Merge 6c44f5b3fd into be7eb8b2b5

style: `.mes_block { overflow-y: clip; }`

public/scripts/extensions/tts/gsvi.js

@@ -0,0 +1,276 @@
+
+import { saveTtsProviderSettings } from './index.js';
+
+export { GSVITtsProvider };
+
+class GSVITtsProvider {
+    //########//
+    // Config //
+    //########//
+
+    settings;
+    ready = false;
+    separator = '. ';
+
+    characterList = {};
+    voices = [];
+    /**
+     * Perform any text processing before passing to TTS engine.
+     * @param {string} text Input text
+     * @returns {string} Processed text
+     */
+    processText(text) {
+        text = text.replace('<br>', '\n'); // Replace <br> with newline
+        return text;
+    }
+
+    languageLabels = {
+        '多语种混合': '多语种混合',
+        '中文': '中文',
+        '英文': '英文',
+        '日文': '日文',
+        '中英混合': '中英混合',
+        '日英混合': '日英混合',
+    };
+    defaultSettings = {
+        provider_endpoint: 'http://127.0.0.1:5000',
+
+        language: '多语种混合',
+
+        cha_name: '',
+        character_emotion: 'default',
+
+        speed: 1,
+
+        top_k: 6,
+        top_p: 0.85,
+        temperature: 0.75,
+        batch_size: 10,
+
+        stream: false,
+        stream_chunk_size: 100,
+    };
+
+
+    // 新增获取角色和情绪的方法
+    async fetchCharacterList() {
+        const response = await fetch(this.settings.provider_endpoint + '/character_list');
+        if (!response.ok) {
+            throw new Error(`HTTP ${response.status}: ${await response.text()}`);
+        }
+        const characterList = await response.json();
+        this.characterList = characterList;
+        this.voices = Object.keys(characterList);
+
+    }
+
+
+
+    get settingsHtml() {
+        let html = `
+        <label for="gsvi_api_language">Text Language</label>
+        <select id="gsvi_api_language">`;
+
+        for (let language in this.languageLabels) {
+            if (this.languageLabels[language] == this.settings?.language) {
+                html += `<option value="${this.languageLabels[language]}" selected="selected">${language}</option>`;
+                continue;
+            }
+
+            html += `<option value="${this.languageLabels[language]}">${language}</option>`;
+        }
+
+        html += `
+        </select>
+        <label>GSVI Settings:</label><br/>
+        <label for="gsvi_tts_endpoint">Provider Endpoint:</label>
+        <input id="gsvi_tts_endpoint" type="text" class="text_pole" maxlength="250" value="${this.defaultSettings.provider_endpoint}"/>
+        
+        
+        
+        <div>Goto <a target="_blank" href="https://www.yuque.com/xter/zibxlp/knu8p82lb5ipufqy">API Document</a>.</div>
+
+        <label for="gsvi_speed">Speed: <span id="gsvi_tts_speed_output">${this.defaultSettings.speed}</span></label>
+        <input id="gsvi_speed" type="range" value="${this.defaultSettings.speed}" min="0.5" max="2" step="0.01" />
+        
+        <label for="gsvi_top_k">Top K: <span id="gsvi_top_k_output">${this.defaultSettings.top_k}</span></label>
+        <input id="gsvi_top_k" type="range" value="${this.defaultSettings.top_k}" min="0" max="100" step="1" />
+        
+        <label for="gsvi_top_p">Top P: <span id="gsvi_top_p_output">${this.defaultSettings.top_p}</span></label>
+        <input id="gsvi_top_p" type="range" value="${this.defaultSettings.top_p}" min="0" max="1" step="0.01" />
+        
+        <label for="gsvi_temperature">Temperature: <span id="gsvi_tts_temperature_output">${this.defaultSettings.temperature}</span></label>
+        <input id="gsvi_temperature" type="range" value="${this.defaultSettings.temperature}" min="0.01" max="1" step="0.01" />
+
+        <label for="gsvi_batch_size">Batch Size: <span id="gsvi_batch_size_output">${this.defaultSettings.batch_size}</span></label>
+        <input id="gsvi_batch_size" type="range" value="${this.defaultSettings.batch_size}" min="1" max="35" step="1" />
+        
+        <label for="gsvi_tts_streaming" class="checkbox_label">
+            <input id="gsvi_tts_streaming" type="checkbox" ${this.defaultSettings.stream ? 'checked' : ''}/>
+            <span>Streaming</span>
+        </label>
+
+        <label for="gsvi_stream_chunk_size">Stream Chunk Size: <span id="gsvi_stream_chunk_size_output">${this.defaultSettings.stream_chunk_size}</span></label>
+        <input id="gsvi_stream_chunk_size" type="range" value="${this.defaultSettings.stream_chunk_size}" min="100" max="400" step="1" />
+        <title>About GSVI (GPT-Sovits Inference)</title>
+        <p>
+        GSVI (GPT-Sovits Inference) is an inference enhancement project based on 
+            <a href="https://github.com/RVC-Boss/GPT-SoVITS" target="_blank">GPT-Sovits</a>, allowing you to run an API interface locally, offering emotion-rich speech-to-text and convenient model management features.
+        </p>
+        <p>
+            For more information, visit the 
+            <a href="https://github.com/X-T-E-R/GPT-SoVITS-Inference" target="_blank">GSVI project page</a>.
+        </p>
+        `;
+
+        return html;
+    }
+
+    onSettingsChange() {
+        // Update provider settings based on input fields
+        this.settings.provider_endpoint = $('#gsvi_tts_endpoint').val();
+        this.settings.language = $('#gsvi_api_language').val();
+
+
+        // Update the rest of TTS settings based on input fields
+        this.settings.speed = parseFloat($('#gsvi_speed').val());
+        this.settings.temperature = parseFloat($('#gsvi_temperature').val());
+        this.settings.top_k = parseInt($('#gsvi_top_k').val(), 10);
+        this.settings.top_p = parseFloat($('#gsvi_top_p').val());
+        this.settings.batch_size = parseInt($('#gsvi_batch_size').val(), 10);
+        this.settings.stream = $('#gsvi_tts_streaming').is(':checked');
+        this.settings.stream_chunk_size = parseInt($('#gsvi_stream_chunk_size').val(), 10);
+
+        // Update UI to reflect changes
+
+        $('#gsvi_tts_speed_output').text(this.settings.speed);
+        $('#gsvi_tts_temperature_output').text(this.settings.temperature);
+        $('#gsvi_top_k_output').text(this.settings.top_k);
+        $('#gsvi_top_p_output').text(this.settings.top_p);
+        $('#gsvi_stream_chunk_size_output').text(this.settings.stream_chunk_size);
+        $('#gsvi_batch_size_output').text(this.settings.batch_size);
+
+
+
+
+        // Persist settings changes
+        saveTtsProviderSettings();
+
+    }
+
+    async loadSettings(settings) {
+        // Populate Provider UI given input settings
+        if (Object.keys(settings).length === 0) {
+            console.info('Using default TTS Provider settings');
+        }
+
+        // Only accept keys defined in defaultSettings
+        this.settings = { ...this.defaultSettings, ...settings };
+
+        // Fetch character and emotion list
+        // Set initial values from the settings
+        $('#gsvi_tts_endpoint').val(this.settings.provider_endpoint);
+        $('#gsvi_api_language').val(this.settings.language);
+
+        $('#gsvi_speed').val(this.settings.speed);
+        $('#gsvi_temperature').val(this.settings.temperature);
+        $('#gsvi_top_k').val(this.settings.top_k);
+        $('#gsvi_top_p').val(this.settings.top_p);
+        $('#gsvi_batch_size').val(this.settings.batch_size);
+        $('#gsvi_tts_streaming').prop('checked', this.settings.stream);
+        $('#gsvi_stream_chunk_size').val(this.settings.stream_chunk_size);
+
+        // Update UI to reflect initial settings
+        $('#gsvi_tts_speed_output').text(this.settings.speed);
+        $('#gsvi_tts_temperature_output').text(this.settings.temperature);
+        $('#gsvi_top_k_output').text(this.settings.top_k);
+        $('#gsvi_top_p_output').text(this.settings.top_p);
+        $('#gsvi_stream_chunk_size_output').text(this.settings.stream_chunk_size);
+
+        // Register event listeners to update settings on user interaction
+        // (Similar to before, ensure event listeners for character and emotion selection are included)
+        // Register input/change event listeners to update settings on user interaction
+        $('#gsvi_tts_endpoint').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_api_language').on('change', () => { this.onSettingsChange(); });
+
+        $('#gsvi_speed').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_temperature').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_top_k').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_top_p').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_batch_size').on('input', () => { this.onSettingsChange(); });
+        $('#gsvi_tts_streaming').on('change', () => { this.onSettingsChange(); });
+        $('#gsvi_stream_chunk_size').on('input', () => { this.onSettingsChange(); });
+
+        await this.checkReady();
+        console.debug('GSVI: Settings loaded');
+    }
+
+
+
+    // Perform a simple readiness check by trying to fetch voiceIds
+    async checkReady() {
+        await Promise.allSettled([this.fetchCharacterList()]);
+    }
+
+    async onRefreshClick() {
+        return;
+    }
+
+    //#################//
+    //  TTS Interfaces //
+    //#################//
+
+    async getVoice(voiceName) {
+        if (this.voices.length == 0) {
+            this.fetchCharacterList();
+        }
+        if (!this.voices.includes(voiceName)) {
+            throw `TTS Voice name ${voiceName} not found`;
+        }
+        return { name: voiceName, voice_id: voiceName, preview_url: false, lang: 'zh-CN' };
+    }
+
+    async generateTts(text, voiceId) {
+        const response = await this.fetchTtsGeneration(text, voiceId);
+        return response;
+    }
+
+    //###########//
+    // API CALLS //
+    //###########//
+    async fetchTtsVoiceObjects() {
+        if (this.voices.length == 0) {
+            await this.fetchCharacterList();
+        }
+        console.log(this.voices);
+        const voices = this.voices.map(x => ({ name: x, voice_id: x, preview_url: false, lang: 'zh-CN' }));
+        return voices;
+    }
+
+
+    async fetchTtsGeneration(inputText, voiceId) {
+        console.info(`Generating new TTS for voice_id ${voiceId}`);
+
+
+        const params = new URLSearchParams();
+        params.append('text', inputText);
+        params.append('cha_name', voiceId);
+        params.append('text_language', this.settings.language);
+        params.append('batch_size', this.settings.batch_size.toString());
+        params.append('speed', this.settings.speed.toString());
+        params.append('top_k', this.settings.top_k.toString());
+        params.append('top_p', this.settings.top_p.toString());
+        params.append('temperature', this.settings.temperature.toString());
+        params.append('stream', this.settings.stream.toString());
+
+        
+        return `${this.settings.provider_endpoint}/tts?${params.toString()}`;
+       
+    }
+
+    // Interface not used by GSVI TTS
+    async fetchTtsFromHistory(history_item_id) {
+        return Promise.resolve(history_item_id);
+    }
+
+}

public/scripts/extensions/tts/index.js

@@ -11,6 +11,7 @@ import { power_user } from '../../power-user.js';
 import { registerSlashCommand } from '../../slash-commands.js';
 import { OpenAITtsProvider } from './openai.js';
 import { XTTSTtsProvider } from './xtts.js';
+import { GSVITtsProvider } from './gsvi.js';
 import { AllTalkTtsProvider } from './alltalk.js';
 import { SpeechT5TtsProvider } from './speecht5.js';
 export { talkingAnimation };
@@ -72,6 +73,7 @@ const ttsProviders = {
     ElevenLabs: ElevenLabsTtsProvider,
     Silero: SileroTtsProvider,
     XTTSv2: XTTSTtsProvider,
+    GSVI: GSVITtsProvider,
     System: SystemTtsProvider,
     Coqui: CoquiTtsProvider,
     Edge: EdgeTtsProvider,

public/scripts/openai.js

@@ -3250,7 +3250,8 @@ async function onExportPresetClick() {
     delete preset.proxy_password;
 
     const presetJsonString = JSON.stringify(preset, null, 4);
-    download(presetJsonString, oai_settings.preset_settings_openai, 'application/json');
+    const presetFileName = `${oai_settings.preset_settings_openai}.json`;
+    download(presetJsonString, presetFileName, 'application/json');
 }
 
 async function onLogitBiasPresetImportFileChange(e) {
@@ -3298,7 +3299,8 @@ function onLogitBiasPresetExportClick() {
     }
 
     const presetJsonString = JSON.stringify(oai_settings.bias_presets[oai_settings.bias_preset_selected], null, 4);
-    download(presetJsonString, oai_settings.bias_preset_selected, 'application/json');
+    const presetFileName = `${oai_settings.bias_preset_selected}.json`;
+    download(presetJsonString, presetFileName, 'application/json');
 }
 
 async function onDeletePresetClick() {

public/style.css

@@ -1000,6 +1000,7 @@ body .panelControlBar {
     padding-left: 10px;
     width: 100%;
     overflow-x: hidden;
+    overflow-y: clip;
 }
 
 .mes_text {

server.js

@@ -45,7 +45,6 @@ const {
     forwardFetchResponse,
 } = require('./src/util');
 const { ensureThumbnailCache } = require('./src/endpoints/thumbnails');
-const { loadTokenizers } = require('./src/endpoints/tokenizers');
 
 // Work around a node v20.0.0, v20.1.0, and v20.2.0 bug. The issue was fixed in v20.3.0.
 // https://github.com/nodejs/node/issues/47822#issuecomment-1564708870
@@ -543,22 +542,12 @@ const setupTasks = async function () {
     }
     console.log();
 
-    // TODO: do endpoint init functions depend on certain directories existing or not existing? They should be callable
-    // in any order for encapsulation reasons, but right now it's unknown if that would break anything.
-    await userModule.initUserStorage(dataRoot);
-
-    if (listen && !basicAuthMode && enableAccounts) {
-        await userModule.checkAccountsProtection();
-    }
-
-    await settingsEndpoint.init();
-    const directories = await userModule.ensurePublicDirectoriesExist();
-    await userModule.migrateUserData();
+    const directories = await userModule.getUserDirectoriesList();
     await contentManager.checkForNewContent(directories);
     await ensureThumbnailCache();
     cleanUploads();
 
-    await loadTokenizers();
+    await settingsEndpoint.init();
     await statsEndpoint.init();
 
     const cleanupPlugins = await loadPlugins();
@@ -581,7 +570,6 @@ const setupTasks = async function () {
         exitProcess();
     });
 
-
     console.log('Launching...');
 
     if (autorun) open(autorunUrl.toString());
@@ -601,6 +589,9 @@ const setupTasks = async function () {
         }
     }
 
+    if (listen && !basicAuthMode && enableAccounts) {
+        await userModule.checkAccountsProtection();
+    }
 };
 
 /**
@@ -642,21 +633,27 @@ function setWindowTitle(title) {
     }
 }
 
-if (cliArguments.ssl) {
-    https.createServer(
-        {
-            cert: fs.readFileSync(cliArguments.certPath),
-            key: fs.readFileSync(cliArguments.keyPath),
-        }, app)
-        .listen(
-            Number(tavernUrl.port) || 443,
-            tavernUrl.hostname,
-            setupTasks,
-        );
-} else {
-    http.createServer(app).listen(
-        Number(tavernUrl.port) || 80,
-        tavernUrl.hostname,
-        setupTasks,
-    );
-}
+// User storage module needs to be initialized before starting the server
+userModule.initUserStorage(dataRoot)
+    .then(userModule.ensurePublicDirectoriesExist)
+    .then(userModule.migrateUserData)
+    .finally(() => {
+        if (cliArguments.ssl) {
+            https.createServer(
+                {
+                    cert: fs.readFileSync(cliArguments.certPath),
+                    key: fs.readFileSync(cliArguments.keyPath),
+                }, app)
+                .listen(
+                    Number(tavernUrl.port) || 443,
+                    tavernUrl.hostname,
+                    setupTasks,
+                );
+        } else {
+            http.createServer(app).listen(
+                Number(tavernUrl.port) || 80,
+                tavernUrl.hostname,
+                setupTasks,
+            );
+        }
+    });

src/endpoints/tokenizers.js

@@ -10,6 +10,10 @@ const { TEXTGEN_TYPES } = require('../constants');
 const { jsonParser } = require('../express-common');
 const { setAdditionalHeaders } = require('../additional-headers');
 
+/**
+ * @typedef { (req: import('express').Request, res: import('express').Response) => Promise<any> } TokenizationHandler
+ */
+
 /**
  * @type {{[key: string]: import("@dqbd/tiktoken").Tiktoken}} Tokenizers cache
  */
@@ -48,16 +52,30 @@ const TEXT_COMPLETION_MODELS = [
 
 const CHARS_PER_TOKEN = 3.35;
 
+/**
+ * Sentencepiece tokenizer for tokenizing text.
+ */
 class SentencePieceTokenizer {
+    /**
+     * @type {import('@agnai/sentencepiece-js').SentencePieceProcessor} Sentencepiece tokenizer instance
+     */
     #instance;
+    /**
+     * @type {string} Path to the tokenizer model
+     */
     #model;
 
+    /**
+     * Creates a new Sentencepiece tokenizer.
+     * @param {string} model Path to the tokenizer model
+     */
     constructor(model) {
         this.#model = model;
     }
 
     /**
      * Gets the Sentencepiece tokenizer instance.
+     * @returns {Promise<import('@agnai/sentencepiece-js').SentencePieceProcessor|null>} Sentencepiece tokenizer instance
      */
     async get() {
         if (this.#instance) {
@@ -76,18 +94,61 @@ class SentencePieceTokenizer {
     }
 }
 
-const spp_llama = new SentencePieceTokenizer('src/sentencepiece/llama.model');
-const spp_nerd = new SentencePieceTokenizer('src/sentencepiece/nerdstash.model');
-const spp_nerd_v2 = new SentencePieceTokenizer('src/sentencepiece/nerdstash_v2.model');
-const spp_mistral = new SentencePieceTokenizer('src/sentencepiece/mistral.model');
-const spp_yi = new SentencePieceTokenizer('src/sentencepiece/yi.model');
-let claude_tokenizer;
+/**
+ * Web tokenizer for tokenizing text.
+ */
+class WebTokenizer {
+    /**
+     * @type {Tokenizer} Web tokenizer instance
+     */
+    #instance;
+    /**
+     * @type {string} Path to the tokenizer model
+     */
+    #model;
+
+    /**
+     * Creates a new Web tokenizer.
+     * @param {string} model Path to the tokenizer model
+     */
+    constructor(model) {
+        this.#model = model;
+    }
+
+    /**
+     * Gets the Web tokenizer instance.
+     * @returns {Promise<Tokenizer|null>} Web tokenizer instance
+     */
+    async get() {
+        if (this.#instance) {
+            return this.#instance;
+        }
+
+        try {
+            const arrayBuffer = fs.readFileSync(this.#model).buffer;
+            this.#instance = await Tokenizer.fromJSON(arrayBuffer);
+            console.log('Instantiated the tokenizer for', path.parse(this.#model).name);
+            return this.#instance;
+        } catch (error) {
+            console.error('Web tokenizer failed to load: ' + this.#model, error);
+            return null;
+        }
+    }
+}
+
+const spp_llama = new SentencePieceTokenizer('src/tokenizers/llama.model');
+const spp_nerd = new SentencePieceTokenizer('src/tokenizers/nerdstash.model');
+const spp_nerd_v2 = new SentencePieceTokenizer('src/tokenizers/nerdstash_v2.model');
+const spp_mistral = new SentencePieceTokenizer('src/tokenizers/mistral.model');
+const spp_yi = new SentencePieceTokenizer('src/tokenizers/yi.model');
+const claude_tokenizer = new WebTokenizer('src/tokenizers/claude.json');
 
 const sentencepieceTokenizers = [
     'llama',
     'nerdstash',
     'nerdstash_v2',
     'mistral',
+    'yi',
 ];
 
 /**
@@ -112,6 +173,10 @@ function getSentencepiceTokenizer(model) {
         return spp_nerd_v2;
     }
 
+    if (model.includes('yi')) {
+        return spp_yi;
+    }
+
     return null;
 }
 
@@ -168,13 +233,23 @@ async function getTiktokenChunks(tokenizer, ids) {
     return chunks;
 }
 
-async function getWebTokenizersChunks(tokenizer, ids) {
+/**
+ * Gets the token chunks for the given token IDs using the Web tokenizer.
+ * @param {Tokenizer} tokenizer Web tokenizer instance
+ * @param {number[]} ids Token IDs
+ * @returns {string[]} Token chunks
+ */
+function getWebTokenizersChunks(tokenizer, ids) {
     const chunks = [];
 
-    for (let i = 0; i < ids.length; i++) {
-        const id = ids[i];
-        const chunkText = await tokenizer.decode(new Uint32Array([id]));
+    for (let i = 0, lastProcessed = 0; i < ids.length; i++) {
+        const chunkIds = ids.slice(lastProcessed, i + 1);
+        const chunkText = tokenizer.decode(new Int32Array(chunkIds));
+        if (chunkText === '<27>') {
+            continue;
+        }
         chunks.push(chunkText);
+        lastProcessed = i + 1;
     }
 
     return chunks;
@@ -237,17 +312,12 @@ function getTiktokenTokenizer(model) {
     return tokenizer;
 }
 
-async function loadClaudeTokenizer(modelPath) {
-    try {
-        const arrayBuffer = fs.readFileSync(modelPath).buffer;
-        const instance = await Tokenizer.fromJSON(arrayBuffer);
-        return instance;
-    } catch (error) {
-        console.error('Claude tokenizer failed to load: ' + modelPath, error);
-        return null;
-    }
-}
-
+/**
+ * Counts the tokens for the given messages using the Claude tokenizer.
+ * @param {Tokenizer} tokenizer Web tokenizer
+ * @param {object[]} messages Array of messages
+ * @returns {number} Number of tokens
+ */
 function countClaudeTokens(tokenizer, messages) {
     // Should be fine if we use the old conversion method instead of the messages API one i think?
     const convertedPrompt = convertClaudePrompt(messages, false, '', false, false, '', false);
@@ -264,9 +334,14 @@ function countClaudeTokens(tokenizer, messages) {
 /**
  * Creates an API handler for encoding Sentencepiece tokens.
  * @param {SentencePieceTokenizer} tokenizer Sentencepiece tokenizer
- * @returns {any} Handler function
+ * @returns {TokenizationHandler} Handler function
  */
 function createSentencepieceEncodingHandler(tokenizer) {
+    /**
+     * Request handler for encoding Sentencepiece tokens.
+     * @param {import('express').Request} request
+     * @param {import('express').Response} response
+     */
     return async function (request, response) {
         try {
             if (!request.body) {
@@ -276,7 +351,7 @@ function createSentencepieceEncodingHandler(tokenizer) {
             const text = request.body.text || '';
             const instance = await tokenizer?.get();
             const { ids, count } = await countSentencepieceTokens(tokenizer, text);
-            const chunks = await instance?.encodePieces(text);
+            const chunks = instance?.encodePieces(text);
             return response.send({ ids, count, chunks });
         } catch (error) {
             console.log(error);
@@ -288,9 +363,14 @@ function createSentencepieceEncodingHandler(tokenizer) {
 /**
  * Creates an API handler for decoding Sentencepiece tokens.
  * @param {SentencePieceTokenizer} tokenizer Sentencepiece tokenizer
- * @returns {any} Handler function
+ * @returns {TokenizationHandler} Handler function
  */
 function createSentencepieceDecodingHandler(tokenizer) {
+    /**
+     * Request handler for decoding Sentencepiece tokens.
+     * @param {import('express').Request} request
+     * @param {import('express').Response} response
+     */
     return async function (request, response) {
         try {
             if (!request.body) {
@@ -299,6 +379,7 @@ function createSentencepieceDecodingHandler(tokenizer) {
 
             const ids = request.body.ids || [];
             const instance = await tokenizer?.get();
+            if (!instance) throw new Error('Failed to load the Sentencepiece tokenizer');
             const ops = ids.map(id => instance.decodeIds([id]));
             const chunks = await Promise.all(ops);
             const text = chunks.join('');
@@ -313,9 +394,14 @@ function createSentencepieceDecodingHandler(tokenizer) {
 /**
  * Creates an API handler for encoding Tiktoken tokens.
  * @param {string} modelId Tiktoken model ID
- * @returns {any} Handler function
+ * @returns {TokenizationHandler} Handler function
  */
 function createTiktokenEncodingHandler(modelId) {
+    /**
+     * Request handler for encoding Tiktoken tokens.
+     * @param {import('express').Request} request
+     * @param {import('express').Response} response
+     */
     return async function (request, response) {
         try {
             if (!request.body) {
@@ -337,9 +423,14 @@ function createTiktokenEncodingHandler(modelId) {
 /**
  * Creates an API handler for decoding Tiktoken tokens.
  * @param {string} modelId Tiktoken model ID
- * @returns {any} Handler function
+ * @returns {TokenizationHandler} Handler function
  */
 function createTiktokenDecodingHandler(modelId) {
+    /**
+     * Request handler for decoding Tiktoken tokens.
+     * @param {import('express').Request} request
+     * @param {import('express').Response} response
+     */
     return async function (request, response) {
         try {
             if (!request.body) {
@@ -358,14 +449,6 @@ function createTiktokenDecodingHandler(modelId) {
     };
 }
 
-/**
- * Loads the model tokenizers.
- * @returns {Promise<void>} Promise that resolves when the tokenizers are loaded
- */
-async function loadTokenizers() {
-    claude_tokenizer = await loadClaudeTokenizer('src/claude.json');
-}
-
 const router = express.Router();
 
 router.post('/ai21/count', jsonParser, async function (req, res) {
@@ -446,8 +529,10 @@ router.post('/openai/encode', jsonParser, async function (req, res) {
 
         if (queryModel.includes('claude')) {
             const text = req.body.text || '';
-            const tokens = Object.values(claude_tokenizer.encode(text));
-            const chunks = await getWebTokenizersChunks(claude_tokenizer, tokens);
+            const instance = await claude_tokenizer.get();
+            if (!instance) throw new Error('Failed to load the Claude tokenizer');
+            const tokens = Object.values(instance.encode(text));
+            const chunks = getWebTokenizersChunks(instance, tokens);
             return res.send({ ids: tokens, count: tokens.length, chunks });
         }
 
@@ -481,7 +566,9 @@ router.post('/openai/decode', jsonParser, async function (req, res) {
 
         if (queryModel.includes('claude')) {
             const ids = req.body.ids || [];
-            const chunkText = await claude_tokenizer.decode(new Uint32Array(ids));
+            const instance = await claude_tokenizer.get();
+            if (!instance) throw new Error('Failed to load the Claude tokenizer');
+            const chunkText = instance.decode(new Int32Array(ids));
             return res.send({ text: chunkText });
         }
 
@@ -503,7 +590,9 @@ router.post('/openai/count', jsonParser, async function (req, res) {
         const model = getTokenizerModel(queryModel);
 
         if (model === 'claude') {
-            num_tokens = countClaudeTokens(claude_tokenizer, req.body);
+            const instance = await claude_tokenizer.get();
+            if (!instance) throw new Error('Failed to load the Claude tokenizer');
+            num_tokens = countClaudeTokens(instance, req.body);
             return res.send({ 'token_count': num_tokens });
         }
 
@@ -665,7 +754,6 @@ module.exports = {
     getTokenizerModel,
     getTiktokenTokenizer,
     countClaudeTokens,
-    loadTokenizers,
     getSentencepiceTokenizer,
     sentencepieceTokenizers,
     router,

src/users.js

@@ -112,6 +112,16 @@ async function ensurePublicDirectoriesExist() {
     return directoriesList;
 }
 
+/**
+ * Gets a list of all user directories.
+ * @returns {Promise<import('./users').UserDirectoryList[]>} - The list of user directories
+ */
+async function getUserDirectoriesList() {
+    const userHandles = await getAllUserHandles();
+    const directoriesList = userHandles.map(handle => getUserDirectories(handle));
+    return directoriesList;
+}
+
 /**
  * Perform migration from the old user data format to the new one.
  */
@@ -707,6 +717,7 @@ module.exports = {
     toAvatarKey,
     initUserStorage,
     ensurePublicDirectoriesExist,
+    getUserDirectoriesList,
     getAllUserHandles,
     getUserDirectories,
     setUserDataMiddleware,