Merge branch 'staging' into pr/Cohee1207/2711

127a46e42092bcecae67618650f9b239aed1950a

RossAscends <124905043+RossAscends@users.noreply.github.com>

Parents: 93bfe7f40b747b45df7f48288188e9ad374b6db1 f1a630376086491f65762c50f260893d5404c520

default/content/presets/openai/Default.json+1 -4 Viewed

     "openrouter_force_instruct": false,
     "openrouter_group_models": false,
     "openrouter_sort_models": "alphabetically",
-    "ai21_model": "j2-ultra",
+    "ai21_model": "jamba-1.5-large",
     "mistralai_model": "mistral-large-latest",
     "custom_model": "",
     "custom_url": "",

     "temperature": 1,
     "frequency_penalty": 0,
     "presence_penalty": 0,
-    "count_penalty": 0,
     "top_p": 1,
     "top_k": 0,
     "top_a": 0,

     "assistant_prefill": "",
     "assistant_impersonation": "",
     "human_sysprompt_message": "Let's get started. Please generate your response based on the information and instructions provided above.",
-    "use_ai21_tokenizer": false,
-    "use_google_tokenizer": false,
     "claude_use_sysprompt": false,
     "use_alt_scale": false,
     "squash_system_messages": false,

default/content/settings.json+11 -5 Viewed

         "enableLabMode": false,
         "enableZenSliders": false,
         "ui_mode": 1,
-        "forbid_external_media": true
+        "forbid_external_media": true,
+        "stscript": {
+            "parser": {
+                "flags": {
+                    "1": true,
+                    "2": true
+                }
+            }
+        }
     },
     "extension_settings": {
         "apiUrl": "http://localhost:5100",

         "temp_openai": 1.0,
         "freq_pen_openai": 0,
         "pres_pen_openai": 0,
-        "count_pen": 0,
         "top_p_openai": 1,
         "top_k_openai": 0,
         "stream_openai": true,

         "wi_format": "{0}",
         "openai_model": "gpt-4-turbo",
         "claude_model": "claude-3-5-sonnet-20240620",
-        "ai21_model": "j2-ultra",
+        "ai21_model": "jamba-1.5-large",
         "windowai_model": "",
         "openrouter_model": "OR_Website",
         "jailbreak_system": true,

@@ -625,7 +632,6 @@
625	"show_external_models": false,	632	"show_external_models": false,
626	"proxy_password": "",	633	"proxy_password": "",
627	"assistant_prefill": "",	634	"assistant_prefill": "",
628	"assistant_impersonation": "",	635	"assistant_impersonation": ""
629	"use_ai21_tokenizer": false
630	}	636	}
631	}	637	}

public/index.html+44 -42 Viewed

     <title>SillyTavern</title>
     <base href="/">
     <meta charset="utf-8">
-    <meta name="viewport" content="width=device-width, viewport-fit=cover, initial-scale=1, maximum-scale=1.0, user-scalable=no">
+    <meta name="viewport" content="width=device-width, viewport-fit=cover, initial-scale=1, maximum-scale=1.0, user-scalable=no, interactive-widget=resizes-content">
     <meta name="apple-mobile-web-app-capable" content="yes">
     <meta name="darkreader-lock">
     <meta name="robots" content="noindex, nofollow" />

                                         </div>
                                     </div>
                                 </div>
-                                <div data-newbie-hidden class="range-block" data-source="openai,openrouter,ai21,custom,cohere,perplexity,groq">
+                                <div data-newbie-hidden class="range-block" data-source="openai,openrouter,custom,cohere,perplexity,groq">
                                     <div class="range-block-title" data-i18n="Frequency Penalty">
                                         Frequency Penalty
                                     </div>

                                         </div>
                                     </div>
                                 </div>
-                                <div data-newbie-hidden class="range-block" data-source="openai,openrouter,ai21,custom,cohere,perplexity,groq">
+                                <div data-newbie-hidden class="range-block" data-source="openai,openrouter,custom,cohere,perplexity,groq">
                                     <div class="range-block-title" data-i18n="Presence Penalty">
                                         Presence Penalty
                                     </div>

                                         </div>
                                     </div>
                                 </div>
-                                <div data-newbie-hidden class="range-block" data-source="ai21">
+                                <div data-newbie-hidden class="range-block" data-source="claude,openrouter,makersuite,cohere,perplexity">
-                                    <div class="range-block-title" data-i18n="Count Penalty">
-                                        Count Penalty
-                                    </div>
-                                    <div class="range-block-range-and-counter">
-                                        <div class="range-block-range">
-                                            <input type="range" id="count_pen" name="volume" min="0" max="1" step="0.01">
-                                        </div>
-                                        <div class="range-block-counter">
-                                            <input type="number" min="0" max="1" step="0.01" data-for="count_pen" id="count_pen_counter">
-                                        </div>
-                                    </div>
-                                </div>
-                                <div data-newbie-hidden class="range-block" data-source="claude,openrouter,ai21,makersuite,cohere,perplexity">
                                     <div class="range-block-title" data-i18n="Top K">
                                         Top K
                                     </div>

                                             </div>
                                         </div>
                                     </div>
+                                    <div data-newbie-hidden data-tg-type="koboldcpp" id="xtc_block" class="wide100p">
+                                        <h4 class="wide100p textAlignCenter">
+                                            <label data-i18n="Exclude Top Choices (XTC)">Exclude Top Choices (XTC)</label>
+                                            <a href="https://github.com/oobabooga/text-generation-webui/pull/6335" target="_blank">
+                                                <div class=" fa-solid fa-circle-info opacity50p"></div>
+                                            </a>
+                                        </h4>
+                                        <div class="flex-container flexFlowRow gap10px flexShrink">
+                                            <div class="alignitemscenter flex-container flexFlowColumn flexBasis48p flexGrow flexShrink gap0">
+                                                <small data-i18n="Threshold">Threshold</small>
+                                                <input class="neo-range-slider" type="range" id="xtc_threshold_textgenerationwebui" name="volume" min="0" max="0.5" step="0.01" />
+                                                <input class="neo-range-input" type="number" min="0" max="0.5" step="0.01" data-for="xtc_threshold_textgenerationwebui" id="xtc_threshold_counter_textgenerationwebui">
+                                            </div>
+                                            <div class="alignitemscenter flex-container flexFlowColumn flexBasis48p flexGrow flexShrink gap0">
+                                                <small data-i18n="Probability">Probability</small>
+                                                <input class="neo-range-slider" type="range" id="xtc_probability_textgenerationwebui" name="volume" min="0" max="1" step="0.01" />
+                                                <input class="neo-range-input" type="number" min="0" max="1" step="0.01" data-for="xtc_probability_textgenerationwebui" id="xtc_probability_counter_textgenerationwebui">
+                                            </div>
+                                        </div>
+                                    </div>
                                     <!-- Enable for llama.cpp when the PR is merged: https://github.com/ggerganov/llama.cpp/pull/6839 -->
                                     <div data-newbie-hidden data-tg-type="ooba, koboldcpp" 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">

                                     <div id="banned_tokens_block_ooba" data-newbie-hidden class="wide100p">
                                         <hr data-newbie-hidden class="width100p">
                                         <h4 class="range-block-title justifyCenter">
-                                            <span data-i18n="Banned Tokens">Banned Tokens</span>
+                                            <span data-i18n="Banned Tokens">Banned Tokens/Strings</span>
-                                            <div class="margin5 fa-solid fa-circle-info opacity50p " data-i18n="[title]LLaMA / Mistral / Yi models only" title="LLaMA / Mistral / Yi models only. Make sure to select an appropriate tokenizer first.&#13;Sequences you don't want to appear in the output.&#13;One per line. Text or [token ids].&#13;Most tokens have a leading space. Use token counter if unsure."></div>
+                                            <div class="margin5 fa-solid fa-circle-info opacity50p " data-i18n="[title]LLaMA / Mistral / Yi models only" title="Enter sequences you don't want to appear in the output.&#13;Unquoted text will be tokenized in the back end and banned as tokens.&#13;[token ids] will be banned as-is.&#13;Most tokens have a leading space. Use token counter (with the correct tokenizer selected first!) if you are unsure.&#13;Enclose text in double quotes to ban the entire string as a set.&#13;Quoted Strings and [Token ids] must be on their own line."></div>
                                         </h4>
                                         <div class="wide100p">
-                                            <textarea id="banned_tokens_textgenerationwebui" class="text_pole textarea_compact" name="banned_tokens_textgenerationwebui" rows="3" data-i18n="[placeholder]Example: some text [42, 69, 1337]" placeholder="Example:&#10;some text&#10;[42, 69, 1337]"></textarea>
+                                            <textarea id="banned_tokens_textgenerationwebui" class="text_pole textarea_compact" name="banned_tokens_textgenerationwebui" rows="3" data-i18n="[placeholder]Example: some text [42, 69, 1337]" placeholder='some text as tokens&#10;[420, 69, 1337]&#10;"Some verbatim string"'></textarea>
                                         </div>
                                     </div>
                                     <div class="range-block wide100p">

                                             </select>
                                         </div>
                                     </div>
-                                    <div class="range-block" data-source="ai21">
-                                        <label for="use_ai21_tokenizer" title="Use AI21 Tokenizer" data-i18n="[title]Use AI21 Tokenizer" class="checkbox_label widthFreeExpand">
-                                            <input id="use_ai21_tokenizer" type="checkbox" /><span data-i18n="Use AI21 Tokenizer">Use AI21 Tokenizer</span>
-                                        </label>
-                                        <div class="toggle-description justifyLeft">
-                                            <span data-i18n="Use the appropriate tokenizer for Jurassic models, which is more efficient than GPT's.">Use the appropriate tokenizer for Jurassic models, which is more efficient than GPT's.</span>
-                                        </div>
-                                    </div>
-                                    <div class="range-block" data-source="makersuite">
-                                        <label for="use_google_tokenizer" title="Use Google Tokenizer" data-i18n="[title]Use Google Tokenizer" class="checkbox_label widthFreeExpand">
-                                            <input id="use_google_tokenizer" type="checkbox" /><span data-i18n="Use Google Tokenizer">Use Google Tokenizer</span>
-                                        </label>
-                                        <div class="toggle-description justifyLeft">
-                                            <span data-i18n="Use the appropriate tokenizer for Google models via their API. Slower prompt processing, but offers much more accurate token counting.">Use the appropriate tokenizer for Google models via their API. Slower prompt processing, but offers much more accurate token counting.</span>
-                                        </div>
-                                    </div>
                                     <div class="range-block" data-source="makersuite">
                                         <label for="use_makersuite_sysprompt" class="checkbox_label widthFreeExpand">
                                             <input id="use_makersuite_sysprompt" type="checkbox" />

                                                     <div class="fa-solid fa-clock-rotate-left"></div>
                                                 </div>
                                             </div>
-                                            <textarea id="claude_human_sysprompt_textarea" class="text_pole textarea_compact" rows="4" maxlength="10000" 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>
+                                            <textarea id="claude_human_sysprompt_textarea" class="text_pole textarea_compact" rows="4" 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>
                                 <h4 data-i18n="AI21 Model">AI21 Model</h4>
                                 <select id="model_ai21_select">
-                                    <optgroup label="Latest">
+                                    <optgroup label="Jamba 1.5">
-                                        <option value="j2-ultra">j2-ultra</option>
+                                        <option value="jamba-1.5-mini">jamba-1.5-mini</option>
-                                        <option value="j2-mid">j2-mid</option>
+                                        <option value="jamba-1.5-large">jamba-1.5-large</option>
-                                        <option value="j2-light">j2-light</option>
+                                    </optgroup>
+                                    <optgroup label="Jamba-Instruct (Deprecated)">
+                                        <option value="jamba-instruct-preview">jamba-instruct-preview</option>
                                     </optgroup>
                                 </select>
                             </div>

                                     </optgroup>
                                     <optgroup label="Subversions">
                                         <option value="gemini-1.5-pro-exp-0801">Gemini 1.5 Pro Experiment 2024-08-01</option>
+                                        <option value="gemini-1.5-pro-exp-0827">Gemini 1.5 Pro Experiment 2024-08-27</option>
                                         <option value="gemini-1.5-pro-latest">Gemini 1.5 Pro [latest]</option>
                                         <option value="gemini-1.5-pro-001">Gemini 1.5 Pro [001]</option>
+                                        <option value="gemini-1.5-flash-exp-0827">Gemini 1.5 Flash Experiment 2024-08-27</option>
+                                        <option value="gemini-1.5-flash-8b-exp-0827">Gemini 1.5 Flash 8B Experiment 2024-08-27</option>
                                         <option value="gemini-1.5-flash-latest">Gemini 1.5 Flash [latest]</option>
                                         <option value="gemini-1.5-flash-001">Gemini 1.5 Flash [001]</option>
                                         <option value="gemini-1.0-pro-latest">Gemini 1.0 Pro [latest]</option>

                                         <option value="command">command</option>
                                         <option value="command-r">command-r</option>
                                         <option value="command-r-plus">command-r-plus</option>
+                                        <option value="command-r-08-2024">command-r-08-2024</option>
+                                        <option value="command-r-plus-08-2024">command-r-plus-08-2024</option>
                                     </optgroup>
                                     <optgroup label="Nightly">
                                         <option value="command-light-nightly">command-light-nightly</option>

                                     <input id="movingUImode" type="checkbox" />
                                     <small data-i18n="Movable UI Panels">MovingUI&nbsp;<i class="fa-solid fa-desktop"></i></small>
                                 </label>
-                                <div data-newbie-hidden id="movingUIreset" title="Reset MovingUI panel sizes/locations." class="menu_button margin0" data-i18n="[title]Reset MovingUI panel sizes/locations."><i class=" fa-solid fa-recycle margin-r5"></i> Reset</div>
+                                <div data-newbie-hidden id="movingUIreset" title="Reset MovingUI panel sizes/locations." class="menu_button margin0" data-i18n="[title]Reset MovingUI panel sizes/locations."><i class=" fa-solid fa-recycle margin-r5"></i> <span data-i18n="mui_reset">Reset</span></div>
                             </div>
                             <div data-newbie-hidden id="MovingUI-presets-block" class="flex-container alignitemscenter">
                                 <div class="flex-container alignItemsFlexEnd">

                                 <div data-newbie-hidden>
                                     <label for="persona_description_position" data-i18n="Position:">Position:</label>
                                     <select id="persona_description_position">
+                                        <option value="9" data-i18n="None (disabled)">None (disabled)</option>
                                         <option value="0" data-i18n="In Story String / Prompt Manager">In Story String / Prompt Manager</option>
                                         <option value="2" data-i18n="Top of Author's Note">Top of Author's Note</option>
                                         <option value="3" data-i18n="Bottom of Author's Note">Bottom of Author's Note</option>

     <script src="lib/jquery-cookie-1.4.1.min.js"></script>
     <script src="lib/jquery.ui.touch-punch.min.js"></script>
     <script src="lib/showdown.min.js"></script>
+    <script src="lib/showdown-patch.js"></script>
     <script src="lib/showdown-katex.min.js"></script>
     <script src="lib/popper.js"></script>
     <script src="lib/purify.min.js"></script>

public/lib/showdown-patch.js+24 -0 Viewed

+showdown.subParser('unhashHTMLSpans', function (text, options, globals) {
+    'use strict';
+    text = globals.converter._dispatch('unhashHTMLSpans.before', text, options, globals);
+    for (var i = 0; i < globals.gHtmlSpans.length; ++i) {
+        var repText = globals.gHtmlSpans[i],
+            // limiter to prevent infinite loop (assume 10 as limit for recurse)
+            limit = 0;
+        while (/¨C(\d+)C/.test(repText)) {
+            var num = RegExp.$1;
+            repText = repText.replace('¨C' + num + 'C', globals.gHtmlSpans[num]);
+            if (limit === 10000) {
+                console.error('maximum nesting of 10000 spans reached!!!');
+                break;
+            }
+            ++limit;
+        }
+        text = text.replace('¨C' + i + 'C', repText);
+    }
+    text = globals.converter._dispatch('unhashHTMLSpans.after', text, options, globals);
+    return text;
+});

public/locales/ru-ru.json+54 -7 Viewed

     "Smoothing Factor": "Коэффициент сглаживания",
     "No Repeat Ngram Size": "Размер no_repeat_ngram",
     "Min Length": "Мин. длина",
-    "Alternative server URL (leave empty to use the default value).": "Альтернативный URL сервера (оставьте пустым для стандартного значения)",
+    "Alternative server URL (leave empty to use the default value).": "URL альтернативного сервера (оставьте пустым для стандартного значения)",
     "Remove your real OAI API Key from the API panel BEFORE typing anything into this box": "Удалите свой личный OAI API Key из панели API, и ТОЛЬКО ПОСЛЕ ЭТОГО вводите что-то сюда",
     "We cannot provide support for problems encountered while using an unofficial OpenAI proxy": "Мы не сможем предоставить помощь с проблемами, с которыми вы столкнетесь при использовании неофициальных прокси для OpenAI",
     "Context Size (tokens)": "Размер контекста (в токенах)",

     "Generate only one line per request": "Генерировать одну строку на запрос",
     "Misc. Settings": "Доп. настройки",
     "Auto-Continue": "Авто-продолжение",
-    "Collapse Consecutive Newlines": "Сворачивать последовательные новые строки",
+    "Collapse Consecutive Newlines": "Сворачивать неск. новых строк в одну",
     "Allow for Chat Completion APIs": "Разрешить для Chat Completion API",
     "Target length (tokens)": "Целевая длина (в токенах)",
     "World Info": "Информация о мире",

     "Prefer Character Card Prompt": "Приоритет промпту из карточки персонажа",
     "Prefer Character Card Jailbreak": "Приоритет джейлбрейку из карточки персонажа",
     "Press Send to continue": "Кнопка отправки продолжает сообщение",
-    "Quick 'Continue' button": "Кнопка быстрого продолжения",
+    "Quick 'Continue' button": "Быстрое продолжение",
     "Log prompts to console": "Выводить промпты в консоль",
     "Never resize avatars": "Не менять размер аватарок",
     "Show avatar filenames": "Показывать названия файлов аватарок",

     "(Write a comma-separated list of tags)": "(Список тегов через запятую)",
     "(A brief description of the personality)": "(Краткое описание личности)",
     "(Circumstances and context of the interaction)": "(Обстоятельства и контекст этого диалога)",
-    "(Examples of chat dialog. Begin each example with START on a new line.)": "(Примеры диалога. Начинайте каждый пример с START или новой линией.)",
+    "(Examples of chat dialog. Begin each example with START on a new line.)": "(Примеры диалога. Начинайте каждый пример с START или новой строкой.)",
     "Type here...": "Пишите здесь...",
     "Comma separated (required)": "Через запятую (обязательное поле)",
     "What this keyword should mean to the AI, sent verbatim": "Что это ключевое слово должно означать для ИИ, отправляется дословно",

     "Click to set a new User Name": "Нажмите, чтобы задать новое имя пользователя.",
     "Click to lock your selected persona to the current chat. Click again to remove the lock.": "Нажмите, чтобы закрепить выбранную персону для текущего чата. Нажмите еще раз, чтобы открепить.",
     "Click to set user name for all messages": "Нажмите, чтобы задать имя пользователя для всех сообщений.",
-    "Create a dummy persona": "Создать персону-болванку",
+    "Create a dummy persona": "Создать пустую персону",
     "Character Management": "Управление персонажами",
     "Locked = Character Management panel will stay open": "Закреплено = Панель управление персонажами останется открытой",
     "Select/Create Characters": "Выбрать/Создать персонажа",

     "Automatically select a background based on the chat context": "Автоматический выбор фона в зависимости от контекста чата",
     "Filter": "Фильтр",
     "Exclude message from prompts": "Исключить сообщение из промпта",
-    "Include message in prompts": "Включить сообщение в подсказки",
+    "Include message in prompts": "Включить сообщение в промпт",
     "Create checkpoint": "Создать чекпоинт",
     "Create Branch": "Создать ветку",
     "Embed file or image": "Вставить файл или изображение",

     "ext_regex_ai_output_desc": "Сообщения, полученные от API",
     "ext_regex_sts_desc": "Сообщения, отправленные с помощью команд STscript",
     "ext_regex_wi_desc": "Содержимое лорбуков и миров. Для работы требует включения флажка \"Только промпт\"!",
-    "ext_regex_only_format_display_desc": "История чата не изменится, замена будет осуществляться только в отображаемом сообщении (в UI)"
+    "ext_regex_only_format_display_desc": "История чата не изменится, замена будет осуществляться только в отображаемом сообщении (в UI)",
+    "Prefer Character Card Instructions": "Приоритет инструкциям из карточек",
+    "If checked and the character card contains a Post-History Instructions override, use that instead": "Если в карточке персонажа имеется собственная инструкция после истории, в промпт попадёт именно она",
+    "Auto-select Input Text": "Автовыделение вводимого текста",
+    "Enable auto-select of input text in some text fields when clicking/selecting them. Applies to popup input textboxes, and possible other custom input fields.": "Автоматически выделять вводимый текст в некоторых текстовых полях при клике/выборе. Работает для вспл. окон и различных кастомных полей ввода.",
+    "Reset MovingUI panel sizes/locations.": "Сбросить расположение и размеры панелей MovingUI.",
+    "mui_reset": "Сброс",
+    "Quick 'Impersonate' button": "Быстрое перевоплощение",
+    "Show a button in the input area to ask the AI to impersonate your character for a single message": "Показать в поле ввода кнопку, по нажатии на которую ИИ сгенерирует одно сообщение от лица вашего персонажа.",
+    "Separators as Stop Strings": "Разделители как стоп-строки",
+    "Names as Stop Strings": "Имена как стоп-строки",
+    "Add Character and User names to a list of stopping strings.": "Добавлять имена персонажа и пользователя в список стоп-строк.",
+    "Allow Post-History Instructions": "Разрешить инструкции после истории",
+    "context_allow_post_history_instructions": "Добавлять в конец промпта инструкции после истории. Работает только при наличии таких инструкций в карточке И при включенной опции ''Приоритет инструкциям из карточек''.\nНЕ РЕКОМЕНДУЕТСЯ ДЛЯ МОДЕЛЕЙ TEXT COMPLETION, МОЖЕТ ПОРТИТЬ ВЫХОДНОЙ ТЕКСТ.",
+    "First User Prefix": "Первый префикс пользователя",
+    "Inserted before the first User's message.": "Вставляется перед первым сообщением пользователя.",
+    "Last User Prefix": "Последний префикс пользователя",
+    "instruct_last_input_sequence": "Вставляется перед последним сообщением пользователя.",
+    "Inserted before a User message and as a last prompt line when impersonating.": "Вставляется перед сообщением пользователя в качестве последней строки промпта при перевоплощении.",
+    "Inserted before a System (added by slash commands or extensions) message.": "Вставляется перед сообщением системы (может быть добавлено слэш-командой или расширением).",
+    "Load Asset List": "Загрузить список ресурсов",
+    "Never add character names.": "Не вставлять имя персонажа.",
+    "Don't add character names unless necessary.": "Вставлять имя персонажа только когда это необходимо.",
+    "character_names_none": "Не добавлять имена персонажей в качестве префикса. Может повредить качеству ответов в групповых чатах, используйте с осторожностью.",
+    "Auxiliary": "Вспомогательный",
+    "Post-History Instructions": "Инструкции после истории",
+    "Current persona updated": "Текущая персона изменена",
+    "Your messages will now be sent as ${0}": "Ваши сообщения будут отправляться от лица ${0}",
+    "Copied!": "Скопировано!",
+    "Are you sure you want to delete this message?": "Вы точно хотите удалить это сообщение?",
+    "Delete Message": "Удалить сообщение",
+    "Delete Swipe": "Удалить свайп",
+    "Could not get a reply from API. Check your connection settings / API key and try again.": "Не удалось получить ответ от API. Проверьте настройки соединения и API-ключ и повторите попытку.",
+    "Connecting To Proxy": "Подключение к прокси",
+    "Are you sure you want to connect to the following proxy URL?": "Вы точно хотите соединиться с прокси по этому адресу?",
+    "API connection successful!": "Соединение с API установлено!",
+    "Proxy Saved": "Прокси сохранена",
+    "Proxy Deleted": "Прокси удалена",
+    "Could not find proxy with name '${0}'": "Не удалось найти прокси с названием '${0}'",
+    "Proxy preset '${0}' not found in proxies array.": "Пресет с названием '${0}' не найден в списке прокси.",
+    "Please wait for the previous request to complete.": "Пожалуйста, дождитесь окончания обработки предыдущего запроса.",
+    "Start new chat?": "Начать новый чат?",
+    "If necessary, you can later restore this chat file from the /backups folder": "При необходимости этот чат можно будет восстановить из папки /backups",
+    "Also delete the current chat file": "Также удалить текущий чат",
+    "chat_rename_1": "Введите новое имя чата:",
+    "chat_rename_2": "!!Не используйте имя уже существующего файла, это приведёт к ошибке!!",
+    "chat_rename_3": "Будут разрушены связи между чатами-чекпоинтами.",
+    "chat_rename_4": "Расширение '.jsonl' дописывать не нужно."
+}

public/script.js+53 -56 Viewed

 export let name1 = default_user_name;
 export let name2 = 'SillyTavern System';
 export let chat = [];
-let safetychat = [
-    {
-        name: systemUserName,
-        is_user: false,
-        create_date: 0,
-        mes: 'You deleted a character/chat and arrived back here for safety reasons! Pick another character!',
-    },
-];
 let chatSaveTimeout;
 let importFlashTimeout;
 export let isChatSaving = false;

 export const MAX_INJECTION_DEPTH = 1000;
+const SAFETY_CHAT = [
+    {
+        name: systemUserName,
+        force_avatar: system_avatar,
+        is_system: true,
+        is_user: false,
+        create_date: 0,
+        mes: 'You deleted a character/chat and arrived back here for safety reasons! Pick another character!',
+    },
+];
 export let system_messages = {};
 async function getSystemMessages() {

@@ -2192,7 +2195,7 @@ export function addCopyToCodeBlocks(messageElement) {
2192	codeBlocks.get(i).appendChild(copyButton);	2195	codeBlocks.get(i).appendChild(copyButton);
2193	copyButton.addEventListener('pointerup', function (event) {	2196	copyButton.addEventListener('pointerup', function (event) {
2194	navigator.clipboard.writeText(codeBlocks.get(i).innerText);	2197	navigator.clipboard.writeText(codeBlocks.get(i).innerText);
2195	toastr.info('Copied!', '', { timeOut: 2000 });	2198	toastr.info(t`Copied!`, '', { timeOut: 2000 });
2196	});	2199	});
2197	}	2200	}
2198	}	2201	}

     const INJECT_TAG = 'PERSONA_DESCRIPTION';
     setExtensionPrompt(INJECT_TAG, '', extension_prompt_types.IN_PROMPT, 0);
-    if (!power_user.persona_description) {
+    if (!power_user.persona_description || power_user.persona_description_position === persona_description_positions.NONE) {
         return;
+    }

+}
 export function isStreamingEnabled() {
-    const noStreamSources = [chat_completion_sources.SCALE, chat_completion_sources.AI21];
+    const noStreamSources = [chat_completion_sources.SCALE];
     return ((main_api == 'openai' && oai_settings.stream_openai && !noStreamSources.includes(oai_settings.chat_completion_source) && !(oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE && oai_settings.google_model.includes('bison')))
         || (main_api == 'kobold' && kai_settings.streaming_kobold && kai_flags.can_use_streaming)
         || (main_api == 'novel' && nai_settings.streaming_novel)

     const storyStringParams = {
         description: description,
         personality: personality,
-        persona: persona,
+        persona: power_user.persona_description_position == persona_description_positions.IN_PROMPT ? persona : '',
         scenario: scenario,
         system: isInstruct ? system : '',
         char: name2,

+    }
     let examplesString = '';
-    let chatString = '';
+    let chatString = addChatsPreamble(addChatsSeparator(''));
     let cyclePrompt = '';
     async function getMessagesTokenCount() {

             storyString,
             afterScenarioAnchor,
             examplesString,
+            userAlignmentMessage,
             chatString,
-            quiet_prompt,
+            modifyLastPromptLine(''),
             cyclePrompt,
-            userAlignmentMessage,
         ].join('').replace(/\r/gm, '');
         return getTokenCountAsync(encodeString, power_user.token_padding);
+    }

+        }
         tokenCount += await getTokenCountAsync(item.replace(/\r/gm, ''));
-        chatString = item + chatString;
         if (tokenCount < this_max_context) {
+            chatString = chatString + item;
             arrMes[index] = item;
             lastAddedIndex = Math.max(lastAddedIndex, index);
         } else {

+        }
         tokenCount += await getTokenCountAsync(item.replace(/\r/gm, ''));
-        chatString = item + chatString;
         if (tokenCount < this_max_context) {
+            chatString = chatString + item;
             arrMes[i] = item;
             lastAddedIndex = Math.max(lastAddedIndex, i);
         } else {

     async function checkPromptSize() {
         console.debug('---checking Prompt size');
         setPromptString();
+        const jointMessages = mesSend.map((e) => `${e.extensionPrompts.join('')}${e.message}`).join('');
         const prompt = [
             beforeScenarioAnchor,
             storyString,
             afterScenarioAnchor,
             mesExmString,
-            mesSend.map((e) => `${e.extensionPrompts.join('')}${e.message}`).join(''),
+            addChatsPreamble(addChatsSeparator(jointMessages)),
             '\n',
+            modifyLastPromptLine(''),
             generatedPromptCache,
-            quiet_prompt,
         ].join('').replace(/\r/gm, '');
         let thisPromptContextSize = await getTokenCountAsync(prompt, power_user.token_padding);

             padding: power_user.token_padding,
             main_api: main_api,
             instruction: isInstruct ? substituteParams(power_user.prefer_character_prompt && system ? system : power_user.instruct.system_prompt) : '',
-            userPersona: (power_user.persona_description || ''),
+            userPersona: (power_user.persona_description_position == persona_description_positions.IN_PROMPT ? (persona || '') : ''),
         };
         //console.log(additionalPromptStuff);

+        }
         navigator.clipboard.writeText(rawPromptValues);
-        toastr.info('Copied!');
+        toastr.info(t`Copied!`);
     });
     popup.dlg.querySelector('#showRawPrompt').addEventListener('click', function () {

     // replaces deleted charcter name with system user since it will be displayed next.
     name2 = systemUserName;
     // sets up system user to tell user about having deleted a character
-    chat = [...safetychat];
+    chat.splice(0, chat.length, ...SAFETY_CHAT);
     // resets chat metadata
     chat_metadata = {};
     // resets the characters array, forcing getcharacters to reset

@@ -6310,7 +6314,7 @@ export function setUserName(value) {
6310	console.log(`User name changed to ${name1}`);	6314	console.log(`User name changed to ${name1}`);
6311	$('#your_name').val(name1);	6315	$('#your_name').val(name1);
6312	if (power_user.persona_show_notifications) {	6316	if (power_user.persona_show_notifications) {
6313	toastr.success(`Your messages will now be sent as ${name1}`, 'Current persona updated');	6317	toastr.success(t`Your messages will now be sent as ${name1}`, t`Current persona updated`);
6314	}	6318	}
6315	saveSettingsDebounced();	6319	saveSettingsDebounced();
6316	}	6320	}

 /**
  * Deletes a character completely, including associated chats if specified
+ *
- * @param {string} characterKey - The key (avatar) of the character to be deleted
+ * @param {string|string[]} characterKey - The key (avatar) of the character to be deleted
  * @param {Object} [options] - Optional parameters for the deletion
  * @param {boolean} [options.deleteChats=true] - Whether to delete associated chats or not
  * @return {Promise<void>} - A promise that resolves when the character is successfully deleted
  */
 export async function deleteCharacter(characterKey, { deleteChats = true } = {}) {
-    const character = characters.find(x => x.avatar == characterKey);
+    if (!Array.isArray(characterKey)) {
+        characterKey = [characterKey];
+    }
+    for (const key of characterKey) {
+        const character = characters.find(x => x.avatar == key);
         if (!character) {
-        toastr.warning(`Character ${characterKey} not found. Cannot be deleted.`);
+            toastr.warning(`Character ${key} not found. Skipping deletion.`);
-        return;
+            continue;
+        }
         const chid = characters.indexOf(character);

         });
         if (!response.ok) {
-        throw new Error(`Failed to delete character: ${response.status} ${response.statusText}`);
+            toastr.error(`${response.status} ${response.statusText}`, 'Failed to delete character');
+            continue;
+        }
-    await removeCharacterFromUI(character.name, character.avatar);
+        delete tag_map[character.avatar];
+        select_rm_info('char_delete', character.name);
         if (deleteChats) {
             for (const chat of pastChats) {

+            }
+        }
-    eventSource.emit(event_types.CHARACTER_DELETED, { id: this_chid, character: characters[this_chid] });
+        await eventSource.emit(event_types.CHARACTER_DELETED, { id: chid, character: character });
+    }
+    await removeCharacterFromUI();
+}
 /**
  * Function to delete a character from UI after character deletion API success.
  * It manages necessary UI changes such as closing advanced editing popup, unsetting
  * character ID, resetting characters array and chat metadata, deselecting character's tab
- * panel, removing character name from navigation tabs, clearing chat, removing character's
+ * panel, removing character name from navigation tabs, clearing chat, fetching updated list of characters.
- * avatar from tag_map, fetching updated list of characters and updating the 'deleted
- * character' message.
  * It also ensures to save the settings after all the operations.
- *
- * @param {string} name - The name of the character to be deleted.
- * @param {string} avatar - The avatar URL of the character to be deleted.
- * @param {boolean} reloadCharacters - Whether the character list should be refreshed after deletion.
  */
-async function removeCharacterFromUI(name, avatar, reloadCharacters = true) {
+async function removeCharacterFromUI() {
     await clearChat();
     $('#character_cross').click();
     this_chid = undefined;
     characters.length = 0;
     name2 = systemUserName;
-    chat = [...safetychat];
+    chat.splice(0, chat.length, ...SAFETY_CHAT);
     chat_metadata = {};
     $(document.getElementById('rm_button_selected_ch')).children('h2').text('');
     this_chid = undefined;
-    delete tag_map[avatar];
+    await getCharacters();
-    if (reloadCharacters) await getCharacters();
-    select_rm_info('char_delete', name);
     await printMessages();
     saveSettingsDebounced();
+}

         const oldFileNameFull = $(this).closest('.select_chat_block_wrapper').find('.select_chat_block_filename').text();
         const oldFileName = oldFileNameFull.replace('.jsonl', '');
-        const popupText = `<h3>Enter the new name for the chat:<h3>
+        const popupText = await renderTemplateAsync('chatRename');
-        <small>!!Using an existing filename will produce an error!!<br>
-        This will break the link between checkpoint chats.<br>
-        No need to add '.jsonl' at the end.<br>
-        </small>`;
         const newName = await callPopup(popupText, 'input', oldFileName);
         if (!newName || newName == oldFileName) {

         else if (id == 'option_start_new_chat') {
             if ((selected_group || this_chid !== undefined) && !is_send_press) {
                 let deleteCurrentChat = false;
-                const result = await Popup.show.confirm('Start new chat?', `
+                const result = await Popup.show.confirm(t`Start new chat?`, await renderTemplateAsync('newChatConfirm'), {
-                    <label for="del_chat_checkbox" class="checkbox_label justifyCenter"
-                    title="If necessary, you can later restore this chat file from the /backups folder">
-                        <input type="checkbox" id="del_chat_checkbox" />
-                        <small>Also delete the current chat file</small>
-                    </label>`, {
                     onClose: () => deleteCurrentChat = !!$('#del_chat_checkbox').prop('checked'),
                 });
                 if (!result) {

         let deleteOnlySwipe = false;
         if (power_user.confirm_message_delete && fromSlashCommand !== true) {
-            const result = await callGenericPopup('Are you sure you want to delete this message?', POPUP_TYPE.CONFIRM, null, {
+            const result = await callGenericPopup(t`Are you sure you want to delete this message?`, POPUP_TYPE.CONFIRM, null, {
-                okButton: canDeleteSwipe ? 'Delete Swipe' : 'Delete Message',
+                okButton: canDeleteSwipe ? t`Delete Swipe` : t`Delete Message`,
                 cancelButton: 'Cancel',
-                customButtons: canDeleteSwipe ? ['Delete Message'] : null,
+                customButtons: canDeleteSwipe ? [t`Delete Message`] : null,
             });
             if (!result) {
                 return;

public/scripts/BulkEditOverlay.js+11 -11 Viewed

      * Delete one or more characters,
      * opens a popup.
+     *
-     * @param {number} characterId
+     * @param {string|string[]} characterKey
      * @param {boolean} [deleteChats]
      * @returns {Promise<void>}
      */
-    static delete = async (characterId, deleteChats = false) => {
+    static delete = async (characterKey, deleteChats = false) => {
-        const character = CharacterContextMenu.#getCharacter(characterId);
+        await deleteCharacter(characterKey, { deleteChats: deleteChats });
-        await deleteCharacter(character.avatar, { deleteChats: deleteChats });
     };
     static #getCharacter = (characterId) => characters[characterId] ?? null;

             this.container.removeEventListener('mouseup', cancelHold);
             this.container.removeEventListener('touchend', cancelHold);
-        },
+        }, BulkEditOverlay.longPressDelay);
-            BulkEditOverlay.longPressDelay);
     };
     handleLongPressEnd = (event) => {

                 const deleteChats = document.getElementById('del_char_checkbox').checked ?? false;
                 showLoader();
-                toastr.info('We\'re deleting your characters, please wait...', 'Working on it');
+                const toast = toastr.info('We\'re deleting your characters, please wait...', 'Working on it');
-                return Promise.allSettled(characterIds.map(async characterId => CharacterContextMenu.delete(characterId, deleteChats)))
+                const avatarList = characterIds.map(id => characters[id]?.avatar).filter(a => a);
-                    .then(() => getCharacters())
+                return CharacterContextMenu.delete(avatarList, deleteChats)
                     .then(() => this.browseState())
-                    .finally(() => hideLoader());
+                    .finally(() => {
+                        toastr.clear(toast);
+                        hideLoader();
+                    });
             });
         // At this moment the popup is already changed in the dom, but not yet closed/resolved. We build the avatar list here

public/scripts/extensions/caption/settings.html+3 -0 Viewed

                         <option data-type="anthropic" value="claude-3-haiku-20240307">claude-3-haiku-20240307</option>
                         <option data-type="google" value="gemini-pro-vision">gemini-pro-vision</option>
                         <option data-type="google" value="gemini-1.5-flash-latest">gemini-1.5-flash-latest</option>
+                        <option data-type="google" value="gemini-1.5-flash-exp-0827">gemini-1.5-flash-exp-0827</option>
+                        <option data-type="google" value="gemini-1.5-flash-8b-exp-0827">gemini-1.5-flash-8b-exp-0827</option>
                         <option data-type="google" value="gemini-1.5-pro-latest">gemini-1.5-pro-latest</option>
                         <option data-type="google" value="gemini-1.5-pro-exp-0801">gemini-1.5-pro-exp-0801</option>
+                        <option data-type="google" value="gemini-1.5-pro-exp-0827">gemini-1.5-pro-exp-0827</option>
                         <option data-type="openrouter" value="openai/gpt-4-vision-preview">openai/gpt-4-vision-preview</option>
                         <option data-type="openrouter" value="openai/gpt-4o">openai/gpt-4o</option>
                         <option data-type="openrouter" value="openai/gpt-4-turbo">openai/gpt-4-turbo</option>

public/scripts/extensions/tts/index.js+1 -1 Viewed

 async function tts(text, voiceId, char) {
     async function processResponse(response) {
         // RVC injection
-        if (extension_settings.rvc.enabled && typeof window['rvcVoiceConversion'] === 'function')
+        if (typeof window['rvcVoiceConversion'] === 'function' && extension_settings.rvc.enabled)
             response = await window['rvcVoiceConversion'](response, char, text);
         await addAudioJob(response, char);

public/scripts/nai-settings.js+16 -1 Viewed

     return null;
+}
+export function getKayraMaxResponseTokens() {
+    switch (novel_data?.tier) {
+        case 1:
+            return 100;
+        case 2:
+            return 100;
+        case 3:
+            return 150;
+    }
+    return maximum_output_length;
+}
 export function getNovelTier() {
     return nai_tiers[novel_data?.tier] ?? 'no_connection';
+}

         console.log(finalPrompt);
+    }
+    const adjustedMaxLength = nai_settings.model_novel.includes('kayra') ? getKayraMaxResponseTokens() : maximum_output_length;
     return {
         'input': finalPrompt,
         'model': nai_settings.model_novel,
         'use_string': true,
         'temperature': Number(nai_settings.temperature),
-        'max_length': maxLength < maximum_output_length ? maxLength : maximum_output_length,
+        'max_length': maxLength < adjustedMaxLength ? maxLength : adjustedMaxLength,
         'min_length': Number(nai_settings.min_length),
         'tail_free_sampling': Number(nai_settings.tail_free_sampling),
         'repetition_penalty': Number(nai_settings.repetition_penalty),

public/scripts/openai.js+46 -112 Viewed

 import { renderTemplateAsync } from './templates.js';
 import { SlashCommandEnumValue } from './slash-commands/SlashCommandEnumValue.js';
 import { Popup, POPUP_RESULT } from './popup.js';
+import { t } from './i18n.js';
 export {
     openai_messages_count,

 const scale_max = 8191;
 const claude_max = 9000; // We have a proper tokenizer, so theoretically could be larger (up to 9k)
 const claude_100k_max = 99000;
-let ai21_max = 9200; //can easily fit 9k gpt tokens because j2's tokenizer is efficient af
+const unlocked_max = max_2mil;
-const unlocked_max = max_200k;
 const oai_max_temp = 2.0;
-const claude_max_temp = 1.0; //same as j2
+const claude_max_temp = 1.0;
-const j2_max_topk = 10.0;
-const j2_max_freq = 5.0;
-const j2_max_pres = 5.0;
 const openrouter_website_model = 'OR_Website';
 const openai_max_stop_strings = 4;

     'custom_include_headers',
 ];
-function getPrefixMap() {
-    return selected_group ? {
-        assistant: '',
-        user: '',
-        system: 'OOC: ',
-    }
-        : {
-            assistant: '{{char}}:',
-            user: '{{user}}:',
-            system: '',
-        };
-}
 const default_settings = {
     preset_settings_openai: 'Default',
     temp_openai: 1.0,
     freq_pen_openai: 0,
     pres_pen_openai: 0,
-    count_pen: 0.0,
     top_p_openai: 1.0,
     top_k_openai: 0,
     min_p_openai: 0,

     openai_model: 'gpt-4-turbo',
     claude_model: 'claude-3-5-sonnet-20240620',
     google_model: 'gemini-1.5-pro',
-    ai21_model: 'j2-ultra',
+    ai21_model: 'jamba-1.5-large',
     mistralai_model: 'mistral-large-latest',
     cohere_model: 'command-r-plus',
     perplexity_model: 'llama-3.1-70b-instruct',

     assistant_prefill: '',
     assistant_impersonation: '',
     human_sysprompt_message: default_claude_human_sysprompt_message,
-    use_ai21_tokenizer: false,
-    use_google_tokenizer: false,
     claude_use_sysprompt: false,
     use_makersuite_sysprompt: true,
     use_alt_scale: false,

     temp_openai: 1.0,
     freq_pen_openai: 0,
     pres_pen_openai: 0,
-    count_pen: 0.0,
     top_p_openai: 1.0,
     top_k_openai: 0,
     min_p_openai: 0,

     openai_model: 'gpt-4-turbo',
     claude_model: 'claude-3-5-sonnet-20240620',
     google_model: 'gemini-1.5-pro',
-    ai21_model: 'j2-ultra',
+    ai21_model: 'jamba-1.5-large',
     mistralai_model: 'mistral-large-latest',
     cohere_model: 'command-r-plus',
     perplexity_model: 'llama-3.1-70b-instruct',

     assistant_prefill: '',
     assistant_impersonation: '',
     human_sysprompt_message: default_claude_human_sysprompt_message,
-    use_ai21_tokenizer: false,
-    use_google_tokenizer: false,
     claude_use_sysprompt: false,
     use_makersuite_sysprompt: true,
     use_alt_scale: false,

     const rememberKey = `Proxy_SkipConfirm_${getStringHash(oai_settings.reverse_proxy)}`;
     const skipConfirm = localStorage.getItem(rememberKey) === 'true';
-    const confirmation = skipConfirm || await Popup.show.confirm('Connecting To Proxy', `<span>Are you sure you want to connect to the following proxy URL?</span><var>${DOMPurify.sanitize(oai_settings.reverse_proxy)}</var>`);
+    const confirmation = skipConfirm || await Popup.show.confirm(t`Connecting To Proxy`, await renderTemplateAsync('proxyConnectionWarning', { proxyURL: DOMPurify.sanitize(oai_settings.reverse_proxy) }));
     if (!confirmation) {
         toastr.error('Update or remove your reverse proxy settings.');

         { role: 'system', content: charDescription, identifier: 'charDescription' },
         { role: 'system', content: charPersonalityText, identifier: 'charPersonality' },
         { role: 'system', content: scenarioText, identifier: 'scenario' },
-        { role: 'system', content: personaDescription, identifier: 'personaDescription' },
         // Unordered prompts without marker
         { role: 'system', content: impersonationPrompt, identifier: 'impersonate' },
         { role: 'system', content: quietPrompt, identifier: 'quietPrompt' },

             toastr.error(data.error.message || response.statusText, 'Chat Completion API');
             throw new Error(data);
+        }
+        if (data.message) {
+            toastr.error(data.message, 'Chat Completion API');
+            throw new Error(data);
+        }
+    }
     catch {
         // No JSON. Do nothing.

     const isClaude = oai_settings.chat_completion_source == chat_completion_sources.CLAUDE;
     const isOpenRouter = oai_settings.chat_completion_source == chat_completion_sources.OPENROUTER;
     const isScale = oai_settings.chat_completion_source == chat_completion_sources.SCALE;
-    const isAI21 = oai_settings.chat_completion_source == chat_completion_sources.AI21;
     const isGoogle = oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE;
     const isOAI = oai_settings.chat_completion_source == chat_completion_sources.OPENAI;
     const isMistral = oai_settings.chat_completion_source == chat_completion_sources.MISTRALAI;

     const isQuiet = type === 'quiet';
     const isImpersonate = type === 'impersonate';
     const isContinue = type === 'continue';
-    const stream = oai_settings.stream_openai && !isQuiet && !isScale && !isAI21 && !(isGoogle && oai_settings.google_model.includes('bison'));
+    const stream = oai_settings.stream_openai && !isQuiet && !isScale && !(isGoogle && oai_settings.google_model.includes('bison'));
     const useLogprobs = !!power_user.request_token_probabilities;
     const canMultiSwipe = oai_settings.n > 1 && !isContinue && !isImpersonate && !isQuiet && (isOAI || isCustom);

         replaceItemizedPromptText(messageId, messages);
+    }
-    if (isAI21) {
-        const joinedMsgs = messages.reduce((acc, obj) => {
-            const prefix = getPrefixMap()[obj.role];
-            return acc + (prefix ? (selected_group ? '\n' : prefix + ' ') : '') + obj.content + '\n';
-        }, '');
-        messages = substituteParams(joinedMsgs) + (isImpersonate ? `${name1}:` : `${name2}:`);
-        replaceItemizedPromptText(messageId, messages);
-    }
     // If we're using the window.ai extension, use that instead
     // Doesn't support logit bias yet
     if (oai_settings.chat_completion_source == chat_completion_sources.WINDOWAI) {

         generate_data['use_makersuite_sysprompt'] = oai_settings.use_makersuite_sysprompt;
+    }
-    if (isAI21) {
-        generate_data['top_k'] = Number(oai_settings.top_k_openai);
-        generate_data['count_pen'] = Number(oai_settings.count_pen);
-        generate_data['stop_tokens'] = [name1 + ':', substituteParams(oai_settings.new_chat_prompt), substituteParams(oai_settings.new_group_chat_prompt)];
-    }
     if (isMistral) {
         generate_data['safe_prompt'] = false; // already defaults to false, but just incase they change that in the future.
+    }

     oai_settings.temp_openai = settings.temp_openai ?? default_settings.temp_openai;
     oai_settings.freq_pen_openai = settings.freq_pen_openai ?? default_settings.freq_pen_openai;
     oai_settings.pres_pen_openai = settings.pres_pen_openai ?? default_settings.pres_pen_openai;
-    oai_settings.count_pen = settings.count_pen ?? default_settings.count_pen;
     oai_settings.top_p_openai = settings.top_p_openai ?? default_settings.top_p_openai;
     oai_settings.top_k_openai = settings.top_k_openai ?? default_settings.top_k_openai;
     oai_settings.top_a_openai = settings.top_a_openai ?? default_settings.top_a_openai;

         oai_settings.names_behavior = character_names_behavior.COMPLETION;
+    }
+    if (oai_settings.ai21_model.startsWith('j2-')) {
+        oai_settings.ai21_model = 'jamba-1.5-large';
+    }
     if (settings.wrap_in_quotes !== undefined) oai_settings.wrap_in_quotes = !!settings.wrap_in_quotes;
     if (settings.openai_model !== undefined) oai_settings.openai_model = settings.openai_model;
-    if (settings.use_ai21_tokenizer !== undefined) { oai_settings.use_ai21_tokenizer = !!settings.use_ai21_tokenizer; oai_settings.use_ai21_tokenizer ? ai21_max = 8191 : ai21_max = 9200; }
-    if (settings.use_google_tokenizer !== undefined) oai_settings.use_google_tokenizer = !!settings.use_google_tokenizer;
     if (settings.claude_use_sysprompt !== undefined) oai_settings.claude_use_sysprompt = !!settings.claude_use_sysprompt;
     if (settings.use_makersuite_sysprompt !== undefined) oai_settings.use_makersuite_sysprompt = !!settings.use_makersuite_sysprompt;
     if (settings.use_alt_scale !== undefined) { oai_settings.use_alt_scale = !!settings.use_alt_scale; updateScaleForm(); }

     $('#jailbreak_system').prop('checked', oai_settings.jailbreak_system);
     $('#openai_show_external_models').prop('checked', oai_settings.show_external_models);
     $('#openai_external_category').toggle(oai_settings.show_external_models);
-    $('#use_ai21_tokenizer').prop('checked', oai_settings.use_ai21_tokenizer);
-    $('#use_google_tokenizer').prop('checked', oai_settings.use_google_tokenizer);
     $('#claude_use_sysprompt').prop('checked', oai_settings.claude_use_sysprompt);
     $('#use_makersuite_sysprompt').prop('checked', oai_settings.use_makersuite_sysprompt);
     $('#scale-alt').prop('checked', oai_settings.use_alt_scale);

     $('#pres_pen_openai').val(oai_settings.pres_pen_openai);
     $('#pres_pen_counter_openai').val(Number(oai_settings.pres_pen_openai).toFixed(2));
-    $('#count_pen').val(oai_settings.count_pen);
-    $('#count_pen_counter').val(Number(oai_settings.count_pen).toFixed(2));
     $('#top_p_openai').val(oai_settings.top_p_openai);
     $('#top_p_counter_openai').val(Number(oai_settings.top_p_openai).toFixed(2));

         temperature: settings.temp_openai,
         frequency_penalty: settings.freq_pen_openai,
         presence_penalty: settings.pres_pen_openai,
-        count_penalty: settings.count_pen,
         top_p: settings.top_p_openai,
         top_k: settings.top_k_openai,
         top_a: settings.top_a_openai,

         assistant_prefill: settings.assistant_prefill,
         assistant_impersonation: settings.assistant_impersonation,
         human_sysprompt_message: settings.human_sysprompt_message,
-        use_ai21_tokenizer: settings.use_ai21_tokenizer,
-        use_google_tokenizer: settings.use_google_tokenizer,
         claude_use_sysprompt: settings.claude_use_sysprompt,
         use_makersuite_sysprompt: settings.use_makersuite_sysprompt,
         use_alt_scale: settings.use_alt_scale,

         temperature: ['#temp_openai', 'temp_openai', false],
         frequency_penalty: ['#freq_pen_openai', 'freq_pen_openai', false],
         presence_penalty: ['#pres_pen_openai', 'pres_pen_openai', false],
-        count_penalty: ['#count_pen', 'count_pen', false],
         top_p: ['#top_p_openai', 'top_p_openai', false],
         top_k: ['#top_k_openai', 'top_k_openai', false],
         top_a: ['#top_a_openai', 'top_a_openai', 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],
-        use_ai21_tokenizer: ['#use_ai21_tokenizer', 'use_ai21_tokenizer', true],
-        use_google_tokenizer: ['#use_google_tokenizer', 'use_google_tokenizer', true],
         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],

@@ -4028,6 +3984,11 @@ async function onModelChange() {
4028	}	3984	}
4029		3985
4030	if ($(this).is('#model_ai21_select')) {	3986	if ($(this).is('#model_ai21_select')) {
		3987	if (value === '' \|\| value.startsWith('j2-')) {
		3988	value = 'jamba-1.5-large';
		3989	$('#model_ai21_select').val(value);
		3990	}
		3991
4031	console.log('AI21 model changed to', value);	3992	console.log('AI21 model changed to', value);
4032	oai_settings.ai21_model = value;	3993	oai_settings.ai21_model = value;
4033	}	3994	}

@@ -4230,7 +4191,7 @@ async function onModelChange() {
4230	else if (['command-light-nightly', 'command-nightly'].includes(oai_settings.cohere_model)) {	4191	else if (['command-light-nightly', 'command-nightly'].includes(oai_settings.cohere_model)) {
4231	$('#openai_max_context').attr('max', max_8k);	4192	$('#openai_max_context').attr('max', max_8k);
4232	}	4193	}
4233	else if (['command-r', 'command-r-plus'].includes(oai_settings.cohere_model)) {	4194	else if (oai_settings.cohere_model.includes('command-r')) {
4234	$('#openai_max_context').attr('max', max_128k);	4195	$('#openai_max_context').attr('max', max_128k);
4235	}	4196	}
4236	else if (['c4ai-aya-23'].includes(oai_settings.cohere_model)) {	4197	else if (['c4ai-aya-23'].includes(oai_settings.cohere_model)) {

     if (oai_settings.chat_completion_source == chat_completion_sources.AI21) {
         if (oai_settings.max_context_unlocked) {
             $('#openai_max_context').attr('max', unlocked_max);
-        } else {
+        } else if (oai_settings.ai21_model.includes('jamba-1.5') || oai_settings.ai21_model.includes('jamba-instruct')) {
-            $('#openai_max_context').attr('max', ai21_max);
+            $('#openai_max_context').attr('max', max_256k);
+        }
-        oai_settings.openai_max_context = Math.min(oai_settings.openai_max_context, Number($('#openai_max_context').attr('max')));
+        oai_settings.openai_max_context = Math.min(Number($('#openai_max_context').attr('max')), oai_settings.openai_max_context);
         $('#openai_max_context').val(oai_settings.openai_max_context).trigger('input');
+        $('#temp_openai').attr('max', oai_max_temp).val(oai_settings.temp_openai).trigger('input');
-        oai_settings.temp_openai = Math.min(claude_max_temp, oai_settings.temp_openai);
-        $('#temp_openai').attr('max', claude_max_temp).val(oai_settings.temp_openai).trigger('input');
-        oai_settings.freq_pen_openai = Math.min(j2_max_freq, oai_settings.freq_pen_openai < 0 ? 0 : oai_settings.freq_pen_openai);
-        $('#freq_pen_openai').attr('min', 0).attr('max', j2_max_freq).val(oai_settings.freq_pen_openai).trigger('input');
-        oai_settings.pres_pen_openai = Math.min(j2_max_pres, oai_settings.pres_pen_openai < 0 ? 0 : oai_settings.pres_pen_openai);
-        $('#pres_pen_openai').attr('min', 0).attr('max', j2_max_pres).val(oai_settings.pres_pen_openai).trigger('input');
-        oai_settings.top_k_openai = Math.min(j2_max_topk, oai_settings.top_k_openai);
-        $('#top_k_openai').attr('max', j2_max_topk).val(oai_settings.top_k_openai).trigger('input');
-    } else if (oai_settings.chat_completion_source != chat_completion_sources.AI21) {
-        oai_settings.freq_pen_openai = Math.min(2.0, oai_settings.freq_pen_openai);
-        $('#freq_pen_openai').attr('min', -2.0).attr('max', 2.0).val(oai_settings.freq_pen_openai).trigger('input');
-        oai_settings.pres_pen_openai = Math.min(2.0, oai_settings.pres_pen_openai);
-        $('#pres_pen_openai').attr('min', -2.0).attr('max', 2.0).val(oai_settings.pres_pen_openai).trigger('input');
-        oai_settings.top_k_openai = Math.min(200, oai_settings.top_k_openai);
-        $('#top_k_openai').attr('max', 200).val(oai_settings.top_k_openai).trigger('input');
+    }
     if (oai_settings.chat_completion_source == chat_completion_sources.CUSTOM) {

         $('#temp_openai').attr('max', oai_max_temp).val(oai_settings.temp_openai).trigger('input');
+    }
+    if (oai_settings.chat_completion_source === chat_completion_sources.COHERE) {
+        oai_settings.pres_pen_openai = Math.min(Math.max(0, oai_settings.pres_pen_openai), 1);
+        $('#pres_pen_openai').attr('max', 1).attr('min', 0).val(oai_settings.pres_pen_openai).trigger('input');
+        oai_settings.freq_pen_openai = Math.min(Math.max(0, oai_settings.freq_pen_openai), 1);
+        $('#freq_pen_openai').attr('max', 1).attr('min', 0).val(oai_settings.freq_pen_openai).trigger('input');
+    } else {
+        $('#pres_pen_openai').attr('max', 2).attr('min', -2).val(oai_settings.pres_pen_openai).trigger('input');
+        $('#freq_pen_openai').attr('max', 2).attr('min', -2).val(oai_settings.freq_pen_openai).trigger('input');
+    }
     $('#openai_max_context_counter').attr('max', Number($('#openai_max_context').attr('max')));
     saveSettingsDebounced();

@@ -4673,17 +4624,17 @@ function toggleChatCompletionForms() {
4673	async function testApiConnection() {	4624	async function testApiConnection() {
4674	// Check if the previous request is still in progress	4625	// Check if the previous request is still in progress
4675	if (is_send_press) {	4626	if (is_send_press) {
4676	toastr.info('Please wait for the previous request to complete.');	4627	toastr.info(t`Please wait for the previous request to complete.`);
4677	return;	4628	return;
4678	}	4629	}
4679		4630
4680	try {	4631	try {
4681	const reply = await sendOpenAIRequest('quiet', [{ 'role': 'user', 'content': 'Hi' }]);	4632	const reply = await sendOpenAIRequest('quiet', [{ 'role': 'user', 'content': 'Hi' }]);
4682	console.log(reply);	4633	console.log(reply);
4683	toastr.success('API connection successful!');	4634	toastr.success(t`API connection successful!`);
4684	}	4635	}
4685	catch (err) {	4636	catch (err) {
4686	toastr.error('Could not get a reply from API. Check your connection settings / API key and try again.');	4637	toastr.error(t`Could not get a reply from API. Check your connection settings / API key and try again.`);
4687	}	4638	}
4688	}	4639	}
4689		4640

         'gemini-1.5-flash',
         'gemini-1.5-flash-latest',
         'gemini-1.5-flash-001',
+        'gemini-1.5-flash-exp-0827',
+        'gemini-1.5-flash-8b-exp-0827',
         'gemini-1.0-pro-vision-latest',
         'gemini-1.5-pro',
         'gemini-1.5-pro-latest',
         'gemini-1.5-pro-001',
         'gemini-1.5-pro-exp-0801',
+        'gemini-1.5-pro-exp-0827',
         'gemini-pro-vision',
         'claude-3',
         'claude-3-5',

@@ -4837,7 +4791,7 @@ function onProxyPresetChange() {
4837	if (selectedPreset) {	4791	if (selectedPreset) {
4838	setProxyPreset(selectedPreset.name, selectedPreset.url, selectedPreset.password);	4792	setProxyPreset(selectedPreset.name, selectedPreset.url, selectedPreset.password);
4839	} else {	4793	} else {
4840	console.error(`Proxy preset "${value}" not found in proxies array.`);	4794	console.error(t`Proxy preset '${value}' not found in proxies array.`);
4841	}	4795	}
4842	saveSettingsDebounced();	4796	saveSettingsDebounced();
4843	}	4797	}

     setProxyPreset(presetName, reverseProxy, proxyPassword);
     saveSettingsDebounced();
-    toastr.success('Proxy Saved');
+    toastr.success(t`Proxy Saved`);
     if ($('#openai_proxy_preset').val() !== presetName) {
         const option = document.createElement('option');
         option.text = presetName;

         saveSettingsDebounced();
         $('#openai_proxy_preset').val(selected_proxy.name);
-        toastr.success('Proxy Deleted');
+        toastr.success(t`Proxy Deleted`);
     } else {
-        toastr.error(`Could not find proxy with name "${presetName}"`);
+        toastr.error(t`Could not find proxy with name '${presetName}'`);
+    }
 });

         saveSettingsDebounced();
     });
-    $('#count_pen').on('input', function () {
-        oai_settings.count_pen = Number($(this).val());
-        $('#count_pen_counter').val(Number($(this).val()).toFixed(2));
-        saveSettingsDebounced();
-    });
     $('#top_p_openai').on('input', function () {
         oai_settings.top_p_openai = Number($(this).val());
         $('#top_p_counter_openai').val(Number($(this).val()).toFixed(2));

         saveSettingsDebounced();
     });
-    $('#use_ai21_tokenizer').on('change', function () {
-        oai_settings.use_ai21_tokenizer = !!$('#use_ai21_tokenizer').prop('checked');
-        oai_settings.use_ai21_tokenizer ? ai21_max = 8191 : ai21_max = 9200;
-        oai_settings.openai_max_context = Math.min(ai21_max, oai_settings.openai_max_context);
-        $('#openai_max_context').attr('max', ai21_max).val(oai_settings.openai_max_context).trigger('input');
-        $('#openai_max_context_counter').attr('max', Number($('#openai_max_context').attr('max')));
-        saveSettingsDebounced();
-    });
-    $('#use_google_tokenizer').on('change', function () {
-        oai_settings.use_google_tokenizer = !!$('#use_google_tokenizer').prop('checked');
-        saveSettingsDebounced();
-    });
     $('#claude_use_sysprompt').on('change', function () {
         oai_settings.claude_use_sysprompt = !!$('#claude_use_sysprompt').prop('checked');
         $('#claude_human_sysprompt_message_block').toggle(oai_settings.claude_use_sysprompt);

public/scripts/power-user.js+1 -0 Viewed

     TOP_AN: 2,
     BOTTOM_AN: 3,
     AT_DEPTH: 4,
+    NONE: 9,
 };
 let power_user = {

public/scripts/templates/chatRename.html+6 -0 Viewed

+<h3 data-i18n="chat_rename_1">Enter the new name for the chat:</h3>
+<small>
+    <span data-i18n="chat_rename_2">!!Using an existing filename will produce an error!!</span><br>
+    <span data-i18n="chat_rename_3">This will break the link between checkpoint chats.</span><br>
+    <span data-i18n="chat_rename_4">No need to add '.jsonl' at the end.</span><br>
+</small>
 \ No newline at end of file

public/scripts/templates/newChatConfirm.html+4 -0 Viewed

+<label for="del_chat_checkbox" class="checkbox_label justifyCenter" data-i18n="[title]If necessary, you can later restore this chat file from the /backups folder" title="If necessary, you can later restore this chat file from the /backups folder">
+    <input type="checkbox" id="del_chat_checkbox" />
+    <small data-i18n="Also delete the current chat file">Also delete the current chat file</small>
+</label>
 \ No newline at end of file

public/scripts/templates/proxyConnectionWarning.html+1 -0 Viewed

@@ -0,0 +1 @@
		1	<span data-i18n="Are you sure you want to connect to the following proxy URL?">Are you sure you want to connect to the following proxy URL?</span><var>{{proxyURL}}</var>
	\ No newline at end of file	1	\ No newline at end of file

public/scripts/textgen-settings.js+8 -0 Viewed

     custom_model: '',
     bypass_status_check: false,
     openrouter_allow_fallbacks: true,
+    xtc_threshold: 0.1,
+    xtc_probability: 0,
 };
 export let textgenerationwebui_banned_in_macros = [];

     'custom_model',
     'bypass_status_check',
     'openrouter_allow_fallbacks',
+    'xtc_threshold',
+    'xtc_probability',
 ];
 const DYNATEMP_BLOCK = document.getElementById('dynatemp_block_ooba');

             'dry_multiplier_textgenerationwebui': 0,
             'dry_base_textgenerationwebui': 1.75,
             'dry_penalty_last_n_textgenerationwebui': 0,
+            'xtc_threshold_textgenerationwebui': 0.1,
+            'xtc_probability_textgenerationwebui': 0,
         };
         for (const [id, value] of Object.entries(inputs)) {

         'api_server': getTextGenServer(),
         'legacy_api': settings.legacy_api && (settings.type === OOBA || settings.type === APHRODITE),
         'sampler_order': settings.type === textgen_types.KOBOLDCPP ? settings.sampler_order : undefined,
+        'xtc_threshold': settings.xtc_threshold,
+        'xtc_probability': settings.xtc_probability,
     };
     const nonAphroditeParams = {
         'rep_pen': settings.rep_pen,

public/scripts/tokenizers.js+26 -23 Viewed

     CLAUDE: 11,
     LLAMA3: 12,
     GEMMA: 13,
+    JAMBA: 14,
     BEST_MATCH: 99,
 };

     tokenizers.YI,
     tokenizers.LLAMA3,
     tokenizers.GEMMA,
+    tokenizers.JAMBA,
     // uncomment when NovelAI releases Kayra and Clio weights, lol
     //tokenizers.NERD,
     //tokenizers.NERD2,

         decode: '/api/tokenizers/gemma/decode',
         count: '/api/tokenizers/gemma/encode',
     },
+    [tokenizers.JAMBA]: {
+        encode: '/api/tokenizers/jamba/encode',
+        decode: '/api/tokenizers/jamba/decode',
+        count: '/api/tokenizers/jamba/encode',
+    },
     [tokenizers.API_TEXTGENERATIONWEBUI]: {
         encode: '/api/tokenizers/remote/textgenerationwebui/encode',
         count: '/api/tokenizers/remote/textgenerationwebui/encode',

         tokenizerId: Number(tokenizerOption.value),
         tokenizerKey: Object.entries(tokenizers).find(([_, value]) => value === Number(tokenizerOption.value))[0].toLocaleLowerCase(),
         tokenizerName: tokenizerOption.text,
-    }))
+    }));
+}
 /**

             if (model.includes('gemma')) {
                 return tokenizers.GEMMA;
+            }
+            if (model.includes('yi')) {
+                return tokenizers.YI;
+            }
+            if (model.includes('jamba')) {
+                return tokenizers.JAMBA;
+            }
+        }
         return tokenizers.LLAMA;

     const mistralTokenizer = 'mistral';
     const yiTokenizer = 'yi';
     const gemmaTokenizer = 'gemma';
+    const jambaTokenizer = 'jamba';
     // Assuming no one would use it for different models.. right?
     if (oai_settings.chat_completion_source == chat_completion_sources.SCALE) {

@@ -562,12 +576,19 @@ export function getTokenizerModel() {
562	else if (oai_settings.openrouter_model.includes('GPT-NeoXT')) {	576	else if (oai_settings.openrouter_model.includes('GPT-NeoXT')) {
563	return gpt2Tokenizer;	577	return gpt2Tokenizer;
564	}	578	}
		579	else if (oai_settings.openrouter_model.includes('jamba')) {
		580	return jambaTokenizer;
		581	}
565	}	582	}
566		583
567	if (oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE) {	584	if (oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE) {
568	return gemmaTokenizer;	585	return gemmaTokenizer;
569	}	586	}
570		587
		588	if (oai_settings.chat_completion_source == chat_completion_sources.AI21) {
		589	return jambaTokenizer;
		590	}
		591
571	if (oai_settings.chat_completion_source == chat_completion_sources.CLAUDE) {	592	if (oai_settings.chat_completion_source == chat_completion_sources.CLAUDE) {
572	return claudeTokenizer;	593	return claudeTokenizer;
573	}	594	}

  * @deprecated Use countTokensOpenAIAsync instead.
  */
 export function countTokensOpenAI(messages, full = false) {
-    const shouldTokenizeAI21 = oai_settings.chat_completion_source === chat_completion_sources.AI21 && oai_settings.use_ai21_tokenizer;
+    const tokenizerEndpoint = `/api/tokenizers/openai/count?model=${getTokenizerModel()}`;
-    const shouldTokenizeGoogle = oai_settings.chat_completion_source === chat_completion_sources.MAKERSUITE && oai_settings.use_google_tokenizer;
-    let tokenizerEndpoint = '';
-    if (shouldTokenizeAI21) {
-        tokenizerEndpoint = '/api/tokenizers/ai21/count';
-    } else if (shouldTokenizeGoogle) {
-        tokenizerEndpoint = `/api/tokenizers/google/count?model=${getTokenizerModel()}&reverse_proxy=${oai_settings.reverse_proxy}&proxy_password=${oai_settings.proxy_password}`;
-    } else {
-        tokenizerEndpoint = `/api/tokenizers/openai/count?model=${getTokenizerModel()}`;
-    }
     const cacheObject = getTokenCacheObject();
     if (!Array.isArray(messages)) {

     for (const message of messages) {
         const model = getTokenizerModel();
-        if (model === 'claude' || shouldTokenizeAI21 || shouldTokenizeGoogle) {
+        if (model === 'claude') {
             full = true;
+        }

  * @returns {Promise<number>} Token count.
  */
 export async function countTokensOpenAIAsync(messages, full = false) {
-    const shouldTokenizeAI21 = oai_settings.chat_completion_source === chat_completion_sources.AI21 && oai_settings.use_ai21_tokenizer;
+    const tokenizerEndpoint = `/api/tokenizers/openai/count?model=${getTokenizerModel()}`;
-    const shouldTokenizeGoogle = oai_settings.chat_completion_source === chat_completion_sources.MAKERSUITE && oai_settings.use_google_tokenizer;
-    let tokenizerEndpoint = '';
-    if (shouldTokenizeAI21) {
-        tokenizerEndpoint = '/api/tokenizers/ai21/count';
-    } else if (shouldTokenizeGoogle) {
-        tokenizerEndpoint = `/api/tokenizers/google/count?model=${getTokenizerModel()}`;
-    } else {
-        tokenizerEndpoint = `/api/tokenizers/openai/count?model=${getTokenizerModel()}`;
-    }
     const cacheObject = getTokenCacheObject();
     if (!Array.isArray(messages)) {

     for (const message of messages) {
         const model = getTokenizerModel();
-        if (model === 'claude' || shouldTokenizeAI21 || shouldTokenizeGoogle) {
+        if (model === 'claude') {
             full = true;
+        }

public/scripts/world-info.js+6 -2 Viewed

         world_info_min_activations = Number($(this).val());
         $('#world_info_min_activations_counter').val(world_info_min_activations);
-        if (world_info_min_activations !== 0) {
+        if (world_info_min_activations !== 0 && world_info_max_recursion_steps !== 0) {
             $('#world_info_max_recursion_steps').val(0).trigger('input');
+            flashHighlight($('#world_info_max_recursion_steps').parent()); // flash the other control to show it has changed
+            console.info('[WI] Max recursion steps set to 0, as min activations is set to', world_info_min_activations);
         } else {
             saveSettings();
+        }

     $('#world_info_max_recursion_steps').on('input', function () {
         world_info_max_recursion_steps = Number($(this).val());
         $('#world_info_max_recursion_steps_counter').val(world_info_max_recursion_steps);
-        if (world_info_max_recursion_steps !== 0) {
+        if (world_info_max_recursion_steps !== 0 && world_info_min_activations !== 0) {
             $('#world_info_min_activations').val(0).trigger('input');
+            flashHighlight($('#world_info_min_activations').parent()); // flash the other control to show it has changed
+            console.info('[WI] Min activations set to 0, as max recursion steps is set to', world_info_max_recursion_steps);
         } else {
             saveSettings();
+        }

public/style.css+0 -1 Viewed

+}
 #rm_print_characters_block .text_block {
-    height: 100%;
     width: 100%;
     opacity: 0.5;
     margin: 0 auto 1px auto;

src/cohere-stream.js+126 -0 Viewed

+const DATA_PREFIX = 'data:';
+/**
+ * Borrowed from Cohere SDK (MIT License)
+ * https://github.com/cohere-ai/cohere-typescript/blob/main/src/core/streaming-fetcher/Stream.ts
+ * Copyright (c) 2021 Cohere
+ */
+class CohereStream {
+    /** @type {ReadableStream} */
+    stream;
+    /** @type {string} */
+    prefix;
+    /** @type {string} */
+    messageTerminator;
+    /** @type {string|undefined} */
+    streamTerminator;
+    /** @type {AbortController} */
+    controller = new AbortController();
+    constructor({ stream, eventShape }) {
+        this.stream = stream;
+        if (eventShape.type === 'sse') {
+            this.prefix = DATA_PREFIX;
+            this.messageTerminator = '\n';
+            this.streamTerminator = eventShape.streamTerminator;
+        } else {
+            this.messageTerminator = eventShape.messageTerminator;
+        }
+    }
+    async *iterMessages()  {
+        const stream = readableStreamAsyncIterable(this.stream);
+        let buf = '';
+        let prefixSeen = false;
+        let parsedAnyMessages = false;
+        for await (const chunk of stream) {
+            buf += this.decodeChunk(chunk);
+            let terminatorIndex;
+            // Parse the chunk into as many messages as possible
+            while ((terminatorIndex = buf.indexOf(this.messageTerminator)) >= 0) {
+                // Extract the line from the buffer
+                let line = buf.slice(0, terminatorIndex + 1);
+                buf = buf.slice(terminatorIndex + 1);
+                // Skip empty lines
+                if (line.length === 0) {
+                    continue;
+                }
+                // Skip the chunk until the prefix is found
+                if (!prefixSeen && this.prefix != null) {
+                    const prefixIndex = line.indexOf(this.prefix);
+                    if (prefixIndex === -1) {
+                        continue;
+                    }
+                    prefixSeen = true;
+                    line = line.slice(prefixIndex + this.prefix.length);
+                }
+                // If the stream terminator is present, return
+                if (this.streamTerminator != null && line.includes(this.streamTerminator)) {
+                    return;
+                }
+                // Otherwise, yield message from the prefix to the terminator
+                const message = JSON.parse(line);
+                yield message;
+                prefixSeen = false;
+                parsedAnyMessages = true;
+            }
+        }
+        if (!parsedAnyMessages && buf.length > 0) {
+            try {
+                yield JSON.parse(buf);
+            } catch (e) {
+                console.error('Error parsing message:', e);
+            }
+        }
+    }
+    async *[Symbol.asyncIterator]() {
+        for await (const message of this.iterMessages()) {
+            yield message;
+        }
+    }
+    decodeChunk(chunk) {
+        const decoder = new TextDecoder('utf8');
+        return decoder.decode(chunk);
+    }
+}
+function readableStreamAsyncIterable(stream) {
+    if (stream[Symbol.asyncIterator]) {
+        return stream;
+    }
+    const reader = stream.getReader();
+    return {
+        async next() {
+            try {
+                const result = await reader.read();
+                if (result?.done) {
+                    reader.releaseLock();
+                } // release lock when stream becomes closed
+                return result;
+            } catch (e) {
+                reader.releaseLock(); // release lock when stream becomes errored
+                throw e;
+            }
+        },
+        async return() {
+            const cancelPromise = reader.cancel();
+            reader.releaseLock();
+            await cancelPromise;
+            return { done: true, value: undefined };
+        },
+        [Symbol.asyncIterator]() {
+            return this;
+        },
+    };
+}
+module.exports = CohereStream;

src/endpoints/backends/chat-completions.js+56 -86 Viewed

 const express = require('express');
 const fetch = require('node-fetch').default;
-const Readable = require('stream').Readable;
 const { jsonParser } = require('../../express-common');
 const { CHAT_COMPLETION_SOURCES, GEMINI_SAFETY, BISON_SAFETY, OPENROUTER_HEADERS } = require('../../constants');
 const { forwardFetchResponse, getConfigValue, tryParse, uuidv4, mergeObjectWithYaml, excludeKeysByYaml, color } = require('../../util');
-const { convertClaudeMessages, convertGooglePrompt, convertTextCompletionPrompt, convertCohereMessages, convertMistralMessages, convertCohereTools } = require('../../prompt-converters');
+const { convertClaudeMessages, convertGooglePrompt, convertTextCompletionPrompt, convertCohereMessages, convertMistralMessages, convertCohereTools, convertAI21Messages } = require('../../prompt-converters');
+const CohereStream = require('../../cohere-stream');
 const { readSecret, SECRET_KEYS } = require('../secrets');
 const { getTokenizerModel, getSentencepiceTokenizer, getTiktokenTokenizer, sentencepieceTokenizers, TEXT_COMPLETION_MODELS } = require('../tokenizers');

 const API_MAKERSUITE = 'https://generativelanguage.googleapis.com';
 const API_01AI = 'https://api.01.ai/v1';
 const API_BLOCKENTROPY = 'https://api.blockentropy.ai/v1';
+const API_AI21 = 'https://api.ai21.com/studio/v1';
 /**
  * Applies a post-processing step to the generated messages.

 /**
  * Ollama strikes back. Special boy #2's steaming routine.
  * Wrap this abomination into proper SSE stream, again.
- * @param {import('node-fetch').Response} jsonStream JSON stream
+ * @param {Response} jsonStream JSON stream
  * @param {import('express').Request} request Express request
  * @param {import('express').Response} response Express response
  * @returns {Promise<any>} Nothing valuable
  */
 async function parseCohereStream(jsonStream, request, response) {
     try {
-        let partialData = '';
+        const stream = new CohereStream({ stream: jsonStream.body, eventShape: { type: 'json', messageTerminator: '\n' } });
-        jsonStream.body.on('data', (data) => {
-            const chunk = data.toString();
+        for await (const json of stream.iterMessages()) {
-            partialData += chunk;
-            while (true) {
-                let json;
-                try {
-                    json = JSON.parse(partialData);
-                } catch (e) {
-                    break;
-                }
             if (json.message) {
                 const message = json.message || 'Unknown error';
                 const chunk = { error: { message: message } };
                 response.write(`data: ${JSON.stringify(chunk)}\n\n`);
-                    partialData = '';
-                    break;
             } else if (json.event_type === 'text-generation') {
                 const text = json.text || '';
                 const chunk = { choices: [{ text }] };
                 response.write(`data: ${JSON.stringify(chunk)}\n\n`);
-                    partialData = '';
-                } else {
-                    partialData = '';
-                    break;
+            }
+        }
-        });
-        request.socket.on('close', function () {
-            if (jsonStream.body instanceof Readable) jsonStream.body.destroy();
-            response.end();
-        });
-        jsonStream.body.on('end', () => {
         console.log('Streaming request finished');
         response.write('data: [DONE]\n\n');
         response.end();
-        });
     } catch (error) {
         console.log('Error forwarding streaming response:', error);
         if (!response.headersSent) {

         if (!generateResponse.ok) {
             console.log(`Scale API returned error: ${generateResponse.status} ${generateResponse.statusText} ${await generateResponse.text()}`);
-            return response.status(generateResponse.status).send({ error: true });
+            return response.status(500).send({ error: true });
+        }
         const generateResponseJson = await generateResponse.json();

     request.socket.on('close', function () {
         controller.abort();
     });
+    const convertedPrompt = convertAI21Messages(request.body.messages, request.body.char_name, request.body.user_name);
+    const body = {
+        messages: convertedPrompt,
+        model: request.body.model,
+        max_tokens: request.body.max_tokens,
+        temperature: request.body.temperature,
+        top_p: request.body.top_p,
+        stop: request.body.stop,
+        stream: request.body.stream,
+    };
     const options = {
         method: 'POST',
         headers: {

             'content-type': 'application/json',
             Authorization: `Bearer ${readSecret(request.user.directories, SECRET_KEYS.AI21)}`,
         },
-        body: JSON.stringify({
+        body: JSON.stringify(body),
-            numResults: 1,
-            maxTokens: request.body.max_tokens,
-            minTokens: 0,
-            temperature: request.body.temperature,
-            topP: request.body.top_p,
-            stopSequences: request.body.stop_tokens,
-            topKReturn: request.body.top_k,
-            frequencyPenalty: {
-                scale: request.body.frequency_penalty * 100,
-                applyToWhitespaces: false,
-                applyToPunctuations: false,
-                applyToNumbers: false,
-                applyToStopwords: false,
-                applyToEmojis: false,
-            },
-            presencePenalty: {
-                scale: request.body.presence_penalty,
-                applyToWhitespaces: false,
-                applyToPunctuations: false,
-                applyToNumbers: false,
-                applyToStopwords: false,
-                applyToEmojis: false,
-            },
-            countPenalty: {
-                scale: request.body.count_pen,
-                applyToWhitespaces: false,
-                applyToPunctuations: false,
-                applyToNumbers: false,
-                applyToStopwords: false,
-                applyToEmojis: false,
-            },
-            prompt: request.body.messages,
-        }),
         signal: controller.signal,
     };
-    fetch(`https://api.ai21.com/studio/v1/${request.body.model}/complete`, options)
+    console.log('AI21 request:', body);
-        .then(r => r.json())
-        .then(r => {
+    try {
-            if (r.completions === undefined) {
+        const generateResponse = await fetch(API_AI21 + '/chat/completions', options);
-                console.log(r);
+        if (request.body.stream) {
+            forwardFetchResponse(generateResponse, response);
         } else {
-                console.log(r.completions[0].data.text);
+            if (!generateResponse.ok) {
+                const errorText = await generateResponse.text();
+                console.log(`AI21 API returned error: ${generateResponse.status} ${generateResponse.statusText} ${errorText}`);
+                const errorJson = tryParse(errorText) ?? { error: true };
+                return response.status(500).send(errorJson);
+            }
+            const generateResponseJson = await generateResponse.json();
+            console.log('AI21 response:', generateResponseJson);
+            return response.send(generateResponseJson);
+        }
+    } catch (error) {
+        console.log('Error communicating with AI21 API: ', error);
+        if (!response.headersSent) {
+            response.send({ error: true });
+        } else {
+            response.end();
+        }
+    }
-            const reply = { choices: [{ 'message': { 'content': r.completions?.[0]?.data?.text } }] };
-            return response.send(reply);
-        })
-        .catch(err => {
-            console.error(err);
-            return response.send({ error: true });
-        });
+}
 /**

             forwardFetchResponse(generateResponse, response);
         } else {
             if (!generateResponse.ok) {
-                console.log(`MistralAI API returned error: ${generateResponse.status} ${generateResponse.statusText} ${await generateResponse.text()}`);
+                const errorText = await generateResponse.text();
-                // a 401 unauthorized response breaks the frontend auth, so return a 500 instead. prob a better way of dealing with this.
+                console.log(`MistralAI API returned error: ${generateResponse.status} ${generateResponse.statusText} ${errorText}`);
-                // 401s are already handled by the streaming processor and dont pop up an error toast, that should probably be fixed too.
+                const errorJson = tryParse(errorText) ?? { error: true };
-                return response.status(generateResponse.status === 401 ? 500 : generateResponse.status).send({ error: true });
+                return response.status(500).send(errorJson);
+            }
             const generateResponseJson = await generateResponse.json();
             console.log('MistralAI response:', generateResponseJson);

             search_queries_only: false,
         };
+        const canDoSafetyMode = String(request.body.model).endsWith('08-2024');
+        if (canDoSafetyMode) {
+            requestBody.safety_mode = 'NONE';
+        }
         console.log('Cohere request:', requestBody);
         const config = {

         const apiUrl = API_COHERE + '/chat';
         if (request.body.stream) {
-            const stream = await fetch(apiUrl, config);
+            const stream = await global.fetch(apiUrl, config);
             parseCohereStream(stream, request, response);
         } else {
             const generateResponse = await fetch(apiUrl, config);
             if (!generateResponse.ok) {
-                console.log(`Cohere API returned error: ${generateResponse.status} ${generateResponse.statusText} ${await generateResponse.text()}`);
+                const errorText = await generateResponse.text();
-                // a 401 unauthorized response breaks the frontend auth, so return a 500 instead. prob a better way of dealing with this.
+                console.log(`Cohere API returned error: ${generateResponse.status} ${generateResponse.statusText} ${errorText}`);
-                // 401s are already handled by the streaming processor and dont pop up an error toast, that should probably be fixed too.
+                const errorJson = tryParse(errorText) ?? { error: true };
-                return response.status(generateResponse.status === 401 ? 500 : generateResponse.status).send({ error: true });
+                return response.status(500).send(errorJson);
+            }
             const generateResponseJson = await generateResponse.json();
             console.log('Cohere response:', generateResponseJson);

src/endpoints/novelai.js+5 -5 Viewed

 const { jsonParser } = require('../express-common');
 const API_NOVELAI = 'https://api.novelai.net';
+const TEXT_NOVELAI = 'https://text.novelai.net';
 const IMAGE_NOVELAI = 'https://image.novelai.net';
 // Ban bracket generation, plus defaults

             'repetition_penalty_slope': req.body.repetition_penalty_slope,
             'repetition_penalty_frequency': req.body.repetition_penalty_frequency,
             'repetition_penalty_presence': req.body.repetition_penalty_presence,
-            'repetition_penalty_whitelist': isNewModel ? repPenaltyAllowList : null,
+            'repetition_penalty_whitelist': isNewModel ? repPenaltyAllowList.flat() : null,
             'top_a': req.body.top_a,
             'top_p': req.body.top_p,
             'top_k': req.body.top_k,

     };
     // Tells the model to stop generation at '>'
-    if ('theme_textadventure' === req.body.prefix &&
+    if ('theme_textadventure' === req.body.prefix && isNewModel) {
-        (true === req.body.model.includes('clio') ||
-         true === req.body.model.includes('kayra'))) {
         data.parameters.eos_token_id = 49405;
+    }

     };
     try {
-        const url = req.body.streaming ? `${API_NOVELAI}/ai/generate-stream` : `${API_NOVELAI}/ai/generate`;
+        const baseURL = req.body.model.includes('kayra') ? TEXT_NOVELAI : API_NOVELAI;
+        const url = req.body.streaming ? `${baseURL}/ai/generate-stream` : `${baseURL}/ai/generate`;
         const response = await fetch(url, { method: 'POST', timeout: 0, ...args });
         if (req.body.streaming) {

src/endpoints/tokenizers.js+27 -47 Viewed

 const spp_mistral = new SentencePieceTokenizer('src/tokenizers/mistral.model');
 const spp_yi = new SentencePieceTokenizer('src/tokenizers/yi.model');
 const spp_gemma = new SentencePieceTokenizer('src/tokenizers/gemma.model');
+const spp_jamba = new SentencePieceTokenizer('src/tokenizers/jamba.model');
 const claude_tokenizer = new WebTokenizer('src/tokenizers/claude.json');
 const llama3_tokenizer = new WebTokenizer('src/tokenizers/llama3.json');

     'mistral',
     'yi',
     'gemma',
+    'jamba',
 ];
 /**

@@ -186,6 +188,10 @@ function getSentencepiceTokenizer(model) {
186	return spp_gemma;	188	return spp_gemma;
187	}	189	}
188		190
		191	if (model.includes('jamba')) {
		192	return spp_jamba;
		193	}
		194
189	return null;	195	return null;
190	}	196	}
191		197

@@ -322,6 +328,10 @@ function getTokenizerModel(requestModel) {
322	return 'gemma';	328	return 'gemma';
323	}	329	}
324		330
		331	if (requestModel.includes('jamba')) {
		332	return 'jamba';
		333	}
		334
325	// default	335	// default
326	return 'gpt-3.5-turbo';	336	return 'gpt-3.5-turbo';
327	}	337	}

 const router = express.Router();
-router.post('/ai21/count', jsonParser, async function (req, res) {
-    if (!req.body) return res.sendStatus(400);
-    const key = readSecret(req.user.directories, SECRET_KEYS.AI21);
-    const options = {
-        method: 'POST',
-        headers: {
-            accept: 'application/json',
-            'content-type': 'application/json',
-            Authorization: `Bearer ${key}`,
-        },
-        body: JSON.stringify({ text: req.body[0].content }),
-    };
-    try {
-        const response = await fetch('https://api.ai21.com/studio/v1/tokenize', options);
-        const data = await response.json();
-        return res.send({ 'token_count': data?.tokens?.length || 0 });
-    } catch (err) {
-        console.error(err);
-        return res.send({ 'token_count': 0 });
-    }
-});
-router.post('/google/count', jsonParser, async function (req, res) {
-    if (!req.body) return res.sendStatus(400);
-    const options = {
-        method: 'POST',
-        headers: {
-            accept: 'application/json',
-            'content-type': 'application/json',
-        },
-        body: JSON.stringify({ contents: convertGooglePrompt(req.body, String(req.query.model)).contents }),
-    };
-    try {
-        const reverseProxy = req.query.reverse_proxy?.toString() || '';
-        const proxyPassword = req.query.proxy_password?.toString() || '';
-        const apiKey = reverseProxy ? proxyPassword : readSecret(req.user.directories, SECRET_KEYS.MAKERSUITE);
-        const apiUrl = new URL(reverseProxy || API_MAKERSUITE);
-        const response = await fetch(`${apiUrl.origin}/v1beta/models/${req.query.model}:countTokens?key=${apiKey}`, options);
-        const data = await response.json();
-        return res.send({ 'token_count': data?.totalTokens || 0 });
-    } catch (err) {
-        console.error(err);
-        return res.send({ 'token_count': 0 });
-    }
-});
 router.post('/llama/encode', jsonParser, createSentencepieceEncodingHandler(spp_llama));
 router.post('/nerdstash/encode', jsonParser, createSentencepieceEncodingHandler(spp_nerd));
 router.post('/nerdstash_v2/encode', jsonParser, createSentencepieceEncodingHandler(spp_nerd_v2));
 router.post('/mistral/encode', jsonParser, createSentencepieceEncodingHandler(spp_mistral));
 router.post('/yi/encode', jsonParser, createSentencepieceEncodingHandler(spp_yi));
 router.post('/gemma/encode', jsonParser, createSentencepieceEncodingHandler(spp_gemma));
+router.post('/jamba/encode', jsonParser, createSentencepieceEncodingHandler(spp_jamba));
 router.post('/gpt2/encode', jsonParser, createTiktokenEncodingHandler('gpt2'));
 router.post('/claude/encode', jsonParser, createWebTokenizerEncodingHandler(claude_tokenizer));
 router.post('/llama3/encode', jsonParser, createWebTokenizerEncodingHandler(llama3_tokenizer));

 router.post('/mistral/decode', jsonParser, createSentencepieceDecodingHandler(spp_mistral));
 router.post('/yi/decode', jsonParser, createSentencepieceDecodingHandler(spp_yi));
 router.post('/gemma/decode', jsonParser, createSentencepieceDecodingHandler(spp_gemma));
+router.post('/jamba/decode', jsonParser, createSentencepieceDecodingHandler(spp_jamba));
 router.post('/gpt2/decode', jsonParser, createTiktokenDecodingHandler('gpt2'));
 router.post('/claude/decode', jsonParser, createWebTokenizerDecodingHandler(claude_tokenizer));
 router.post('/llama3/decode', jsonParser, createWebTokenizerDecodingHandler(llama3_tokenizer));

             return handler(req, res);
+        }
+        if (queryModel.includes('jamba')) {
+            const handler = createSentencepieceEncodingHandler(spp_jamba);
+            return handler(req, res);
+        }
         const model = getTokenizerModel(queryModel);
         const handler = createTiktokenEncodingHandler(model);
         return handler(req, res);

             return handler(req, res);
+        }
+        if (queryModel.includes('jamba')) {
+            const handler = createSentencepieceDecodingHandler(spp_jamba);
+            return handler(req, res);
+        }
         const model = getTokenizerModel(queryModel);
         const handler = createTiktokenDecodingHandler(model);
         return handler(req, res);

             return res.send({ 'token_count': num_tokens });
+        }
+        if (model === 'jamba') {
+            num_tokens = await countSentencepieceArrayTokens(spp_jamba, req.body);
+            return res.send({ 'token_count': num_tokens });
+        }
         const tokensPerName = queryModel.includes('gpt-3.5-turbo-0301') ? -1 : 1;
         const tokensPerMessage = queryModel.includes('gpt-3.5-turbo-0301') ? 4 : 3;
         const tokensPadding = 3;

src/endpoints/translate.js+1 -1 Viewed

 });
 router.post('/yandex', jsonParser, async (request, response) => {
+    try {
         const chunks = request.body.chunks;
         const lang = request.body.lang;

         console.log('Input text: ' + inputText);
-    try {
         const result = await fetch(`https://translate.yandex.net/api/v1/tr.json/translate?ucid=${ucid}&srv=android&format=text`, {
             method: 'POST',
             body: params,

src/prompt-converters.js+76 -0 Viewed

         'gemini-1.5-flash',
         'gemini-1.5-flash-latest',
         'gemini-1.5-flash-001',
+        'gemini-1.5-flash-exp-0827',
+        'gemini-1.5-flash-8b-exp-0827',
         'gemini-1.5-pro',
         'gemini-1.5-pro-latest',
         'gemini-1.5-pro-001',
         'gemini-1.5-pro-exp-0801',
+        'gemini-1.5-pro-exp-0827',
         'gemini-1.0-pro-vision-latest',
         'gemini-pro-vision',
     ];

+}
 /**
+ * Convert AI21 prompt. Classic: system message squash, user/assistant message merge.
+ * @param {object[]} messages Array of messages
+ * @param {string} charName Character name
+ * @param {string} userName User name
+ */
+function convertAI21Messages(messages, charName = '', userName = '') {
+    if (!Array.isArray(messages)) {
+        return [];
+    }
+    // Collect all the system messages up until the first instance of a non-system message, and then remove them from the messages array.
+    let i = 0, systemPrompt = '';
+    for (i = 0; i < messages.length; i++) {
+        if (messages[i].role !== 'system') {
+            break;
+        }
+        // Append example names if not already done by the frontend (e.g. for group chats).
+        if (userName && messages[i].name === 'example_user') {
+            if (!messages[i].content.startsWith(`${userName}: `)) {
+                messages[i].content = `${userName}: ${messages[i].content}`;
+            }
+        }
+        if (charName && messages[i].name === 'example_assistant') {
+            if (!messages[i].content.startsWith(`${charName}: `)) {
+                messages[i].content = `${charName}: ${messages[i].content}`;
+            }
+        }
+        systemPrompt += `${messages[i].content}\n\n`;
+    }
+    messages.splice(0, i);
+    // Prevent erroring out if the messages array is empty.
+    if (messages.length === 0) {
+        messages.unshift({
+            role: 'user',
+            content: '[Start a new chat]',
+        });
+    }
+    if (systemPrompt) {
+        messages.unshift({
+            role: 'system',
+            content: systemPrompt.trim(),
+        });
+    }
+    // Doesn't support completion names, so prepend if not already done by the frontend (e.g. for group chats).
+    messages.forEach(msg => {
+        if ('name' in msg) {
+            if (msg.role !== 'system' && !msg.content.startsWith(`${msg.name}: `)) {
+                msg.content = `${msg.name}: ${msg.content}`;
+            }
+            delete msg.name;
+        }
+    });
+    // Since the messaging endpoint only supports alternating turns, we have to merge messages with the same role if they follow each other
+    let mergedMessages = [];
+    messages.forEach((message) => {
+        if (mergedMessages.length > 0 && mergedMessages[mergedMessages.length - 1].role === message.role) {
+            mergedMessages[mergedMessages.length - 1].content += '\n\n' + message.content;
+        } else {
+            mergedMessages.push(message);
+        }
+    });
+    return mergedMessages;
+}
+/**
  * Convert a prompt from the ChatML objects to the format used by MistralAI.
  * @param {object[]} messages Array of messages
  * @param {string} charName Character name

     convertCohereMessages,
     convertMistralMessages,
     convertCohereTools,
+    convertAI21Messages,
 };

src/tokenizers/jamba.model+0 -0 Viewed

Binary file

src/util.js+5 -4 Viewed

@@ -34,8 +34,9 @@ function getConfig() {
34	CACHED_CONFIG = config;	34	CACHED_CONFIG = config;
35	return config;	35	return config;
36	} catch (error) {	36	} catch (error) {
37	console.warn('Failed to read config.yaml');	37	console.error(color.red('FATAL: Failed to read config.yaml. Please check the file for syntax errors.'));
38	return {};	38	console.error(error.message);
		39	process.exit(1);
39	}	40	}
40	}	41	}
41		42

  * @returns {string} A UUIDv4 string
  */
 function uuidv4() {
-    if ('randomUUID' in crypto) {
+    if ('crypto' in global && 'randomUUID' in global.crypto) {
-        return crypto.randomUUID();
+        return global.crypto.randomUUID();
+    }
     return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, function (c) {
         const r = Math.random() * 16 | 0;