Merge branch 'staging' of https://github.com/joenunezb/SillyTavern into optimize/improve-search

.github/readme-ru_ru.md

@@ -229,6 +229,46 @@ SillyTavern сохраняет ключи от ваших API в файле `sec
 1. Зайдите в файл `config.yaml` и установите `allowKeysExposure` в положение `true`.
 2. Перезапустите сервер SillyTavern.
 
+## Аргументы командной строки
+
+Вы можете передавать аргументы командной строки при запуске сервера SillyTavern, чтобы переопределять настройки из `config.yaml`.
+
+### Примеры
+
+```shell
+node server.js --port 8000 --listen false
+# или
+npm run start -- --port 8000 --listen false
+# или (только на Windows)
+Start.bat --port 8000 --listen false
+```
+
+### Поддерживаемые аргументы
+
+| Аргумент                | Описание                                                                                                       | Тип      |
+|-------------------------|----------------------------------------------------------------------------------------------------------------|----------|
+| `--version`             | Показывает номер версии.                                                                                       | boolean  |
+| `--enableIPv6`          | Включает IPv6.                                                                                                 | boolean  |
+| `--enableIPv4`          | Включает IPv4.                                                                                                 | boolean  |
+| `--port`                | Устанавливает порт, котрый будет использовать SillyTavern. Если не указан, то используется yaml-конфиг 'port'. | number   |
+| `--dnsPreferIPv6`       | Отдает предпочтение IPv6 для dns. Если не указан, то используется yaml-конфиг 'preferIPv6'.                    | boolean  |
+| `--autorun`             | Автоматический запуск SillyTavern в браузере. Если не указан, то используется yaml-конфиг 'autorun'.           | boolean  |
+| `--autorunHostname`     | Имя хоста автозапуска, лучше оставить на 'auto'.                                                               | string   |
+| `--autorunPortOverride` | Переопределяет порт для автозапуска.                                                                           | string   |
+| `--listen`              | SillyTavern будет прослушивать все сетевые интерфейсы. Если не указан, то используется yaml-конфиг 'listen'.   | boolean  |
+| `--corsProxy`           | Включает CORS-прокси. Если не указан, то используется yaml-конфиг 'enableCorsProxy'.                           | boolean  |
+| `--disableCsrf`         | Отключает защиту от CSRF.                                                                                      | boolean  |
+| `--ssl`                 | Включает SSL.                                                                                                  | boolean  |
+| `--certPath`            | Путь к файлу c сертификатом.                                                                                   | string   |
+| `--keyPath`             | Путь к файлу с закрытым ключом.                                                                                | string   |
+| `--whitelist`           | Включает режим белого списка.                                                                                  | boolean  |
+| `--dataRoot`            | Корневой каталог для хранения данных.                                                                          | string   |
+| `--avoidLocalhost`      | Избегает использования 'localhost' для автозапуска в режиме 'auto'.                                            | boolean  |
+| `--basicAuthMode`       | Включает простую аутентификацию.                                                                               | boolean  |
+| `--requestProxyEnabled` | Разрешает использование прокси для исходящих запросов.                                                         | boolean  |
+| `--requestProxyUrl`     | URL-адрес прокси (протоколы HTTP или SOCKS).                                                                   | string   |
+| `--requestProxyBypass`  | Bypass список прокси (список хостов, разделенных пробелами).                                                   | array    |
+
 ## Удалённое подключение
 
 В основном этим пользуются тогда, когда хотят использовать SillyTavern с телефона, запустив сервер SillyTavern на стационарном ПК в той же Wi-Fi-сети.

default/content/presets/openai/Default.json

@@ -230,7 +230,6 @@
     "show_external_models": false,
     "assistant_prefill": "",
     "assistant_impersonation": "",
-    "human_sysprompt_message": "Let's get started. Please generate your response based on the information and instructions provided above.",
     "claude_use_sysprompt": false,
     "use_alt_scale": false,
     "squash_system_messages": false,

public/index.html

@@ -1291,6 +1291,14 @@
                                         <input class="neo-range-slider" type="range" id="epsilon_cutoff_textgenerationwebui" name="volume" min="0" max="9" step="0.01">
                                         <input class="neo-range-input" type="number" min="0" max="9" step="0.01" data-for="epsilon_cutoff_textgenerationwebui" id="epsilon_cutoff_counter_textgenerationwebui">
                                     </div>
+                                    <div data-tg-type="aphrodite" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
+                                        <small>
+                                            <span data-i18n="Top nsigma">Top nsigma</span>
+                                            <div class="fa-solid fa-circle-info opacity50p" title="A sampling method that filters logits based on their statistical properties. It keeps tokens within n standard deviations of the maximum logit value, providing a simpler alternative to top-p/top-k sampling while maintaining sampling stability across different temperatures."></div>
+                                        </small>
+                                        <input class="neo-range-slider" type="range" id="nsigma_textgenerationwebui" name="volume" min="0" max="5" step="0.01">
+                                        <input class="neo-range-input" type="number" min="0" max="5" step="0.01" data-for="nsigma_textgenerationwebui" id="nsigma_counter_textgenerationwebui">
+                                    </div>
                                     <div data-tg-type="ooba,mancer,aphrodite" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
                                         <small>
                                             <span data-i18n="Eta Cutoff">Eta Cutoff</span>
@@ -1334,12 +1342,12 @@
                                         <input class="neo-range-slider" type="range" id="presence_pen_textgenerationwebui" name="volume" min="-2" max="2" step="0.01" />
                                         <input class="neo-range-input" type="number" min="-2" max="2" step="0.01" data-for="presence_pen_textgenerationwebui" id="presence_pen_counter_textgenerationwebui">
                                     </div>
-                                    <div data-tg-type="ooba" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
+                                    <div data-tg-type="aphrodite, ooba" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
                                         <small data-i18n="No Repeat Ngram Size">No Repeat Ngram Size</small>
                                         <input class="neo-range-slider" type="range" id="no_repeat_ngram_size_textgenerationwebui" name="volume" min="0" max="20" step="1">
                                         <input class="neo-range-input" type="number" min="0" max="20" step="1" data-for="no_repeat_ngram_size_textgenerationwebui" id="no_repeat_ngram_size_counter_textgenerationwebui">
                                     </div>
-                                    <div data-tg-type="tabby" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
+                                    <div data-tg-type="tabby, aphrodite" class="alignitemscenter flex-container flexFlowColumn flexBasis30p flexGrow flexShrink gap0">
                                         <small data-i18n="Skew">Skew</small>
                                         <input class="neo-range-slider" type="range" id="skew_textgenerationwebui" name="volume" min="-5" max="5" step="0.01" />
                                         <input class="neo-range-input" type="number" min="-5" max="5" step="0.01" data-for="skew_textgenerationwebui" id="skew_counter_textgenerationwebui">
@@ -1394,7 +1402,7 @@
                                         </div>
                                     </div>
 
-                                    <div data-tg-type="ooba, koboldcpp, tabby, llamacpp" id="dryBlock" class="wide100p">
+                                    <div data-tg-type="aphrodite, ooba, koboldcpp, tabby, llamacpp" id="dryBlock" class="wide100p">
                                         <h4 class="wide100p textAlignCenter" title="DRY penalizes tokens that would extend the end of the input into a sequence that has previously occurred in the input. Set multiplier to 0 to disable." data-i18n="[title]DRY_Repetition_Penalty_desc">
                                             <label data-i18n="DRY Repetition Penalty">DRY Repetition Penalty</label>
                                             <a href="https://github.com/oobabooga/text-generation-webui/pull/5677" target="_blank">
@@ -1951,15 +1959,6 @@
                                                 Send the system prompt for supported models. If disabled, the user message is added to the beginning of the prompt.
                                             </span>
                                         </div>
-                                        <div id="claude_human_sysprompt_message_block" class="wide100p">
-                                            <div class="range-block-title openai_restorable">
-                                                <span data-i18n="User first message">User first message</span>
-                                                <div id="claude_human_sysprompt_message_restore" title="Restore User first message" data-i18n="[title]Restore User first message" class="right_menu_button">
-                                                    <div class="fa-solid fa-clock-rotate-left"></div>
-                                                </div>
-                                            </div>
-                                            <textarea id="claude_human_sysprompt_textarea" class="text_pole textarea_compact autoSetHeight" rows="2" data-i18n="[placeholder]Human message" placeholder="Human message, instruction, etc.&#10;Adds nothing when empty, i.e. requires a new prompt with the role 'user'."></textarea>
-                                        </div>
                                     </div>
                                 </div>
                                 <div class="range-block m-t-1" data-source="openai,openrouter,scale,custom">
@@ -2805,9 +2804,6 @@
                                         <option value="claude-3-haiku-20240307">claude-3-haiku-20240307</option>
                                         <option value="claude-2.1">claude-2.1</option>
                                         <option value="claude-2.0">claude-2.0</option>
-                                        <option value="claude-1.3">claude-1.3</option>
-                                        <option value="claude-instant-1.2">claude-instant-1.2</option>
-                                        <option value="claude-instant-1.1">claude-instant-1.1</option>
                                     </optgroup>
                                 </select>
                             </div>

public/script.js

@@ -2707,8 +2707,7 @@ export async function generateQuietPrompt(quiet_prompt, quietToLoud, skipWIAN, q
             quietName: quietName,
         };
         originalResponseLength = responseLengthCustomized ? saveResponseLength(main_api, responseLength) : -1;
-        const generateFinished = await Generate('quiet', options);
+        return await Generate('quiet', options);
-        return generateFinished;
     } finally {
         if (responseLengthCustomized) {
             restoreResponseLength(main_api, originalResponseLength);
@@ -3363,9 +3362,9 @@ export async function generateRaw(prompt, api, instructOverride, quietToLoud, sy
 
         let data = {};
 
-        if (api == 'koboldhorde') {
+        if (api === 'koboldhorde') {
             data = await generateHorde(prompt, generateData, abortController.signal, false);
-        } else if (api == 'openai') {
+        } else if (api === 'openai') {
             data = await sendOpenAIRequest('quiet', generateData, abortController.signal);
         } else {
             const generateUrl = getGenerateUrl(api);
@@ -3378,13 +3377,15 @@ export async function generateRaw(prompt, api, instructOverride, quietToLoud, sy
             });
 
             if (!response.ok) {
-                const error = await response.json();
+                throw await response.json();
-                throw error;
             }
 
             data = await response.json();
         }
 
+        // should only happen for text completions
+        // other frontend paths do not return data if calling the backend fails,
+        // they throw things instead
         if (data.error) {
             throw new Error(data.response);
         }
@@ -4436,6 +4437,11 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
         return Promise.resolve();
     }
 
+    /**
+     * Saves itemized prompt bits and calls streaming or non-streaming generation API.
+     * @returns {Promise<void|*|Awaited<*>|String|{fromStream}|string|undefined|Object>}
+     * @throws {Error|object} Error with message text, or Error with response JSON (OAI/Horde), or the actual response JSON (novel|textgenerationwebui|kobold)
+     */
     async function finishGenerating() {
         if (power_user.console_log_prompts) {
             console.log(generate_data.prompt);
@@ -4547,6 +4553,12 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
 
     return finishGenerating().then(onSuccess, onError);
 
+    /**
+     * Handles the successful response from the generation API.
+     * @param data
+     * @returns {Promise<String|{fromStream}|*|string|string|void|Awaited<*>|undefined>}
+     * @throws {Error} Throws an error if the response data contains an error message
+     */
     async function onSuccess(data) {
         if (!data) return;
 
@@ -4556,6 +4568,7 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
 
         let messageChunk = '';
 
+        // if an error was returned in data (textgenwebui), show it and throw it
         if (data.error) {
             unblockGeneration(type);
             generatedPromptCache = '';
@@ -4670,9 +4683,15 @@ export async function Generate(type, { automatic_trigger, force_name2, quiet_pro
         return Object.defineProperty(new String(getMessage), 'messageChunk', { value: messageChunk });
     }
 
+    /**
+     * Exception handler for finishGenerating
+     * @param {Error|object} exception Error or response JSON
+     * @throws {Error|object} Re-throws the exception
+     */
     function onError(exception) {
+        // if the response JSON was thrown (novel|textgenerationwebui|kobold), show the error message
         if (typeof exception?.error?.message === 'string') {
-            toastr.error(exception.error.message, t`Error`, { timeOut: 10000, extendedTimeOut: 20000 });
+            toastr.error(exception.error.message, t`Text generation error`, { timeOut: 10000, extendedTimeOut: 20000 });
         }
 
         generatedPromptCache = '';
@@ -5340,6 +5359,7 @@ function setInContextMessages(lastmsg, type) {
  * @param {string} type Generation type
  * @param {object} data Generation data
  * @returns {Promise<object>} Response data from the API
+ * @throws {Error|object}
  */
 export async function sendGenerationRequest(type, data) {
     if (main_api === 'openai') {
@@ -5359,12 +5379,10 @@ export async function sendGenerationRequest(type, data) {
     });
 
     if (!response.ok) {
-        const error = await response.json();
+        throw await response.json();
-        throw error;
     }
 
-    const responseData = await response.json();
+    return await response.json();
-    return responseData;
 }
 
 /**
@@ -5396,6 +5414,7 @@ export async function sendStreamingRequest(type, data) {
  * Gets the generation endpoint URL for the specified API.
  * @param {string} api API name
  * @returns {string} Generation URL
+ * @throws {Error} If the API is unknown
  */
 function getGenerateUrl(api) {
     switch (api) {

public/scripts/chat-templates.js

@@ -65,6 +65,11 @@ const parse_derivation = derivation => (typeof derivation === 'string') ? {
 } : derivation;
 
 export async function deriveTemplatesFromChatTemplate(chat_template, hash) {
+    if (chat_template.trim() === '') {
+        console.log('Missing chat template.');
+        return null;
+    }
+
     if (hash in hash_derivations) {
         return parse_derivation(hash_derivations[hash]);
     }

public/scripts/extensions/stable-diffusion/index.js

@@ -2373,6 +2373,7 @@ function ensureSelectionExists(setting, selector) {
  * @param {string} [message] Chat message
  * @param {function} [callback] Callback function
  * @returns {Promise<string|undefined>} Image path
+ * @throws {Error} If the prompt or image generation fails
  */
 async function generatePicture(initiator, args, trigger, message, callback) {
     if (!trigger || trigger.trim().length === 0) {
@@ -2391,7 +2392,7 @@ async function generatePicture(initiator, args, trigger, message, callback) {
     trigger = trigger.trim();
     const generationType = getGenerationType(trigger);
     const generationTypeKey = Object.keys(generationMode).find(key => generationMode[key] === generationType);
-    console.log(`Generation mode ${generationTypeKey} triggered with "${trigger}"`);
+    console.log(`Image generation mode ${generationTypeKey} triggered with "${trigger}"`);
 
     const quietPrompt = getQuietPrompt(generationType, trigger);
     const context = getContext();
@@ -2428,6 +2429,8 @@ async function generatePicture(initiator, args, trigger, message, callback) {
 
     try {
         const combineNegatives = (prefix) => { negativePromptPrefix = combinePrefixes(negativePromptPrefix, prefix); };
+
+        // generate the text prompt for the image
         const prompt = await getPrompt(generationType, message, trigger, quietPrompt, combineNegatives);
         console.log('Processed image prompt:', prompt);
 
@@ -2438,11 +2441,16 @@ async function generatePicture(initiator, args, trigger, message, callback) {
             args._abortController.addEventListener('abort', stopListener);
         }
 
+        // generate the image
         imagePath = await sendGenerationRequest(generationType, prompt, negativePromptPrefix, characterName, callback, initiator, abortController.signal);
     } catch (err) {
         console.trace(err);
-        toastr.error('SD prompt text generation failed. Reason: ' + err, 'Image Generation');
+        // errors here are most likely due to text generation failure
-        throw new Error('SD prompt text generation failed. Reason: ' + err);
+        // sendGenerationRequest mostly deals with its own errors
+        const reason = err.error?.message || err.message || 'Unknown error';
+        const errorText = 'SD prompt text generation failed. ' + reason;
+        toastr.error(errorText, 'Image Generation');
+        throw new Error(errorText);
     }
     finally {
         $(stopButton).hide();
@@ -2513,7 +2521,7 @@ function restoreOriginalDimensions(savedParams) {
  */
 async function getPrompt(generationType, message, trigger, quietPrompt, combineNegatives) {
     let prompt;
-
+    console.log('getPrompt: Generation mode', generationType, 'triggered with', trigger);
     switch (generationType) {
         case generationMode.RAW_LAST:
             prompt = message || getRawLastMessage();
@@ -2729,7 +2737,7 @@ async function sendGenerationRequest(generationType, prompt, additionalNegativeP
             throw new Error('Endpoint did not return image data.');
         }
     } catch (err) {
-        console.error(err);
+        console.error('Image generation request error: ', err);
         toastr.error('Image generation failed. Please try again.' + '\n\n' + String(err), 'Image Generation');
         return;
     }

public/scripts/horde.js

@@ -181,6 +181,14 @@ function setContextSizePreview() {
     }
 }
 
+/** Generates text using the Horde API.
+ * @param {string} prompt
+ * @param params
+ * @param signal
+ * @param reportProgress
+ * @returns {Promise<{text: *, workerName: string}>}
+ * @throws {Error}
+ */
 async function generateHorde(prompt, params, signal, reportProgress) {
     validateHordeModel();
     delete params.prompt;

public/scripts/openai.js

@@ -99,7 +99,6 @@ const default_wi_format = '{0}';
 const default_new_chat_prompt = '[Start a new Chat]';
 const default_new_group_chat_prompt = '[Start a new group chat. Group members: {{group}}]';
 const default_new_example_chat_prompt = '[Example Chat]';
-const default_claude_human_sysprompt_message = 'Let\'s get started. Please generate your response based on the information and instructions provided above.';
 const default_continue_nudge_prompt = '[Continue the following message. Do not include ANY parts of the original message. Use capitalization and punctuation as if your reply is a part of the original message: {{lastChatMessage}}]';
 const default_bias = 'Default (none)';
 const default_personality_format = '[{{char}}\'s personality: {{personality}}]';
@@ -276,7 +275,6 @@ const default_settings = {
     proxy_password: '',
     assistant_prefill: '',
     assistant_impersonation: '',
-    human_sysprompt_message: default_claude_human_sysprompt_message,
     claude_use_sysprompt: false,
     use_makersuite_sysprompt: true,
     use_alt_scale: false,
@@ -353,7 +351,6 @@ const oai_settings = {
     proxy_password: '',
     assistant_prefill: '',
     assistant_impersonation: '',
-    human_sysprompt_message: default_claude_human_sysprompt_message,
     claude_use_sysprompt: false,
     use_makersuite_sysprompt: true,
     use_alt_scale: false,
@@ -1313,6 +1310,11 @@ export async function prepareOpenAIMessages({
     return [chat, promptManager.tokenHandler.counts];
 }
 
+/**
+ * Handles errors during streaming requests.
+ * @param {Response} response
+ * @param {string} decoded - response text or decoded stream data
+ */
 function tryParseStreamingError(response, decoded) {
     try {
         const data = JSON.parse(decoded);
@@ -1324,6 +1326,9 @@ function tryParseStreamingError(response, decoded) {
         checkQuotaError(data);
         checkModerationError(data);
 
+        // these do not throw correctly (equiv to Error("[object Object]"))
+        // if trying to fix "[object Object]" displayed to users, start here
+
         if (data.error) {
             toastr.error(data.error.message || response.statusText, 'Chat Completion API');
             throw new Error(data);
@@ -1339,15 +1344,22 @@ function tryParseStreamingError(response, decoded) {
     }
 }
 
-async function checkQuotaError(data) {
+/**
-    const errorText = await renderTemplateAsync('quotaError');
+ * Checks if the response contains a quota error and displays a popup if it does.
-
+ * @param data
+ * @returns {void}
+ * @throws {object} - response JSON
+ */
+function checkQuotaError(data) {
     if (!data) {
         return;
     }
 
     if (data.quota_error) {
-        callPopup(errorText, 'text');
+        renderTemplateAsync('quotaError').then((html) => Popup.show.text('Quota Error', html));
+
+        // this does not throw correctly (equiv to Error("[object Object]"))
+        // if trying to fix "[object Object]" displayed to users, start here
         throw new Error(data);
     }
 }
@@ -1766,6 +1778,15 @@ async function sendAltScaleRequest(messages, logit_bias, signal, type) {
     return data.output;
 }
 
+/**
+ * Send a chat completion request to backend
+ * @param {string} type (impersonate, quiet, continue, etc)
+ * @param {Array} messages
+ * @param {AbortSignal?} signal
+ * @returns {Promise<unknown>}
+ * @throws {Error}
+ */
+
 async function sendOpenAIRequest(type, messages, signal) {
     // Provide default abort signal
     if (!signal) {
@@ -1868,7 +1889,6 @@ async function sendOpenAIRequest(type, messages, signal) {
         generate_data['top_k'] = Number(oai_settings.top_k_openai);
         generate_data['claude_use_sysprompt'] = oai_settings.claude_use_sysprompt;
         generate_data['stop'] = getCustomStoppingStrings(); // Claude shouldn't have limits on stop strings.
-        generate_data['human_sysprompt_message'] = substituteParams(oai_settings.human_sysprompt_message);
         // Don't add a prefill on quiet gens (summarization) and when using continue prefill.
         if (!isQuiet && !(isContinue && oai_settings.continue_prefill)) {
             generate_data['assistant_prefill'] = isImpersonate ? substituteParams(oai_settings.assistant_impersonation) : substituteParams(oai_settings.assistant_prefill);
@@ -2028,12 +2048,13 @@ async function sendOpenAIRequest(type, messages, signal) {
     else {
         const data = await response.json();
 
-        await checkQuotaError(data);
+        checkQuotaError(data);
         checkModerationError(data);
 
         if (data.error) {
-            toastr.error(data.error.message || response.statusText, t`API returned an error`);
+            const message = data.error.message || response.statusText || t`Unknown error`;
-            throw new Error(data);
+            toastr.error(message, t`API returned an error`);
+            throw new Error(message);
         }
 
         if (type !== 'quiet') {
@@ -3005,7 +3026,6 @@ function loadOpenAISettings(data, settings) {
     oai_settings.proxy_password = settings.proxy_password ?? default_settings.proxy_password;
     oai_settings.assistant_prefill = settings.assistant_prefill ?? default_settings.assistant_prefill;
     oai_settings.assistant_impersonation = settings.assistant_impersonation ?? default_settings.assistant_impersonation;
-    oai_settings.human_sysprompt_message = settings.human_sysprompt_message ?? default_settings.human_sysprompt_message;
     oai_settings.image_inlining = settings.image_inlining ?? default_settings.image_inlining;
     oai_settings.inline_image_quality = settings.inline_image_quality ?? default_settings.inline_image_quality;
     oai_settings.bypass_status_check = settings.bypass_status_check ?? default_settings.bypass_status_check;
@@ -3045,7 +3065,6 @@ function loadOpenAISettings(data, settings) {
     $('#openai_proxy_password').val(oai_settings.proxy_password);
     $('#claude_assistant_prefill').val(oai_settings.assistant_prefill);
     $('#claude_assistant_impersonation').val(oai_settings.assistant_impersonation);
-    $('#claude_human_sysprompt_textarea').val(oai_settings.human_sysprompt_message);
     $('#openai_image_inlining').prop('checked', oai_settings.image_inlining);
     $('#openai_bypass_status_check').prop('checked', oai_settings.bypass_status_check);
 
@@ -3375,7 +3394,6 @@ async function saveOpenAIPreset(name, settings, triggerUi = true) {
         show_external_models: settings.show_external_models,
         assistant_prefill: settings.assistant_prefill,
         assistant_impersonation: settings.assistant_impersonation,
-        human_sysprompt_message: settings.human_sysprompt_message,
         claude_use_sysprompt: settings.claude_use_sysprompt,
         use_makersuite_sysprompt: settings.use_makersuite_sysprompt,
         use_alt_scale: settings.use_alt_scale,
@@ -3800,7 +3818,6 @@ function onSettingsPresetChange() {
         proxy_password: ['#openai_proxy_password', 'proxy_password', false],
         assistant_prefill: ['#claude_assistant_prefill', 'assistant_prefill', false],
         assistant_impersonation: ['#claude_assistant_impersonation', 'assistant_impersonation', false],
-        human_sysprompt_message: ['#claude_human_sysprompt_textarea', 'human_sysprompt_message', false],
         claude_use_sysprompt: ['#claude_use_sysprompt', 'claude_use_sysprompt', true],
         use_makersuite_sysprompt: ['#use_makersuite_sysprompt', 'use_makersuite_sysprompt', true],
         use_alt_scale: ['#use_alt_scale', 'use_alt_scale', true],
@@ -4652,10 +4669,6 @@ function toggleChatCompletionForms() {
         const validSources = $(this).data('source').split(',');
         $(this).toggle(validSources.includes(oai_settings.chat_completion_source));
     });
-
-    if (chat_completion_sources.CLAUDE == oai_settings.chat_completion_source) {
-        $('#claude_human_sysprompt_message_block').toggle(oai_settings.claude_use_sysprompt);
-    }
 }
 
 async function testApiConnection() {
@@ -5011,7 +5024,6 @@ export function initOpenAI() {
 
     $('#claude_use_sysprompt').on('change', function () {
         oai_settings.claude_use_sysprompt = !!$('#claude_use_sysprompt').prop('checked');
-        $('#claude_human_sysprompt_message_block').toggle(oai_settings.claude_use_sysprompt);
         saveSettingsDebounced();
     });
 
@@ -5088,12 +5100,6 @@ export function initOpenAI() {
         saveSettingsDebounced();
     });
 
-    $('#claude_human_sysprompt_message_restore').on('click', function () {
-        oai_settings.human_sysprompt_message = default_claude_human_sysprompt_message;
-        $('#claude_human_sysprompt_textarea').val(oai_settings.human_sysprompt_message);
-        saveSettingsDebounced();
-    });
-
     $('#newgroupchat_prompt_restore').on('click', function () {
         oai_settings.new_group_chat_prompt = default_new_group_chat_prompt;
         $('#newgroupchat_prompt_textarea').val(oai_settings.new_group_chat_prompt);
@@ -5185,11 +5191,6 @@ export function initOpenAI() {
         saveSettingsDebounced();
     });
 
-    $('#claude_human_sysprompt_textarea').on('input', function () {
-        oai_settings.human_sysprompt_message = String($('#claude_human_sysprompt_textarea').val());
-        saveSettingsDebounced();
-    });
-
     $('#openrouter_use_fallback').on('input', function () {
         oai_settings.openrouter_use_fallback = !!$(this).prop('checked');
         saveSettingsDebounced();

public/scripts/power-user.js

@@ -658,6 +658,10 @@ async function CreateZenSliders(elmnt) {
         numSteps = 50;
         decimals = 1;
     }
+    if (sliderID == 'nsigma') {
+        numSteps = 50;
+        decimals = 1;
+    }
     //customize steps
     if (sliderID == 'mirostat_mode_textgenerationwebui' ||
         sliderID == 'mirostat_mode_kobold') {
@@ -702,6 +706,7 @@ async function CreateZenSliders(elmnt) {
         sliderID == 'penalty_alpha_textgenerationwebui' ||
         sliderID == 'length_penalty_textgenerationwebui' ||
         sliderID == 'epsilon_cutoff_textgenerationwebui' ||
+        sliderID == 'nsigma' ||
         sliderID == 'rep_pen_range' ||
         sliderID == 'eta_cutoff_textgenerationwebui' ||
         sliderID == 'top_a_textgenerationwebui' ||

public/scripts/slash-commands/SlashCommandBrowser.js

@@ -1,5 +1,4 @@
 import { escapeRegex } from '../utils.js';
-import { SlashCommand } from './SlashCommand.js';
 import { SlashCommandParser } from './SlashCommandParser.js';
 
 export class SlashCommandBrowser {
@@ -30,7 +29,7 @@ export class SlashCommandBrowser {
                                 this.details?.remove();
                                 this.details = null;
                                 let query = inp.value.trim();
-                                if (query.slice(-1) == '"' && !/(?:^|\s+)"/.test(query)) {
+                                if (query.slice(-1) === '"' && !/(?:^|\s+)"/.test(query)) {
                                     query = `"${query}`;
                                 }
                                 let fuzzyList = [];
@@ -59,7 +58,7 @@ export class SlashCommandBrowser {
                                         cmd.helpString,
                                     ];
                                     const find = ()=>targets.find(t=>(fuzzyList.find(f=>f.test(t)) ?? quotedList.find(q=>t.includes(q))) !== undefined) !== undefined;
-                                    if (fuzzyList.length + quotedList.length == 0 || find()) {
+                                    if (fuzzyList.length + quotedList.length === 0 || find()) {
                                         this.itemMap[cmd.name].classList.remove('isFiltered');
                                     } else {
                                         this.itemMap[cmd.name].classList.add('isFiltered');
@@ -78,7 +77,7 @@ export class SlashCommandBrowser {
                         list.classList.add('autoComplete');
                         this.cmdList = Object
                             .keys(SlashCommandParser.commands)
-                            .filter(key => SlashCommandParser.commands[key].name == key) // exclude aliases
+                            .filter(key => SlashCommandParser.commands[key].name === key) // exclude aliases
                             .sort((a, b) => a.toLowerCase().localeCompare(b.toLowerCase()))
                             .map(key => SlashCommandParser.commands[key])
                         ;
@@ -97,7 +96,7 @@ export class SlashCommandBrowser {
                                         }
                                     }
                                 }
-                                if (this.details != details) {
+                                if (this.details !== details) {
                                     Array.from(list.querySelectorAll('.selected')).forEach(it=>it.classList.remove('selected'));
                                     item.classList.add('selected');
                                     this.details?.remove();
@@ -124,7 +123,7 @@ export class SlashCommandBrowser {
         parent.append(this.dom);
 
         this.mo = new MutationObserver(muts=>{
-            if (muts.find(mut=>Array.from(mut.removedNodes).find(it=>it == this.dom || it.contains(this.dom)))) {
+            if (muts.find(mut=>Array.from(mut.removedNodes).find(it=>it === this.dom || it.contains(this.dom)))) {
                 this.mo.disconnect();
                 window.removeEventListener('keydown', boundHandler);
             }
@@ -136,7 +135,7 @@ export class SlashCommandBrowser {
     }
 
     handleKeyDown(evt) {
-        if (!evt.shiftKey && !evt.altKey && evt.ctrlKey && evt.key.toLowerCase() == 'f') {
+        if (!evt.shiftKey && !evt.altKey && evt.ctrlKey && evt.key.toLowerCase() === 'f') {
             if (!this.dom.closest('body')) return;
             if (this.dom.closest('.mes') && !this.dom.closest('.last_mes')) return;
             evt.preventDefault();

public/scripts/textgen-settings.js

@@ -193,6 +193,7 @@ const settings = {
     openrouter_allow_fallbacks: true,
     xtc_threshold: 0.1,
     xtc_probability: 0,
+    nsigma: 0.0,
     featherless_model: '',
 };
 
@@ -265,6 +266,7 @@ export const setting_names = [
     'openrouter_allow_fallbacks',
     'xtc_threshold',
     'xtc_probability',
+    'nsigma',
 ];
 
 const DYNATEMP_BLOCK = document.getElementById('dynatemp_block_ooba');
@@ -880,6 +882,13 @@ function setSettingByName(setting, value, trigger) {
     }
 }
 
+/**
+ * Sends a streaming request for textgenerationwebui.
+ * @param generate_data
+ * @param signal
+ * @returns {Promise<(function(): AsyncGenerator<{swipes: [], text: string, toolCalls: [], logprobs: {token: string, topLogprobs: Candidate[]}|null}, void, *>)|*>}
+ * @throws {Error} - If the response status is not OK, or from within the generator
+ */
 async function generateTextGenWithStreaming(generate_data, signal) {
     generate_data.stream = true;
 
@@ -995,6 +1004,7 @@ export function parseTabbyLogprobs(data) {
  * @param {Response} response - Response from the server.
  * @param {string} decoded - Decoded response body.
  * @returns {void} Nothing.
+ * @throws {Error} If the response contains an error message, throws Error with the message.
  */
 function tryParseStreamingError(response, decoded) {
     let data = {};
@@ -1178,6 +1188,7 @@ export function getTextGenGenerationData(finalPrompt, maxTokens, isImpersonate,
         'sampler_order': settings.type === textgen_types.KOBOLDCPP ? settings.sampler_order : undefined,
         'xtc_threshold': settings.xtc_threshold,
         'xtc_probability': settings.xtc_probability,
+        'nsigma': settings.nsigma,
     };
     const nonAphroditeParams = {
         'rep_pen': settings.rep_pen,
@@ -1245,7 +1256,9 @@ export function getTextGenGenerationData(finalPrompt, maxTokens, isImpersonate,
         'dynatemp_exponent': dynatemp ? settings.dynatemp_exponent : undefined,
         'xtc_threshold': settings.xtc_threshold,
         'xtc_probability': settings.xtc_probability,
+        'nsigma': settings.nsigma,
         'custom_token_bans': toIntArray(banned_tokens),
+        'no_repeat_ngram_size': settings.no_repeat_ngram_size,
     };
 
     if (settings.type === OPENROUTER) {

src/endpoints/backends/chat-completions.js

@@ -102,7 +102,7 @@ async function sendClaudeRequest(request, response) {
         const additionalHeaders = {};
         const useTools = request.body.model.startsWith('claude-3') && Array.isArray(request.body.tools) && request.body.tools.length > 0;
         const useSystemPrompt = (request.body.model.startsWith('claude-2') || request.body.model.startsWith('claude-3')) && request.body.claude_use_sysprompt;
-        const convertedPrompt = convertClaudeMessages(request.body.messages, request.body.assistant_prefill, useSystemPrompt, useTools, request.body.human_sysprompt_message, request.body.char_name, request.body.user_name);
+        const convertedPrompt = convertClaudeMessages(request.body.messages, request.body.assistant_prefill, useSystemPrompt, useTools, request.body.char_name, request.body.user_name);
         // Add custom stop sequences
         const stopSequences = [];
         if (Array.isArray(request.body.stop)) {
@@ -1051,8 +1051,12 @@ router.post('/generate', jsonParser, function (request, response) {
             }
         } catch (error) {
             console.log('Generation failed', error);
+            const message = error.code === 'ECONNREFUSED'
+                ? `Connection refused: ${error.message}`
+                : error.message || 'Unknown error occurred';
+
             if (!response.headersSent) {
-                response.send({ error: true });
+                response.status(502).send({ error: { message, ...error } });
             } else {
                 response.end();
             }
@@ -1068,7 +1072,7 @@ router.post('/generate', jsonParser, function (request, response) {
 
         const message = errorResponse.statusText || 'Unknown error occurred';
         const quota_error = errorResponse.status === 429 && errorData?.error?.type === 'insufficient_quota';
-        console.log(message, responseText);
+        console.log('Chat completion request error: ', message, responseText);
 
         if (!response.headersSent) {
             response.send({ error: { message }, quota_error: quota_error });

src/endpoints/characters.js

@@ -24,6 +24,8 @@ const defaultAvatarPath = './public/img/ai4.png';
 
 // KV-store for parsed character data
 const characterDataCache = new Map();
+// Some Android devices require tighter memory management
+const isAndroid = process.platform === 'android';
 
 /**
  * Reads the character card from the specified image file.
@@ -39,7 +41,7 @@ async function readCharacterData(inputFile, inputFormat = 'png') {
     }
 
     const result = parse(inputFile, inputFormat);
-    characterDataCache.set(cacheKey, result);
+    !isAndroid && characterDataCache.set(cacheKey, result);
     return result;
 }
 

src/middleware/webpack-serve.js

@@ -6,7 +6,7 @@ import { publicLibConfig } from '../../webpack.config.js';
 export default function getWebpackServeMiddleware() {
     const compiler = webpack(publicLibConfig);
 
-    if (process.env.NODE_ENV === 'production') {
+    if (process.env.NODE_ENV === 'production' || process.platform === 'android') {
         compiler.hooks.done.tap('serve', () => {
             if (compiler.watching) {
                 compiler.watching.close(() => { });

src/prompt-converters.js

@@ -91,11 +91,10 @@ export function convertClaudePrompt(messages, addAssistantPostfix, addAssistantP
  * @param {string}   prefillString User determined prefill string
  * @param {boolean}  useSysPrompt See if we want to use a system prompt
  * @param {boolean}  useTools See if we want to use tools
- * @param {string}   humanMsgFix Add Human message between system prompt and assistant.
  * @param {string}   charName Character name
  * @param {string}   userName User name
  */
-export function convertClaudeMessages(messages, prefillString, useSysPrompt, useTools, humanMsgFix, charName = '', userName = '') {
+export function convertClaudeMessages(messages, prefillString, useSysPrompt, useTools, charName, userName) {
     let systemPrompt = [];
     if (useSysPrompt) {
         // Collect all the system messages up until the first instance of a non-system message, and then remove them from the messages array.
@@ -122,10 +121,10 @@ export function convertClaudeMessages(messages, prefillString, useSysPrompt, use
 
         // Check if the first message in the array is of type user, if not, interject with humanMsgFix or a blank message.
         // Also prevents erroring out if the messages array is empty.
-        if (messages.length === 0 || (messages.length > 0 && messages[0].role !== 'user')) {
+        if (messages.length === 0) {
             messages.unshift({
                 role: 'user',
-                content: humanMsgFix || PROMPT_PLACEHOLDER,
+                content: PROMPT_PLACEHOLDER,
             });
         }
     }