes.push({
code: String.fromCharCode(i) + String.fromCharCode(j),
name: ''
})
}
}
languageCodes = languageCodes.map(v => {
return {
code: v.code.toLocaleLowerCase(),
name: new Intl.DisplayNames([
DBState.db.language === 'cn' ? 'zh' : DBState.db.language
], {
type: 'language',
fallback: 'none'
}).of(v.code)
}
}).filter((a) => {
return a.name
}).sort((a, b) => a.name.localeCompare(b.name))
return languageCodes
}

View File

@@ -26,7 +26,21 @@ class StreamChunkWriter{
break
}
if(typeString === 'tEXt'){
pos += 12 + len
const endPos = 12 + len + pos
//get key
let key=''
while(data[pos+8] !== 0){
key += String.fromCharCode(data[pos+8])
pos++
if(pos === endPos || key.length > 6){
break
}
}
if(key !== 'ccv3' && key !== 'chara'){
await this.pushData(data.slice(pos,endPos))
}
pos = endPos
}
else{
await this.pushData(data.slice(pos,pos+12+len))

View File

@@ -41,7 +41,7 @@ export class CharXWriter{
//do nothing, just to make compatible with other writer
}
async write(key:string,data:Uint8Array|string){
async write(key:string,data:Uint8Array|string, level?:0|1|2|3|4|5|6|7|8|9){
console.log('write',key)
let dat:Uint8Array
if(typeof data === 'string'){
@@ -52,7 +52,7 @@ export class CharXWriter{
}
this.writeEnd = false
const file = new fflate.ZipDeflate(key, {
level: 0
level: level ?? 0
});
await this.zip.add(file)
await file.push(dat, true)
@@ -102,6 +102,9 @@ export class CharXReader{
else if(file.name === 'module.risum'){
this.moduleData = assetData
}
else if(file.name.endsWith('.json')){
//do nothing
}
else{
this.assetPromises.push((async () => {
const assetId = await saveAsset(assetData)

View File

@@ -226,6 +226,19 @@ export async function requestChatData(arg:requestDataArgument, model:ModelModeEx
}
}
if(da.type === 'success' && db.banCharacterset?.length > 0){
for(const set of db.banCharacterset){
const checkRegex = new RegExp(`\\p{Script=${set}}`, 'gu')
if(checkRegex.test(da.result)){
return {
type: 'fail',
result: 'Banned character found'
}
}
}
}
if(da.type !== 'fail' || da.noRetry){
return da

View File

@@ -467,6 +467,7 @@ export function setDatabase(data:Database){
data.enableCustomFlags ??= false
data.assetMaxDifference ??= 4
data.showSavingIcon ??= false
data.banCharacterset ??= []
changeLanguage(data.language)
setDatabaseLite(data)
}
@@ -866,6 +867,7 @@ export interface Database{
pluginV2: RisuPlugin[]
showSavingIcon:boolean
presetRegex: customscript[]
banCharacterset:string[]
}
interface SeparateParameters{

View File

@@ -17,7 +17,7 @@ let cache={
trans: ['']
}
const LLMCacheStorage = localforage.createInstance({
export const LLMCacheStorage = localforage.createInstance({
name: "LLMTranslateCache"
})
@@ -25,10 +25,6 @@ let waitTrans = 0
export async function translate(text:string, reverse:boolean) {
let db = getDatabase()
const plug = await translatorPlugin(text, reverse ? db.translator: 'en', reverse ? 'en' : db.translator)
if(plug){
return plug.content
}
if(!reverse){
const ind = cache.origin.indexOf(text)
if(ind !== -1){
@@ -112,7 +108,7 @@ async function translateMain(text:string, arg:{from:string, to:string, host:stri
let db = getDatabase()
if(db.translatorType === 'llm'){
const tr = arg.to || 'en'
return translateLLM(text, {to: tr})
return translateLLM(text, {to: tr, from: arg.from})
}
if(db.translatorType === 'deepl'){
const body = {
@@ -197,12 +193,7 @@ async function translateMain(text:string, arg:{from:string, to:string, host:stri
return result
}
export async function translateVox(text:string) {
const plug = await translatorPlugin(text, 'en', 'ja')
if(plug){
return plug.content
}
export async function translateVox(text:string) {
return jaTrans(text)
}
@@ -245,7 +236,8 @@ export async function translateHTML(html: string, reverse:boolean, charArg:simpl
}
if(db.translatorType === 'llm'){
const tr = db.translator || 'en'
return translateLLM(html, {to: tr, regenerate})
const from = db.translatorInputLanguage
return translateLLM(html, {to: tr, from: from, regenerate})
}
const dom = new DOMParser().parseFromString(html, 'text/html');
console.log(html)
@@ -454,7 +446,7 @@ function needSuperChunkedTranslate(){
return getDatabase().translatorType === 'deeplX'
}
async function translateLLM(text:string, arg:{to:string, regenerate?:boolean}):Promise<string>{
async function translateLLM(text:string, arg:{to:string, from:string, regenerate?:boolean}):Promise<string>{
if(!arg.regenerate){
const cacheMatch = await LLMCacheStorage.getItem(text)
if(cacheMatch){
@@ -472,7 +464,7 @@ async function translateLLM(text:string, arg:{to:string, regenerate?:boolean}):P
const charIndex = get(selectedCharID)
const currentChar = db.characters[charIndex]
let translatorNote
if (currentChar.type === "character") {
if (currentChar?.type === "character") {
translatorNote = currentChar.translatorNote ?? ""
} else {
translatorNote = ""
@@ -480,12 +472,12 @@ async function translateLLM(text:string, arg:{to:string, regenerate?:boolean}):P
let formated:OpenAIChat[] = []
let prompt = db.translatorPrompt || `You are a translator. translate the following html or text into {{slot}}. do not output anything other than the translation.`
let parsedPrompt = parseChatML(prompt.replaceAll('{{slot}}', arg.to).replaceAll('{{solt::content}}', text).replaceAll('{{slot::tnote}}', translatorNote))
let parsedPrompt = parseChatML(prompt.replaceAll('{{slot::from}}', arg.from).replaceAll('{{slot}}', arg.to).replaceAll('{{solt::content}}', text).replaceAll('{{slot::tnote}}', translatorNote))
if(parsedPrompt){
formated = parsedPrompt
}
else{
prompt = prompt.replaceAll('{{slot}}', arg.to).replaceAll('{{slot::tnote}}', translatorNote)
prompt = prompt.replaceAll('{{slot}}', arg.to).replaceAll('{{slot::tnote}}', translatorNote).replaceAll('{{slot::from}}', arg.from)
formated = [
{
'role': 'system',