diff --git a/public/script.js b/public/script.js index 2523ff9ba..38f069968 100644 --- a/public/script.js +++ b/public/script.js @@ -154,7 +154,7 @@ import { isValidUrl, ensureImageFormatSupported, flashHighlight, - checkOverwriteExistingData, + isTrueBoolean, } from './scripts/utils.js'; import { debounce_timeout } from './scripts/constants.js'; @@ -232,7 +232,7 @@ import { renderTemplate, renderTemplateAsync } from './scripts/templates.js'; import { ScraperManager } from './scripts/scrapers.js'; import { SlashCommandParser } from './scripts/slash-commands/SlashCommandParser.js'; import { SlashCommand } from './scripts/slash-commands/SlashCommand.js'; -import { ARGUMENT_TYPE, SlashCommandArgument } from './scripts/slash-commands/SlashCommandArgument.js'; +import { ARGUMENT_TYPE, SlashCommandArgument, SlashCommandNamedArgument } from './scripts/slash-commands/SlashCommandArgument.js'; import { SlashCommandBrowser } from './scripts/slash-commands/SlashCommandBrowser.js'; import { initCustomSelectedSamplers, validateDisabledSamplers } from './scripts/samplerSelect.js'; import { DragAndDropHandler } from './scripts/dragdrop.js'; @@ -8446,9 +8446,30 @@ async function importFromURL(items, files) { } } -async function doImpersonate(_, prompt) { - $('#send_textarea').val(''); - $('#option_impersonate').trigger('click', { fromSlashCommand: true, additionalPrompt: prompt }); +async function doImpersonate(args, prompt) { + const options = prompt?.trim() ? { quiet_prompt: prompt.trim(), quietToLoud: true } : {}; + const shouldAwait = isTrueBoolean(args?.await); + const outerPromise = new Promise((outerResolve) => setTimeout(async () => { + try { + await waitUntilCondition(() => !is_send_press && !is_group_generating, 10000, 100); + } catch { + console.warn('Timeout waiting for generation unlock'); + toastr.warning('Cannot run /impersonate command while the reply is being generated.'); + return ''; + } + + // Prevent generate recursion + $('#send_textarea').val('')[0].dispatchEvent(new Event('input', { bubbles: true })); + + outerResolve(new Promise(innerResolve => setTimeout(() => innerResolve(Generate('impersonate', options)), 1))); + }, 1)); + + if (shouldAwait) { + const innerPromise = await outerPromise; + await innerPromise; + } + + return ''; } async function doDeleteChat() { @@ -8764,6 +8785,16 @@ jQuery(async function () { name: 'impersonate', callback: doImpersonate, aliases: ['imp'], + namedArgumentList: [ + new SlashCommandNamedArgument( + 'await', + 'Whether to await for the triggered generation before continuing', + [ARGUMENT_TYPE.BOOLEAN], + false, + false, + 'false', + ), + ], unnamedArgumentList: [ new SlashCommandArgument( 'prompt', [ARGUMENT_TYPE.STRING], false, @@ -8773,6 +8804,9 @@ jQuery(async function () {
Calls an impersonation response, with an optional additional prompt.
+
+ If await=true named argument is passed, the command will wait for the impersonation to end before continuing. +
Example:
+
+ + Disable +
+
+ + Enable +
Delete @@ -88,8 +96,8 @@
- - These files are available the current character in all chats they are in. + + These files are available for the current character in all chats they are in. @@ -113,8 +121,8 @@
- - These files are available to all characters in the current chat. + + These files are available for all characters in the current chat.
diff --git a/public/scripts/extensions/tts/elevenlabs.js b/public/scripts/extensions/tts/elevenlabs.js index 3504c813d..5cdc4cf39 100644 --- a/public/scripts/extensions/tts/elevenlabs.js +++ b/public/scripts/extensions/tts/elevenlabs.js @@ -75,9 +75,9 @@ class ElevenLabsTtsProvider { this.settings.style_exaggeration = $('#elevenlabs_tts_style_exaggeration').val(); this.settings.speaker_boost = $('#elevenlabs_tts_speaker_boost').is(':checked'); this.settings.model = $('#elevenlabs_tts_model').find(':selected').val(); - $('#elevenlabs_tts_stability_output').text(this.settings.stability * 100 + '%'); - $('#elevenlabs_tts_similarity_boost_output').text(this.settings.similarity_boost * 100 + '%'); - $('#elevenlabs_tts_style_exaggeration_output').text(this.settings.style_exaggeration * 100 + '%'); + $('#elevenlabs_tts_stability_output').text(Math.round(this.settings.stability * 100) + '%'); + $('#elevenlabs_tts_similarity_boost_output').text(Math.round(this.settings.similarity_boost * 100) + '%'); + $('#elevenlabs_tts_style_exaggeration_output').text(Math.round(this.settings.style_exaggeration * 100) + '%'); $('#elevenlabs_tts_v2_options').toggle(this.shouldInvolveExtendedSettings()); saveTtsProviderSettings(); } @@ -117,9 +117,9 @@ class ElevenLabsTtsProvider { $('#elevenlabs_tts_style_exaggeration').on('input', this.onSettingsChange.bind(this)); $('#elevenlabs_tts_speaker_boost').on('change', this.onSettingsChange.bind(this)); $('#elevenlabs_tts_model').on('change', this.onSettingsChange.bind(this)); - $('#elevenlabs_tts_stability_output').text(this.settings.stability); - $('#elevenlabs_tts_similarity_boost_output').text(this.settings.similarity_boost); - $('#elevenlabs_tts_style_exaggeration_output').text(this.settings.style_exaggeration); + $('#elevenlabs_tts_stability_output').text(Math.round(this.settings.stability * 100) + '%'); + $('#elevenlabs_tts_similarity_boost_output').text(Math.round(this.settings.similarity_boost * 100) + '%'); + $('#elevenlabs_tts_style_exaggeration_output').text(Math.round(this.settings.style_exaggeration * 100) + '%'); $('#elevenlabs_tts_v2_options').toggle(this.shouldInvolveExtendedSettings()); try { await this.checkReady(); @@ -311,8 +311,8 @@ class ElevenLabsTtsProvider { }, }; if (this.shouldInvolveExtendedSettings()) { - request.voice_settings.style_exaggeration = Number(this.settings.style_exaggeration); - request.voice_settings.speaker_boost = Boolean(this.settings.speaker_boost); + request.voice_settings.style = Number(this.settings.style_exaggeration); + request.voice_settings.use_speaker_boost = Boolean(this.settings.speaker_boost); } const response = await fetch(`https://api.elevenlabs.io/v1/text-to-speech/${voiceId}`, { method: 'POST', diff --git a/public/scripts/extensions/tts/index.js b/public/scripts/extensions/tts/index.js index 1ac1edd8b..c54458efe 100644 --- a/public/scripts/extensions/tts/index.js +++ b/public/scripts/extensions/tts/index.js @@ -11,6 +11,7 @@ import { power_user } from '../../power-user.js'; import { OpenAITtsProvider } from './openai.js'; import { XTTSTtsProvider } from './xtts.js'; import { GSVITtsProvider } from './gsvi.js'; +import { SBVits2TtsProvider } from './sbvits2.js'; import { AllTalkTtsProvider } from './alltalk.js'; import { SpeechT5TtsProvider } from './speecht5.js'; import { AzureTtsProvider } from './azure.js'; @@ -77,6 +78,7 @@ const ttsProviders = { Silero: SileroTtsProvider, XTTSv2: XTTSTtsProvider, GSVI: GSVITtsProvider, + SBVits2: SBVits2TtsProvider, System: SystemTtsProvider, Coqui: CoquiTtsProvider, Edge: EdgeTtsProvider, diff --git a/public/scripts/extensions/tts/sbvits2.js b/public/scripts/extensions/tts/sbvits2.js new file mode 100644 index 000000000..ab41d4f6d --- /dev/null +++ b/public/scripts/extensions/tts/sbvits2.js @@ -0,0 +1,339 @@ +import { getPreviewString, saveTtsProviderSettings } from './index.js'; + +export { SBVits2TtsProvider }; + +class SBVits2TtsProvider { + //########// + // Config // + //########// + + settings; + ready = false; + voices = []; + separator = '. '; + audioElement = document.createElement('audio'); + + /** + * Perform any text processing before passing to TTS engine. + * @param {string} text Input text + * @returns {string} Processed text + */ + processText(text) { + return text; + } + + languageLabels = { + 'Chinese': 'ZH', + 'English': 'EN', + 'Japanese': 'JP', + }; + + langKey2LangCode = { + 'ZH': 'zh-CN', + 'EN': 'en-US', + 'JP': 'ja-JP', + }; + + defaultSettings = { + provider_endpoint: 'http://localhost:5000', + sdp_ratio: 0.2, + noise: 0.6, + noisew: 0.8, + length: 1, + language: 'JP', + auto_split: true, + split_interval: 0.5, + assist_text: '', + assist_text_weight: 1, + style: 'Neutral', + style_weight: 1, + reference_audio_path: '', + }; + + get settingsHtml() { + let html = ` + + + SBVits2 Settings:
+ + + Use Style-Bert-VITS2 API Server.
+ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + `; + + return html; + } + + onSettingsChange() { + // Used when provider settings are updated from UI + this.settings.provider_endpoint = $('#sbvits_tts_endpoint').val(); + this.settings.language = $('#sbvits_api_language').val(); + this.settings.assist_text = $('#sbvits_assist_text').val(); + this.settings.reference_audio_path = $('#sbvits_reference_audio_path').val(); + + // Update the default TTS settings based on input fields + this.settings.sdp_ratio = $('#sbvits_sdp_ratio').val(); + this.settings.noise = $('#sbvits_noise').val(); + this.settings.noisew = $('#sbvits_noisew').val(); + this.settings.length = $('#sbvits_length').val(); + this.settings.auto_split = $('#sbvits_auto_split').is(':checked'); + this.settings.split_interval = $('#sbvits_split_interval').val(); + this.settings.assist_text_weight = $('#sbvits_assist_text_weight').val(); + this.settings.style_weight = $('#sbvits_style_weight').val(); + + // Update the UI to reflect changes + $('#sbvits_sdp_ratio_output').text(this.settings.sdp_ratio); + $('#sbvits_noise_output').text(this.settings.noise); + $('#sbvits_noisew_output').text(this.settings.noisew); + $('#sbvits_length_output').text(this.settings.length); + $('#sbvits_split_interval_output').text(this.settings.split_interval); + $('#sbvits_assist_text_weight_output').text(this.settings.assist_text_weight); + $('#sbvits_style_weight_output').text(this.settings.style_weight); + + saveTtsProviderSettings(); + this.changeTTSSettings(); + } + + async loadSettings(settings) { + // Pupulate 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; + + for (const key in settings) { + if (key in this.settings) { + this.settings[key] = settings[key]; + } else { + console.debug(`Ignoring non-user-configurable setting: ${key}`); + } + } + + // Set initial values from the settings + $('#sbvits_tts_endpoint').val(this.settings.provider_endpoint); + $('#sbvits_api_language').val(this.settings.language); + $('#sbvits_assist_text').val(this.settings.assist_text); + $('#sbvits_reference_audio_path').val(this.settings.reference_audio_path); + $('#sbvits_sdp_ratio').val(this.settings.sdp_ratio); + $('#sbvits_noise').val(this.settings.noise); + $('#sbvits_noisew').val(this.settings.noisew); + $('#sbvits_length').val(this.settings.length); + $('#sbvits_auto_split').prop('checked', this.settings.auto_split); + $('#sbvits_split_interval').val(this.settings.split_interval); + $('#sbvits_assist_text_weight').val(this.settings.assist_text_weight); + $('#sbvits_style_weight').val(this.settings.style_weight); + + // Update the UI to reflect changes + $('#sbvits_sdp_ratio_output').text(this.settings.sdp_ratio); + $('#sbvits_noise_output').text(this.settings.noise); + $('#sbvits_noisew_output').text(this.settings.noisew); + $('#sbvits_length_output').text(this.settings.length); + $('#sbvits_split_interval_output').text(this.settings.split_interval); + $('#sbvits_assist_text_weight_output').text(this.settings.assist_text_weight); + $('#sbvits_style_weight_output').text(this.settings.style_weight); + + // Register input/change event listeners to update settings on user interaction + $('#sbvits_tts_endpoint').on('input', () => { this.onSettingsChange(); }); + $('#sbvits_api_language').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_assist_text').on('input', () => { this.onSettingsChange(); }); + $('#sbvits_reference_audio_path').on('input', () => { this.onSettingsChange(); }); + $('#sbvits_sdp_ratio').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_noise').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_noisew').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_length').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_auto_split').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_split_interval').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_assist_text_weight').on('change', () => { this.onSettingsChange(); }); + $('#sbvits_style_weight').on('change', () => { this.onSettingsChange(); }); + + await this.checkReady(); + + console.info('SBVits2: Settings loaded'); + } + + // Perform a simple readiness check by trying to fetch voiceIds + async checkReady() { + await Promise.allSettled([this.fetchTtsVoiceObjects(), this.changeTTSSettings()]); + } + + async onRefreshClick() { + return; + } + + //#################// + // TTS Interfaces // + //#################// + + /** + * Get a voice from the TTS provider. + * @param {string} voiceName Voice name to get + * @returns {Promise} Voice object + */ + async getVoice(voiceName) { + if (this.voices.length == 0) { + this.voices = await this.fetchTtsVoiceObjects(); + } + const match = this.voices.filter( + v => v.name == voiceName, + )[0]; + if (!match) { + throw `TTS Voice name ${voiceName} not found`; + } + return match; + } + + async generateTts(text, voiceId) { + const response = await this.fetchTtsGeneration(text, voiceId); + return response; + } + + //###########// + // API CALLS // + //###########// + async fetchTtsVoiceObjects() { + const response = await fetch(`${this.settings.provider_endpoint}/models/info`); + if (!response.ok) { + throw new Error(`HTTP ${response.status}: ${await response.json()}`); + } + const data = await response.json(); + const voices = Object.keys(data).flatMap(key => { + const config = data[key]; + const spk2id = config.spk2id; + const style2id = config.style2id; + + return Object.entries(spk2id).flatMap(([speaker, speaker_id]) => { + return Object.entries(style2id).map(([style, styleId]) => { + return { + name: `${speaker} (${style})`, + voice_id: `${key}-${speaker_id}-${style}`, + preview_url: false, + }; + }); + }); + }); + + this.voices = voices; // Assign to the class property + return voices; // Also return this list + } + + // Each time a parameter is changed, we change the configuration + async changeTTSSettings() { + } + + /** + * Fetch TTS generation from the API. + * @param {string} inputText Text to generate TTS for + * @param {string} voiceId Voice ID to use (model_id-speaker_id-style) + * @returns {Promise} Fetch response + */ + async fetchTtsGeneration(inputText, voiceId) { + console.info(`Generating new TTS for voice_id ${voiceId}`); + + const [model_id, speaker_id, style] = voiceId.split('-'); + const params = new URLSearchParams(); + params.append('text', inputText); + params.append('model_id', model_id); + params.append('speaker_id', speaker_id); + params.append('sdp_ratio', this.settings.sdp_ratio); + params.append('noise', this.settings.noise); + params.append('noisew', this.settings.noisew); + params.append('length', this.settings.length); + params.append('language', this.settings.language); + params.append('auto_split', this.settings.auto_split); + params.append('split_interval', this.settings.split_interval); + if (this.settings.assist_text) { + params.append('assist_text', this.settings.assist_text); + params.append('assist_text_weight', this.settings.assist_text_weight); + } + params.append('style', style); + params.append('style_weight', this.settings.style_weight); + if (this.settings.reference_audio_path) { + params.append('reference_audio_path', this.settings.reference_audio_path); + } + const url = `${this.settings.provider_endpoint}/voice?${params.toString()}`; + + const response = await fetch( + url, + { + method: 'POST', + headers: { + }, + }, + ); + if (!response.ok) { + toastr.error(response.statusText, 'TTS Generation Failed'); + throw new Error(`HTTP ${response.status}: ${await response.text()}`); + } + return response; + } + + /** + * Preview TTS for a given voice ID. + * @param {string} id Voice ID + */ + async previewTtsVoice(id) { + this.audioElement.pause(); + this.audioElement.currentTime = 0; + const lang_code = this.langKey2LangCode[this.settings.lang] ?? 'ja-JP'; + const text = getPreviewString(lang_code); + const response = await this.fetchTtsGeneration(text, id); + if (!response.ok) { + throw new Error(`HTTP ${response.status}: ${await response.text()}`); + } + + const audio = await response.blob(); + const url = URL.createObjectURL(audio); + this.audioElement.src = url; + this.audioElement.play(); + } + + // Interface not used + async fetchTtsFromHistory(history_item_id) { + return Promise.resolve(history_item_id); + } +} diff --git a/public/scripts/extensions/vectors/index.js b/public/scripts/extensions/vectors/index.js index e5c42a1e3..cf64c7d7f 100644 --- a/public/scripts/extensions/vectors/index.js +++ b/public/scripts/extensions/vectors/index.js @@ -44,6 +44,7 @@ const settings = { cohere_model: 'embed-english-v3.0', ollama_model: 'mxbai-embed-large', ollama_keep: false, + vllm_model: '', summarize: false, summarize_sent: false, summary_source: 'main', @@ -691,6 +692,9 @@ function getVectorHeaders() { case 'llamacpp': addLlamaCppHeaders(headers); break; + case 'vllm': + addVllmHeaders(headers); + break; default: break; } @@ -761,6 +765,17 @@ function addLlamaCppHeaders(headers) { }); } +/** + * Add headers for the VLLM API source. + * @param {object} headers Header object + */ +function addVllmHeaders(headers) { + Object.assign(headers, { + 'X-Vllm-URL': textgenerationwebui_settings.server_urls[textgen_types.VLLM], + 'X-Vllm-Model': extension_settings.vectors.vllm_model, + }); +} + /** * Inserts vector items into a collection * @param {string} collectionId - The collection to insert into @@ -801,11 +816,12 @@ function throwIfSourceInvalid() { } if (settings.source === 'ollama' && !textgenerationwebui_settings.server_urls[textgen_types.OLLAMA] || + settings.source === 'vllm' && !textgenerationwebui_settings.server_urls[textgen_types.VLLM] || settings.source === 'llamacpp' && !textgenerationwebui_settings.server_urls[textgen_types.LLAMACPP]) { throw new Error('Vectors: API URL missing', { cause: 'api_url_missing' }); } - if (settings.source === 'ollama' && !settings.ollama_model) { + if (settings.source === 'ollama' && !settings.ollama_model || settings.source === 'vllm' && !settings.vllm_model) { throw new Error('Vectors: API model missing', { cause: 'api_model_missing' }); } @@ -965,6 +981,7 @@ function toggleSettings() { $('#cohere_vectorsModel').toggle(settings.source === 'cohere'); $('#ollama_vectorsModel').toggle(settings.source === 'ollama'); $('#llamacpp_vectorsModel').toggle(settings.source === 'llamacpp'); + $('#vllm_vectorsModel').toggle(settings.source === 'vllm'); $('#nomicai_apiKey').toggle(settings.source === 'nomicai'); } @@ -1274,6 +1291,12 @@ jQuery(async () => { Object.assign(extension_settings.vectors, settings); saveSettingsDebounced(); }); + $('#vectors_vllm_model').val(settings.vllm_model).on('input', () => { + $('#vectors_modelWarning').show(); + settings.vllm_model = String($('#vectors_vllm_model').val()); + Object.assign(extension_settings.vectors, settings); + saveSettingsDebounced(); + }); $('#vectors_ollama_keep').prop('checked', settings.ollama_keep).on('input', () => { settings.ollama_keep = $('#vectors_ollama_keep').prop('checked'); Object.assign(extension_settings.vectors, settings); diff --git a/public/scripts/extensions/vectors/settings.html b/public/scripts/extensions/vectors/settings.html index 48865c242..a6033ba83 100644 --- a/public/scripts/extensions/vectors/settings.html +++ b/public/scripts/extensions/vectors/settings.html @@ -20,6 +20,7 @@ +
@@ -82,6 +83,15 @@
+
+ + + + Hint: Set the URL in the API connection settings. + +
diff --git a/public/scripts/textgen-settings.js b/public/scripts/textgen-settings.js index dac12be1c..329cb5d0f 100644 --- a/public/scripts/textgen-settings.js +++ b/public/scripts/textgen-settings.js @@ -1019,6 +1019,14 @@ export function isJsonSchemaSupported() { return [TABBY, LLAMACPP].includes(settings.type) && main_api === 'textgenerationwebui'; } +function getLogprobsNumber() { + if (settings.type === VLLM) { + return 5; + } + + return 10; +} + export function getTextGenGenerationData(finalPrompt, maxTokens, isImpersonate, isContinue, cfgValues, type) { const canMultiSwipe = !isContinue && !isImpersonate && type !== 'quiet'; const { banned_tokens, banned_strings } = getCustomTokenBans(); @@ -1028,7 +1036,7 @@ export function getTextGenGenerationData(finalPrompt, maxTokens, isImpersonate, 'model': getTextGenModel(), 'max_new_tokens': maxTokens, 'max_tokens': maxTokens, - 'logprobs': power_user.request_token_probabilities ? 10 : undefined, + 'logprobs': power_user.request_token_probabilities ? getLogprobsNumber() : undefined, 'temperature': settings.dynatemp ? (settings.min_temp + settings.max_temp) / 2 : settings.temp, 'top_p': settings.top_p, 'typical_p': settings.typical_p, diff --git a/public/scripts/utils.js b/public/scripts/utils.js index 035208f49..5d5fe0f1d 100644 --- a/public/scripts/utils.js +++ b/public/scripts/utils.js @@ -654,11 +654,11 @@ function parseTimestamp(timestamp) { } // Unix time (legacy TAI / tags) - if (typeof timestamp === 'number') { + if (typeof timestamp === 'number' || /^\d+$/.test(timestamp)) { if (isNaN(timestamp) || !isFinite(timestamp) || timestamp < 0) { return moment.invalid(); } - return moment(timestamp); + return moment(Number(timestamp)); } // ST "humanized" format pattern diff --git a/src/endpoints/characters.js b/src/endpoints/characters.js index 8e663db56..c5c15b2e3 100644 --- a/src/endpoints/characters.js +++ b/src/endpoints/characters.js @@ -44,7 +44,7 @@ async function readCharacterData(inputFile, inputFormat = 'png') { /** * Writes the character card to the specified image file. - * @param {string} inputFile - Path to the image file + * @param {string|Buffer} inputFile - Path to the image file or image buffer * @param {string} data - Character card data * @param {string} outputFile - Target image file name * @param {import('express').Request} request - Express request obejct @@ -60,8 +60,20 @@ async function writeCharacterData(inputFile, data, outputFile, request, crop = u break; } } - // Read the image, resize, and save it as a PNG into the buffer - const inputImage = await tryReadImage(inputFile, crop); + + /** + * Read the image, resize, and save it as a PNG into the buffer. + * @returns {Promise} Image buffer + */ + function getInputImage() { + if (Buffer.isBuffer(inputFile)) { + return parseImageBuffer(inputFile, crop); + } + + return tryReadImage(inputFile, crop); + } + + const inputImage = await getInputImage(); // Get the chunks const outputImage = characterCardParser.write(inputImage, data); @@ -84,6 +96,32 @@ async function writeCharacterData(inputFile, data, outputFile, request, crop = u * @property {boolean} want_resize Resize the image to the standard avatar size */ +/** + * Parses an image buffer and applies crop if defined. + * @param {Buffer} buffer Buffer of the image + * @param {Crop|undefined} [crop] Crop parameters + * @returns {Promise} Image buffer + */ +async function parseImageBuffer(buffer, crop) { + const image = await jimp.read(buffer); + let finalWidth = image.bitmap.width, finalHeight = image.bitmap.height; + + // Apply crop if defined + if (typeof crop == 'object' && [crop.x, crop.y, crop.width, crop.height].every(x => typeof x === 'number')) { + image.crop(crop.x, crop.y, crop.width, crop.height); + // Apply standard resize if requested + if (crop.want_resize) { + finalWidth = AVATAR_WIDTH; + finalHeight = AVATAR_HEIGHT; + } else { + finalWidth = crop.width; + finalHeight = crop.height; + } + } + + return image.cover(finalWidth, finalHeight).getBufferAsync(jimp.MIME_PNG); +} + /** * Reads an image file and applies crop if defined. * @param {string} imgPath Path to the image file @@ -509,11 +547,25 @@ async function importFromCharX(uploadPath, { request }) { throw new Error('Invalid CharX card file: missing spec field'); } + /** @type {string|Buffer} */ + let avatar = defaultAvatarPath; + const assets = _.get(card, 'data.assets'); + if (Array.isArray(assets) && assets.length) { + for (const asset of assets.filter(x => x.type === 'icon' && typeof x.uri === 'string')) { + const pathNoProtocol = String(asset.uri.replace(/^(?:\/\/|[^/]+)*\//, '')); + const buffer = await extractFileFromZipBuffer(data, pathNoProtocol); + if (buffer) { + avatar = buffer; + break; + } + } + } + unsetFavFlag(card); card['create_date'] = humanizedISO8601DateTime(); card.name = sanitize(card.name); const fileName = getPngName(card.name, request.user.directories); - const result = await writeCharacterData(defaultAvatarPath, JSON.stringify(card), fileName, request); + const result = await writeCharacterData(avatar, JSON.stringify(card), fileName, request); return result ? fileName : ''; } diff --git a/src/endpoints/novelai.js b/src/endpoints/novelai.js index e26bfe34a..abbfe3ef6 100644 --- a/src/endpoints/novelai.js +++ b/src/endpoints/novelai.js @@ -281,6 +281,12 @@ router.post('/generate-image', jsonParser, async (request, response) => { const archiveBuffer = await generateResult.arrayBuffer(); const imageBuffer = await extractFileFromZipBuffer(archiveBuffer, '.png'); + + if (!imageBuffer) { + console.warn('NovelAI generated an image, but the PNG file was not found.'); + return response.sendStatus(500); + } + const originalBase64 = imageBuffer.toString('base64'); // No upscaling @@ -311,6 +317,11 @@ router.post('/generate-image', jsonParser, async (request, response) => { const upscaledArchiveBuffer = await upscaleResult.arrayBuffer(); const upscaledImageBuffer = await extractFileFromZipBuffer(upscaledArchiveBuffer, '.png'); + + if (!upscaledImageBuffer) { + throw new Error('NovelAI upscaled an image, but the PNG file was not found.'); + } + const upscaledBase64 = upscaledImageBuffer.toString('base64'); return response.send(upscaledBase64); diff --git a/src/endpoints/vectors.js b/src/endpoints/vectors.js index 7b089cd79..74666e584 100644 --- a/src/endpoints/vectors.js +++ b/src/endpoints/vectors.js @@ -16,6 +16,7 @@ const SOURCES = [ 'cohere', 'ollama', 'llamacpp', + 'vllm', ]; /** @@ -45,6 +46,8 @@ async function getVector(source, sourceSettings, text, isQuery, directories) { return require('../vectors/cohere-vectors').getCohereVector(text, isQuery, directories, sourceSettings.model); case 'llamacpp': return require('../vectors/llamacpp-vectors').getLlamaCppVector(text, sourceSettings.apiUrl, directories); + case 'vllm': + return require('../vectors/vllm-vectors').getVllmVector(text, sourceSettings.apiUrl, sourceSettings.model, directories); case 'ollama': return require('../vectors/ollama-vectors').getOllamaVector(text, sourceSettings.apiUrl, sourceSettings.model, sourceSettings.keep, directories); } @@ -91,6 +94,9 @@ async function getBatchVector(source, sourceSettings, texts, isQuery, directorie case 'llamacpp': results.push(...await require('../vectors/llamacpp-vectors').getLlamaCppBatchVector(batch, sourceSettings.apiUrl, directories)); break; + case 'vllm': + results.push(...await require('../vectors/vllm-vectors').getVllmBatchVector(batch, sourceSettings.apiUrl, sourceSettings.model, directories)); + break; case 'ollama': results.push(...await require('../vectors/ollama-vectors').getOllamaBatchVector(batch, sourceSettings.apiUrl, sourceSettings.model, sourceSettings.keep, directories)); break; @@ -278,6 +284,14 @@ function getSourceSettings(source, request) { return { apiUrl: apiUrl, }; + } else if (source === 'vllm') { + const apiUrl = String(request.headers['x-vllm-url']); + const model = String(request.headers['x-vllm-model']); + + return { + apiUrl: apiUrl, + model: model, + }; } else if (source === 'ollama') { const apiUrl = String(request.headers['x-ollama-url']); const model = String(request.headers['x-ollama-model']); diff --git a/src/util.js b/src/util.js index 8ccc7851c..326ad6a36 100644 --- a/src/util.js +++ b/src/util.js @@ -139,7 +139,7 @@ function getHexString(length) { * Extracts a file with given extension from an ArrayBuffer containing a ZIP archive. * @param {ArrayBuffer} archiveBuffer Buffer containing a ZIP archive * @param {string} fileExtension File extension to look for - * @returns {Promise} Buffer containing the extracted file + * @returns {Promise} Buffer containing the extracted file. Null if the file was not found. */ async function extractFileFromZipBuffer(archiveBuffer, fileExtension) { return await new Promise((resolve, reject) => yauzl.fromBuffer(Buffer.from(archiveBuffer), { lazyEntries: true }, (err, zipfile) => { @@ -171,6 +171,7 @@ async function extractFileFromZipBuffer(archiveBuffer, fileExtension) { zipfile.readEntry(); } }); + zipfile.on('end', () => resolve(null)); })); } diff --git a/src/vectors/vllm-vectors.js b/src/vectors/vllm-vectors.js new file mode 100644 index 000000000..64879edaa --- /dev/null +++ b/src/vectors/vllm-vectors.js @@ -0,0 +1,63 @@ +const fetch = require('node-fetch').default; +const { setAdditionalHeadersByType } = require('../additional-headers'); +const { TEXTGEN_TYPES } = require('../constants'); + +/** + * Gets the vector for the given text from VLLM + * @param {string[]} texts - The array of texts to get the vectors for + * @param {string} apiUrl - The API URL + * @param {string} model - The model to use + * @param {import('../users').UserDirectoryList} directories - The directories object for the user + * @returns {Promise} - The array of vectors for the texts + */ +async function getVllmBatchVector(texts, apiUrl, model, directories) { + const url = new URL(apiUrl); + url.pathname = '/v1/embeddings'; + + const headers = {}; + setAdditionalHeadersByType(headers, TEXTGEN_TYPES.VLLM, apiUrl, directories); + + const response = await fetch(url, { + method: 'POST', + headers: { + 'Content-Type': 'application/json', + ...headers, + }, + body: JSON.stringify({ input: texts, model }), + }); + + if (!response.ok) { + const responseText = await response.text(); + throw new Error(`VLLM: Failed to get vector for text: ${response.statusText} ${responseText}`); + } + + const data = await response.json(); + + if (!Array.isArray(data?.data)) { + throw new Error('API response was not an array'); + } + + // Sort data by x.index to ensure the order is correct + data.data.sort((a, b) => a.index - b.index); + + const vectors = data.data.map(x => x.embedding); + return vectors; +} + +/** + * Gets the vector for the given text from VLLM + * @param {string} text - The text to get the vector for + * @param {string} apiUrl - The API URL + * @param {string} model - The model to use + * @param {import('../users').UserDirectoryList} directories - The directories object for the user + * @returns {Promise} - The vector for the text + */ +async function getVllmVector(text, apiUrl, model, directories) { + const vectors = await getVllmBatchVector([text], apiUrl, model, directories); + return vectors[0]; +} + +module.exports = { + getVllmBatchVector, + getVllmVector, +};