Merge remote-tracking branch 'upstream/main'

src/lang/en.ts

@@ -171,6 +171,8 @@ export const languageEnglish = {
         translatorPrompt: "The prompt that is used for translation. if it is blank, it will use the default prompt. you can also use ChatML formating with {{slot}} for the dest language, {{solt::content}} for the content, and {{slot::tnote}} for the translator note.",
         translateBeforeHTMLFormatting: "If enabled, it will translate the text before Regex scripts and HTML formatting. this could make the token lesser but could break the formatting.",
         autoTranslateCachedOnly: "If enabled, it will automatically translate only the text that the user has translated previously.",
+        presetChain: "If it is not blank, the preset will be changed and applied randomly every time when user sends a message in the preset list in this input. preset list should be seperated by comma, for example, `preset1,preset2`.",
+        legacyMediaFindings: "If enabled, it will use the old method to find media assets, without using the additional search algorithm.",
     },
     setup: {
         chooseProvider: "Choose AI Provider",
@@ -817,4 +819,7 @@ export const languageEnglish = {
     permissionDenied: "Permission Denied by Your Browser or OS",
     customFlags: "Custom Flags",
     enableCustomFlags: "Enable Custom Flags",
+    googleCloudTokenization: "Google Cloud Tokenization",
+    presetChain: "Preset Chain",
+    legacyMediaFindings: "Legacy Media Findings",
 }

src/lang/zh-Hant.ts

@@ -20,7 +20,7 @@ export const languageChineseTraditional = {
         "noData": "無法找到檔案中的資料，或者檔案已經損毀",
         "onlyOneChat": "至少需要一個聊天室",
         "alreadyCharInGroup": "該群組中已經有一個同名角色。",
-        "noUserIcon": "請先設定你的個人頭像。",
+        "noUserIcon": "請先設定您的個人頭像。",
         "emptyText": "文字內容為空。",
         "wrongPassword": "密碼錯誤",
         "networkFetch": "這通常是由於網路連線不穩定或伺服器故障引起的。",
@@ -32,18 +32,18 @@ export const languageChineseTraditional = {
     "showHelp": "顯示幫助",
     "help": {
         "model": "此模型是指聊天中使用的主要模型。",
-        "submodel": "輔助模型是一個用於分析情感圖像、產生自動建議等的模型，推薦使用 GPT-3.5。",
+        "submodel": "輔助模型是一個用於分析情感圖片、產生自動建議等的模型，推薦使用 GPT-3.5。",
         "oaiapikey": "OpenAI 的 API 金鑰（Key），可在 https://platform.openai.com/account/api-keys 取得。",
         "mainprompt": "主要提示詞設定用於決定模型的預設行為。",
         "jailbreak": "當角色中的越獄開關被啟動後，越獄提示詞將被使用。",
         "globalNote": "一個對模型行為有強烈影響的備註（也稱為 UJB），適用於所有角色。",
         "autoSuggest": "用於自動建議使用者回應時生成選項的提示詞。",
         "formatOrder": "提示詞的排列順序：越靠下的區塊對模型的影響越大。",
-        "forceUrl": "此欄位不為空時，請求將被發送到你所輸入的網址。",
+        "forceUrl": "此欄位不為空時，請求將被發送到您所輸入的網址。",
         "tempature": "較低的數值會使角色更緊密地遵循提示詞，但會使回應更制式與機械化。\n較高的數值則會增強角色的創意表現，但回應可能會變得不穩定。",
         "frequencyPenalty": "較高的數值可以避免角色在個別回應中重複使用相同的詞彙，但回應也更容易出現語意混亂。",
         "presensePenalty": "較高的數值可以避免角色在整體對話中重複使用相同的詞彙，但這也可能導致回答失去一致性和穩定性。",
-        "sdProvider": "圖像生成提供者。",
+        "sdProvider": "圖片生成提供者。",
         "msgSound": "當角色回應時，播放 *叮* 的提示音",
         "charDesc": "角色的簡要描述。這會影響角色的回應方式。",
         "charFirstMessage": "角色的初始訊息，這會極大地影響角色的回應方式。",
@@ -54,11 +54,11 @@ export const languageChineseTraditional = {
         "loreActivationKey": "當上下文中包含任一關鍵字時，該條目將被啟用，並啟動相應的提示詞。使用逗號分隔。",
         "loreorder": "插入順序越高，對模型的影響力越大。在啟動大量條目時，也更不容易被截斷。",
         "bias": "Bias 是一組鍵值數據，用於修改某些字串出現的機率。\n其數值範圍可以是 -100 到 100。較高的數值會使該字串更可能出現，較低的數值則降低出現機率。\n另外，在某些模型中，若將數值設為 -101，該字串將被標記為「強制禁止詞」。\n警告：若 Tokenizer 設定有誤，可能無法正常運作。",
-        "emotion": "「表情立繪會根據角色情緒顯示對應的圖片。具體的情緒由角色的回應進行分析。你必須輸入情緒名稱作為詞彙*（如：joy, happy, fear 等）*。若存在名為 **neutral** 的情緒，將作為默認情緒。至少需要三張圖片才能正常運作。」",
+        "emotion": "「表情立繪會根據角色情緒顯示對應的圖片。具體的情緒由角色的回應進行分析。您必須輸入情緒名稱作為詞彙*（如：joy, happy, fear 等）*。若存在名為 **neutral** 的情緒，將作為默認情緒。至少需要三張圖片才能正常運作。」",
         "imggen": "分析聊天內容後，將提示套用至 {{slot}}。",
         "regexScript": "正規表達式（Regex Script）是一個自定義工具，用於將符合條件的字串由「IN」替換為「OUT」。\n\n有四種類型選項：\n\n- **修改輸入（Modify Input）**：修改使用者的輸入內容\n\n- **修改輸出（Modify Output）**：修改角色的輸出內容\n\n- **修改請求資料（Modify Request Data）**：在當前聊天資料發送時進行修改\n\n- **修改顯示（Modify Display）**：僅修改顯示的文字，不更改聊天資料\n\n「IN」必須是一個不帶標誌（flags）的正規表達式，且開頭和結尾不包含斜線。\n\n「OUT」是一個可以包含替換模式的字串。這些替換模式如下：\n\n- $$\n\n  - 插入符號「$」\n\n- $&\n\n  - 插入匹配到的子字串\n\n- $`\n\n  - 插入匹配子字串前的部分\n\n- $1\n\n  - 插入第一個匹配群組，可以替換為其他數字（例如 2、3...）\n\n- $(name)\n\n  - 插入命名群組\n\n針對標誌（flags），除了原生支援的標誌之外，還可以使用下列專為進階使用者設計的標誌：\n\n- `<inject>`：將結果注入當前字串中\n- `<move_top>`：將結果移到字串的頂部\n- `<move_bottom>`：將結果移到字串的底部\n- `<repeat_back>`：如果找不到匹配，則使用上一個匹配的結果\n- `<order n>`：設定結果的順序，數值越高顯示越靠前。「n」代表一個數字（例如 `<order 1>`）。如果未設定，默認為 0。\n- `<cbs>`：解析「IN」中的大括號語法\n\n若要與原生標誌結合使用，可以像這樣使用：`gi<cbs><move_top>`。",
         "experimental": "此為實驗性功能，可能不穩定。",
-        "oogaboogaURL": "如果你的 WebUI 支援舊版 API，你的 URL 應類似於 *https://.../run/textgen*。\n\n如果你的 WebUI 支援新版 API，你的 URL 應類似於 *https://.../api/v1/generate*，且將使用 API 伺服器作為主機，並在參數中添加 —api。",
+        "oogaboogaURL": "如果您的 WebUI 支援舊版 API，您的 URL 應類似於 *https://.../run/textgen*。\n\n如果您的 WebUI 支援新版 API，您的 URL 應類似於 *https://.../api/v1/generate*，且將使用 API 伺服器作為主機，並在參數中添加 —api。",
         "exampleMessage": "示範對話會影響角色的回應，但不會永久佔用 Token。\n\n對話格式示例：\n\n```\n<START>\n{{user}}: hi\n{{char}}: hello\n<START>\n{{user}}: hi\nHaruhi: hello\n```\n\n```<START>``` 標記了一段新對話的開始。",
         "creatorQuotes": "說明將顯示在初始訊息之上，用於向使用者提供角色說明。此內容不會進入提示詞中。",
         "systemPrompt": "此欄位不為空時，將替換設定中的主要提示詞為此內容。",
@@ -69,10 +69,10 @@ export const languageChineseTraditional = {
         "loreSelective": "啟用選擇性模式後，需同時匹配關鍵字與次要關鍵字，方可啟用該條目。",
         "loreRandomActivation": "啟用「使用機率條件」後，若同時符合啟用條目的其他條件，則在每次發送聊天時，該條目將依照設定的機率被使用。",
         "additionalAssets": "在聊天中顯示的額外資源。\n\n - 使用 `{{raw::<資源名稱>}}` 作為路徑。\n - 使用 `{{image::<資源名稱>}}` 作為圖片。\n - 使用 `{{video::<資源名稱>}}` 作為影片。\n - 使用 `{{audio::<資源名稱>}}` 作為音訊。\n    - 建議放置在背景 HTML 中。",
-        "superMemory": "SuperMemory 通過向 AI 提供摘要資料來增強角色的記憶能力。\n\nSuperMemory 是一個文本摘要功能，推薦使用 davinci 模型。不建議使用輔助模型，除非它是未經過濾、最大上下文長度超過 2000 Tokens，且具有良好摘要能力的模型。\n\nSuperMemory 提示詞決定了模型如何撰寫摘要。留空將使用預設提示詞，建議保持留空。\n\n完成所有設定後，你可以在角色的設定中啟用此功能。",
+        "superMemory": "SuperMemory 通過向 AI 提供摘要資料來增強角色的記憶能力。\n\nSuperMemory 是一個文本摘要功能，推薦使用 davinci 模型。不建議使用輔助模型，除非它是未經過濾、最大上下文長度超過 2000 Tokens，且具有良好摘要能力的模型。\n\nSuperMemory 提示詞決定了模型如何撰寫摘要。留空將使用預設提示詞，建議保持留空。\n\n完成所有設定後，您可以在角色的設定中啟用此功能。",
         "replaceGlobalNote": "此欄位不為空時，將替換當前的全域備註為此內容。",
-        "backgroundHTML": "將 Markdown/HTML 注入到聊天畫面的背景中。\n\n你也可以使用額外資源。例如，你可以使用 {{audio::<資源名稱}} 作為背景音樂。\n\n此外，你還可以與額外資源搭配使用以下格式：\n - {{bg::<資源名稱>}}：將資源設為背景。",
-        "additionalText": "只有當 AI 認為有必要時，才會將該段文本添加到角色描述中。你可以在此處放置較長的文本。使用雙換行進行內容分隔。",
+        "backgroundHTML": "將 Markdown/HTML 注入到聊天畫面的背景中。\n\n您也可以使用額外資源。例如，您可以使用 {{audio::<資源名稱}} 作為背景音樂。\n\n此外，您還可以與額外資源搭配使用以下格式：\n - {{bg::<資源名稱>}}：將資源設為背景。",
+        "additionalText": "只有當 AI 認為有必要時，才會將該段文本添加到角色描述中。您可以在此處放置較長的文本。使用雙換行進行內容分隔。",
         "charjs": "這是一段會與角色一同運行的 JavaScript。詳情請查看：https://github.com/kwaroran/RisuAI/blob/main/src/etc/example-char.js\n**出於安全原因，目前不建議使用。這些代碼不會被包含在匯出中。**",
         "romanizer": "Romanizer 是一個將非羅馬字母轉換為羅馬字母的外掛程式，用於減少請求資料時的 Token。這可能會導致輸出結果與原始模型不同。如果已在聊天中使用羅馬字母，不建議啟用。",
         "oaiRandomUser": "啟用後，請求中的使用者參數將被隨機 UUID 替代，並在重新整理時修改。這可以用來防止 AI 識別使用者。",
@@ -89,7 +89,7 @@ export const languageChineseTraditional = {
         "forcePlainFetch": "啟用後，將使用瀏覽器的 Fetch API 來替代原生 HTTP 請求。這可能會導致 CORS 錯誤。",
         "autoFillRequestURL": "啟用後，將自動填入請求 URL 以匹配當前模型。",
         "chainOfThought": "啟用後，將在提示詞中添加思維鏈（CoT, Chain-of-Thought）提示。",
-        "gptVisionQuality": "此選項用於設定圖像檢測模型的品質。品質越高，檢測越準確，但會使用更多的 Token。",
+        "gptVisionQuality": "此選項用於設定圖片檢測模型的品質。品質越高，檢測越準確，但會使用更多的 Token。",
         "genTimes": "此設定支援模型上的重滾（reroll）回應數量。除第一則回應外，其他回應將作為快取使用，以降低成本。但若未多次重滾回應，可能增加成本。",
         "requestretrys": "此選項用於設定請求失敗時的重試次數。",
         "emotionPrompt": "此選項用於設定情緒檢測的提示詞。留空將使用預設提示詞。",
@@ -104,10 +104,10 @@ export const languageChineseTraditional = {
         "nickname": "設定後，將在聊天中以此暱稱取代角色名稱，並顯示於 {{char}} 和 <char>。",
         "useRegexLorebook": "啟用後，Lorebook 將改用正規表達式（Regex）搜尋，而不再使用字串匹配。格式為 /regex/flags。",
         "customChainOfThought": "警告：不再建議使用思維鏈（CoT, Chain-of-Thought）切換功能。請將相關提示詞移至其他提示詞欄位。",
-        "customPromptTemplateToggle": "可在此處設定自定義提示詞切換功能。使用 `<toggle variable>=<toggle name>` 格式，每行一個，例如：`cot=Toggle COT`。你可以在提示詞中透過 `{{getglobalvar::toggle_<toggle variable>}}` 語法來使用這些切換功能，如：`{{getglobalvar::toggle_cot}}`。",
+        "customPromptTemplateToggle": "可在此處設定自定義提示詞切換功能。使用 `<toggle variable>=<toggle name>` 格式，每行一個，例如：`cot=Toggle COT`。您可以在提示詞中透過 `{{getglobalvar::toggle_<toggle variable>}}` 語法來使用這些切換功能，如：`{{getglobalvar::toggle_cot}}`。",
         "defaultVariables": "可在此處設定自訂預設變數。使用 `<variable name>=<variable value>` 格式，每行一個。例如：`name=叡甦`，可在觸發式和 CBS 變數中使用，如：`{{getvar::A}}`、`{{setvar::A::B}}` 或 `{{? $A + 1}}`。若提示詞範本的預設變數與角色的預設變數名稱相同，系統將使用角色的預設變數。",
         "lowLevelAccess": "啟用後，將開放需要高計算能力的功能，並允許通過角色中的觸發式執行 AI 模型。除非確實需要這些功能，否則不要啟用此選項。",
-        "triggerLLMPrompt": "這是將發送到模型的提示詞。你可以使用 `@@role user`、`@@role system`、`@@role assistant` 來設定多輪對話及角色。例如：\n```\n@@role system\nrespond as hello\n@@role assistant\nhello\n@@role user\nhi\n```",
+        "triggerLLMPrompt": "這是將發送到模型的提示詞。您可以使用 `@@role user`、`@@role system`、`@@role assistant` 來設定多輪對話及角色。例如：\n```\n@@role system\nrespond as hello\n@@role assistant\nhello\n@@role user\nhi\n```",
         "legacyTranslation": "啟用後，將使用舊版翻譯方法，在翻譯前對 Markdown 和引號進行預處理，而非在翻譯後處理。",
         "luaHelp": "可使用 Lua 作為觸發式，並可定義 onInput、onOutput 和 onStart 函數。當使用者發送消息時，調用 onInput；當角色發送消息時，調用 onOutput；當對話開始時，調用 onStart。詳情請參閱說明文檔。",
         "claudeCachingExperimental": "Claude 快取是實驗性功能，可減少模型成本。但若在不使用重滾（reroll）回應的情況下啟用，則可能增加成本。實驗性功能可能不穩定，且未來可能會有所變動。",
@@ -117,7 +117,7 @@ export const languageChineseTraditional = {
         "customCSS": "自訂 CSS 樣式。若出現問題，可使用 (Ctrl + .) 啟用或禁用。",
         "betaMobileGUI": "啟用後，將在小於 800px 的螢幕上使用測試版行動介面，需重新整理頁面。",
         "unrecommended": "這是一個不建議使用的設定。建議關閉。",
-        "jsonSchema": "JSON Schema 將在 AI 模型支援時發送給模型。\n\n然而，由於 JSON Schema 學習難度較高，在叡甦中，你可以使用 TypeScript 接口的子集來代替 JSON Schema。叡甦將在運行時進行轉換。例如，如果你想發送如下的JSON：\n\n```js\n{\n  \"name\": \"叡甦\", // name 必須是叡甦，\n  \"age\": 1, // age 必須是數字，\n  \"icon\": \"slim\", // icon 必須是 ’slim‘ 或 ’rounded‘\n  \"thoughts\": [\"Good View!\", \"Lorem\"] // thoughts 必須是字符串數組\n}\n```\n\n你可以使用以下 TypeScript 接口：\n\n```typescript\ninterface Schema {\n  name: string;\n  age: number;\n  icon: ’slim‘|’rounded‘\n  thoughts: string[]\n}\n```\n\n接口名稱不重要。欲了解更多資訊，請參閱 TypeScript 說明文件：https://www.typescriptlang.org/docs/handbook/interfaces.html 。要檢查支持的 TypeScript 子集，請查看以下內容。<details><summary>支持的 TypeScript 子集</summary>\n\n支援的類型包括 `boolean`、`number`、`string` 和 `Array`。高級類型不被支援（如：單元類型、交集類型、聯合類型、可選類型、字面量類型等），除了以下幾種情況：\n\n        - 原始資料型別（Primitive Type）的陣列（Array）：（如 `string[]`、`Array<boolean>`）\n        - 字符串之間的單值類型（Unit Types）：（例如 `’slim‘|’rounded‘`）\n\n        屬性必須在同一行內定義。若一行中有多個屬性，將會產生錯誤。屬性和接口名稱僅可使用拉丁字符，並在 ASCII 範圍內。屬性名稱不得以單引號或雙引號包裹。接口內部不支持嵌套。在定義屬性的行中，不能包含 `{` 或 `}`。如果想使用更高級的類型，請使用 JSON Schema。\n        </details>",
+        "jsonSchema": "JSON Schema 將在 AI 模型支援時發送給模型。\n\n然而，由於 JSON Schema 學習難度較高，在叡甦中，您可以使用 TypeScript 接口的子集來代替 JSON Schema。叡甦將在運行時進行轉換。例如，如果您想發送如下的JSON：\n\n```js\n{\n  \"name\": \"叡甦\", // name 必須是叡甦，\n  \"age\": 1, // age 必須是數字，\n  \"icon\": \"slim\", // icon 必須是 ’slim‘ 或 ’rounded‘\n  \"thoughts\": [\"Good View!\", \"Lorem\"] // thoughts 必須是字符串數組\n}\n```\n\n您可以使用以下 TypeScript 接口：\n\n```typescript\ninterface Schema {\n  name: string;\n  age: number;\n  icon: ’slim‘|’rounded‘\n  thoughts: string[]\n}\n```\n\n接口名稱不重要。欲了解更多資訊，請參閱 TypeScript 說明文件：https://www.typescriptlang.org/docs/handbook/interfaces.html 。要檢查支持的 TypeScript 子集，請查看以下內容。<details><summary>支持的 TypeScript 子集</summary>\n\n支援的類型包括 `boolean`、`number`、`string` 和 `Array`。高級類型不被支援（如：單元類型、交集類型、聯合類型、可選類型、字面量類型等），除了以下幾種情況：\n\n        - 原始資料型別（Primitive Type）的陣列（Array）：（如 `string[]`、`Array<boolean>`）\n        - 字符串之間的單值類型（Unit Types）：（例如 `’slim‘|’rounded‘`）\n\n        屬性必須在同一行內定義。若一行中有多個屬性，將會產生錯誤。屬性和接口名稱僅可使用拉丁字符，並在 ASCII 範圍內。屬性名稱不得以單引號或雙引號包裹。接口內部不支持嵌套。在定義屬性的行中，不能包含 `{` 或 `}`。如果想使用更高級的類型，請使用 JSON Schema。\n        </details>",
         "strictJsonSchema": "啟用後，某些模型將嚴格遵循提供的 JSON Schema。若禁用，可能會忽略 JSON Schema。",
         "extractJson": "此欄位不為空時，將從回應中提取特定的 JSON 資料。例如：想從回應 `{\"response\": {\"text\": [\"hello\"]}}` 中提取 `response.text[0]`，可以填寫 `response.text.0`。",
         "translatorNote": "可在此處為每個角色添加獨特的翻譯提示，但僅適用於使用 Ax. 模型進行翻譯。要啟用此功能，請在語言設定中包含 `{{slot::tnote}}`。此功能不適用群組聊天。",
@@ -126,9 +126,11 @@ export const languageChineseTraditional = {
         "chatHTML": "每個聊天插入的 HTML。\n\n可以使用CBS和特殊標籤。\n- `<risutextbox>`：用於呈現文字的文本框\n- `<risuicon>`：用於顯示使用者或助理的頭像\n- `<risubuttons>`：用於聊天編輯、翻譯等圖示按鈕\n- `<risugeninfo>`：生成訊息按鈕。",
         "systemContentReplacement": "若模型不支援系統提示詞，則使用此格式取代系統提示詞內容。",
         "systemRoleReplacement": "若模型不支援系統角色，將使用此角色取代系統角色。",
-        "summarizationPrompt": "用於摘要的提示詞。留空將使用預設提示詞。你也可以使用包含 {{slot}} 的 ChatML 格式來處理聊天數據。",
-        "translatorPrompt": "用於翻譯的提示詞。留空將使用默認提示。你還可以使用帶有 {{slot}} 的 ChatML 格式表示目標語言：用 {{slot::content}} 表示內容，用 {{slot::tnote}} 表示翻譯註釋。",
-        "translateBeforeHTMLFormatting": "啟用後，將在正規表達式和 HTML 格式化之前翻譯文本。這可能減少 Token 數，但可能破壞格式。"
+        "summarizationPrompt": "用於摘要的提示詞。留空將使用預設提示詞。您也可以使用包含 {{slot}} 的 ChatML 格式來處理聊天數據。",
+        "translatorPrompt": "用於翻譯的提示詞。留空將使用默認提示。您還可以使用帶有 {{slot}} 的 ChatML 格式表示目標語言：用 {{slot::content}} 表示內容，用 {{slot::tnote}} 表示翻譯註釋。",
+        "translateBeforeHTMLFormatting": "啟用後，將在正規表達式和 HTML 格式化之前翻譯文本。這可能減少 Token 數，但可能破壞格式。",
+        "autoTranslateCachedOnly": "啟用後，僅會自動翻譯使用者之前已翻譯的內容。",
+        "APIPool": "啟用後，系統將連接到 RisuAI 的 API 資源池。已啟用的使用者可共享免費、速率受限模型的 API 金鑰，從而利用其他使用者未充分使用的金鑰，增加對速率受限模型的請求次數。"
     },
     "setup": {
         "chooseProvider": "選擇 AI 提供者",
@@ -143,17 +145,17 @@ export const languageChineseTraditional = {
         "themeDescWifuCut": "適合在行動裝置上使用",
         "themeDescClassic": "適用於所有裝置",
         "texttheme": "設定文字顏色",
-        "inputName": "最後，請輸入你的暱稱。",
-        "welcome": "歡迎使用 Risu（叡甦）！我將引導你進行設定。請問我該如何稱呼你？",
-        "welcome2": "你好，{username}！在開始之前，我會問你一些問題，稍後可在設定中進行修改。\n\n首先，請選擇 AI 提供者。",
+        "inputName": "最後，請輸入您的暱稱。",
+        "welcome": "歡迎使用 Risu（叡甦）！我將引導您進行設定。請問我該如何稱呼您？",
+        "welcome2": "您好，{username}！在開始之前，我會問您一些問題，稍後可在設定中進行修改。\n\n首先，請選擇 AI 提供者。",
         "openrouterProvider": "Openrouter 提供許多模型，部分免費且未經內容過濾，但品質不如 OpenAI。",
         "hordeProvider": "Horde 提供免費服務，但回應時間較長且品質較低。",
-        "setProviderLater": "還有其他提供者，你可以稍後在設定中配置。如想稍後設定，請選擇此選項。",
+        "setProviderLater": "還有其他提供者，您可以稍後在設定中配置。如想稍後設定，請選擇此選項。",
         "setupOpenAI": "使用 OpenAI，需要取得 API 金鑰（Key）。\n1. 前往 https://beta.openai.com/ \n2. 使用帳號登入 \n3. 前往 https://beta.openai.com/account/api-keys \n4. 點擊「Create New API Key」，並命名金鑰。 \n5. 複製該金鑰。 \n6. 返回叡甦\n7. 貼上金鑰並點擊「發送」。",
         "setupClaude": "使用 Claude，需要取得一個 API 金鑰（Key）。",
         "setupClaudeSteps": [
             "點擊此連結並使用 Google 帳號登錄",
-            "輸入你的資料並點擊「繼續」（Continue）",
+            "輸入您的資料並點擊「繼續」（Continue）",
             "僅在第一個框中輸入任意名稱，然後點擊「建立帳號」（Create Account）",
             "點擊「購買點數」（Buy Credits）",
             "點擊「完成設定」（Complete Setup）",
@@ -167,20 +169,20 @@ export const languageChineseTraditional = {
         ],
         "setupOpenrouter": "使用 Openrouter 需要獲取 API 金鑰（Key）。 \n1. 前往 https://openrouter.ai/keys\n2. 點擊「Create Key」\n3. 任意命名金鑰名稱。\n4. 複製該金鑰。\n5. 返回叡甦\n6. 貼上金鑰並點擊「發送」。",
         "allDone": "完成所有設定！請稍待片刻。",
-        "setupLaterMessage": "歡迎，{username}！你希望我引導你完成設定還是自行設定？",
+        "setupLaterMessage": "歡迎，{username}！您希望我引導您完成設定還是自行設定？",
         "setupMessageOption1": "引導我完成設定",
         "setupMessageOption1Desc": "推薦新使用者使用",
         "setupMessageOption2": "由我自己完成設定",
-        "claudeDesc": "Claude 是由 Antropic 開發的 AI 模型，是 OpenAI 的競爭對手。若你希望使用非英語語言，它優於 GPT。",
+        "claudeDesc": "Claude 是由 Antropic 開發的 AI 模型，是 OpenAI 的競爭對手。若您希望使用非英語語言，它優於 GPT。",
         "openAIDesc": "OpenAI GPT 是高品質的 AI 模型，但它付費且存在內容審核，在非英語環境下表現較弱。",
         "chooseChatType": "很好！現在請選擇聊天語言。",
         "chooseChatTypeOption1": "全英語",
         "chooseChatTypeOption1Desc": "推薦英語使用者。AI 將使用英語進行輸入和輸出。",
         "chooseChatTypeOption2": "英語處理",
-        "chooseChatTypeOption2Desc": "推薦非英語使用者。AI 內部使用英語處理，但輸入輸出為你的語言。",
+        "chooseChatTypeOption2Desc": "推薦非英語使用者。AI 內部使用英語處理，但輸入輸出為您的語言。",
         "chooseChatTypeOption3": "無語言側重",
-        "chooseChatTypeOption3Desc": "AI 將使用你的語言處理，但可能會降低回應品質。",
-        "chooseCheapOrMemory": "除此之外，你更傾向於記憶功能還是節省成本？",
+        "chooseChatTypeOption3Desc": "AI 將使用您的語言處理，但可能會降低回應品質。",
+        "chooseCheapOrMemory": "除此之外，您更傾向於記憶功能還是節省成本？",
         "chooseCheapOrMemoryOption1": "記憶功能",
         "chooseCheapOrMemoryOption1Desc": "AI 會記住更多內容，但費用較高。",
         "chooseCheapOrMemoryOption2": "省錢模式",
@@ -189,7 +191,7 @@ export const languageChineseTraditional = {
         "chooseCheapOrMemoryOption3Desc": "AI 記住的內容多於省錢模式，但不及使用記憶功能。",
         "chooseCheapOrMemoryOption4": "無限制",
         "chooseCheapOrMemoryOption4Desc": "AI 會記住幾乎所有內容，但費用極高。",
-        "finally": "最後，你是否希望使用進階工具？",
+        "finally": "最後，您是否希望使用進階工具？",
         "finallyOption1": "是",
         "finallyOption1Desc": "使用進階工具會使界面變得更複雜。推薦進階使用者使用。",
         "finallyOption2": "否",
@@ -241,9 +243,9 @@ export const languageChineseTraditional = {
     "group": "群組",
     "groupLoreInfo": "群組 Lorebook 適用於該群組的所有對話。",
     "localLoreInfo": "聊天 Lorebook 僅用於此對話。",
-    "removeConfirm": "你確定要刪除：",
-    "removeConfirm2": "你**真的**確定要刪除：",
-    "exportConfirm": "你想要匯出此資料嗎？",
+    "removeConfirm": "您確定要刪除：",
+    "removeConfirm2": "您**真的**確定要刪除：",
+    "exportConfirm": "您想要匯出此資料嗎？",
     "insertOrder": "插入順序",
     "activationKeys": "關鍵字",
     "activationKeysInfo": "使用逗號分隔",
@@ -254,9 +256,9 @@ export const languageChineseTraditional = {
     "removeGroup": "刪除群組",
     "exportCharacter": "匯出角色",
     "userSetting": "使用者設定",
-    "username": "你的名稱",
-    "userIcon": "你的頭像",
-    "successExport": "已成功匯出並保存至你的下載資料夾",
+    "username": "您的名稱",
+    "userIcon": "您的頭像",
+    "successExport": "已成功匯出並保存至您的下載資料夾",
     "successImport": "成功匯入",
     "importedCharacter": "匯入角色",
     "alwaysActive": "始終啟用",
@@ -304,21 +306,21 @@ export const languageChineseTraditional = {
     "savebackup": "備份至 Google",
     "loadbackup": "從 Google 讀取備份",
     "files": "檔案",
-    "backupConfirm": "你確定要保存備份嗎？",
-    "backupLoadConfirm": "你確定要讀取備份嗎？所有資料將被覆蓋！",
-    "backupLoadConfirm2": "你**真的、真的**確定要加載備份嗎？這將會清除所有資料！",
+    "backupConfirm": "您確定要保存備份嗎？",
+    "backupLoadConfirm": "您確定要讀取備份嗎？所有資料將被覆蓋！",
+    "backupLoadConfirm2": "您**真的、真的**確定要加載備份嗎？這將會清除所有資料！",
     "pasteAuthCode": "請從彈出窗口複製驗證碼並貼入：",
     "others": "其他",
     "presets": "預設設定",
-    "imageGeneration": "圖像生成",
+    "imageGeneration": "圖片生成",
     "provider": "提供者",
     "key": "金鑰（Key）",
     "noData": "沒有資料",
-    "currentImageGeneration": "當前圖像生成數據",
+    "currentImageGeneration": "當前圖片生成數據",
     "promptPreprocess": "使用提示詞預處理",
     "SwipeRegenerate": "使用滑動箭頭重新產生訊息",
     "instantRemove": "刪除訊息時連帶刪除後續訊息",
-    "instantRemoveConfirm": "你想只刪除一條訊息嗎？若選擇「否」，後續訊息也將被刪除。",
+    "instantRemoveConfirm": "您想只刪除一條訊息嗎？若選擇「否」，後續訊息也將被刪除。",
     "textColor": "文字顏色",
     "classicRisu": "經典 Risu",
     "highcontrast": "高對比度",
@@ -506,7 +508,7 @@ export const languageChineseTraditional = {
     "innerFormat": "內部格式",
     "HypaMemory": "HypaMemory",
     "ToggleHypaMemory": "啟動 HypaMemory",
-    "resetPromptTemplateConfirm": "你真的確定要重置提示詞模板嗎？",
+    "resetPromptTemplateConfirm": "您真的確定要重置提示詞模板嗎？",
     "emotionMethod": "情緒檢測方式",
     "continueResponse": "繼續回應",
     "showMenuChatList": "在選單中顯示聊天列表",
@@ -520,7 +522,7 @@ export const languageChineseTraditional = {
     "importPersona": "匯入使用者設定",
     "export": "匯出",
     "import": "匯入",
-    "supporterThanks": "支持者感謝",
+    "supporterThanks": "支持者鳴謝",
     "supporterThanksDesc": "感謝您的支持！",
     "donatorPatreonDesc": "為保護隱私，默認不會顯示在名單中。若想顯示您的暱稱，請前往叡甦的 Patreon 頁面並點擊連結按鈕。",
     "useNamePrefix": "使用名稱前綴",
@@ -539,8 +541,8 @@ export const languageChineseTraditional = {
     "exactTokens": "精確 Tokens",
     "fixedTokens": "估算 Tokens",
     "inlayViewScreen": "內嵌視窗",
-    "imgGenPrompt": "圖像生成提示詞",
-    "imgGenNegatives": "圖像生成負面提示詞",
+    "imgGenPrompt": "圖片生成提示詞",
+    "imgGenNegatives": "圖片生成負面提示詞",
     "imgGenInstructions": "系統提示詞",
     "usePlainFetchWarn": "使用 NovelAI 時請關閉此選項，避免出現 CORS 錯誤。",
     "translationPrompt": "翻譯提示詞",
@@ -578,12 +580,12 @@ export const languageChineseTraditional = {
     "inputCardPassword": "輸入角色卡密碼",
     "ccv2Desc": "V2 角色卡是廣泛用於聊天機器人前端的格式。",
     "ccv3Desc": "V3 角色卡是用於聊天機器人前端的新型格式。",
-    "realmDesc": "RisuRealm 是叡甦的內容分享平台，你可以將角色分享給其他使用者。",
+    "realmDesc": "RisuRealm 是叡甦的內容分享平台，您可以將角色分享給其他使用者。",
     "rccDesc": "Risu Refined 角色卡具有密碼保護、完整性驗證等附加功能。",
     "password": "密碼",
     "license": "授權",
-    "licenseDesc": "你可以設定下載授權，限制角色卡對提示詞的使用。",
-    "passwordDesc": "你可以為角色卡設置密碼，防止未經授權的訪問。",
+    "licenseDesc": "您可以設定下載授權，限制角色卡對提示詞的使用。",
+    "passwordDesc": "您可以為角色卡設置密碼，防止未經授權的訪問。",
     "largePersonaPortrait": "使用者肖像",
     "module": "模組",
     "modules": "模組",
@@ -601,9 +603,9 @@ export const languageChineseTraditional = {
     "sideMenuRerollButton": "側欄選單重新載入",
     "persistentStorage": "永久儲存",
     "persistentStorageSuccess": "儲存已成功永久化",
-    "persistentStorageFail": "儲存未能永久化。你是否拒絕了請求，或瀏覽器不支持？",
+    "persistentStorageFail": "儲存未能永久化。您是否拒絕了請求，或瀏覽器不支持？",
     "persistentStorageRecommended": "建議使用永久儲存",
-    "persistentStorageDesc": "你的瀏覽器支持永久儲存，建議啟用以提升效能和使用者體驗。",
+    "persistentStorageDesc": "您的瀏覽器支持永久儲存，建議啟用以提升效能和使用者體驗。",
     "enable": "啟用",
     "postFile": "上傳檔案",
     "requestInfoInsideChat": "在聊天中顯示請求資料",
@@ -617,7 +619,7 @@ export const languageChineseTraditional = {
     "useAdvancedEditor": "使用進階編輯器",
     "noWaitForTranslate": "不等待翻譯",
     "updateRealm": "更新至 RisuRealm",
-    "updateRealmDesc": "你正試圖將角色更新至 RisuRealm。此操作將使角色更新至 RisuRealm，且無法還原。",
+    "updateRealmDesc": "您正試圖將角色更新至 RisuRealm。此操作將使角色更新至 RisuRealm，且無法還原。",
     "antiClaudeOverload": "防止 Claude 超載",
     "activeTabChange": "目前的標籤已停用，因其他標籤處於活動中。若要啟動此標籤，請按「確定」。",
     "maxSupaChunkSize": "最大 SupaMemory Chunk 大小",
@@ -672,11 +674,11 @@ export const languageChineseTraditional = {
     "error": "錯誤",
     "input": "輸入",
     "select": "選擇",
-    "lowLevelAccessConfirm": "此內容使用低層級訪問，可直接存取 AI 模型和你的儲存資料。你確定要匯入嗎？",
+    "lowLevelAccessConfirm": "此內容使用低層級訪問，可直接存取 AI 模型和您的儲存資料。您確定要匯入嗎？",
     "triggerLowLevelOnly": "此觸發僅適用於低層級訪問，需在角色或模組的進階設定中啟用低層級訪問。",
     "truthy": "真值",
     "extractRegex": "使用正規表達式提取文字",
-    "runImgGen": "執行圖像生成",
+    "runImgGen": "執行圖片生成",
     "cutChat": "分割聊天",
     "modifyChat": "修改聊天",
     "regex": "正規表達式",
@@ -703,7 +705,7 @@ export const languageChineseTraditional = {
     "doNotTranslate": "不進行翻譯",
     "includePersonaName": "包含使用者名稱",
     "hidePersonaName": "隱藏使用者名稱",
-    "triggerSwitchWarn": "更改觸發類型後，現有觸發將被清除。你確定要繼續嗎？",
+    "triggerSwitchWarn": "更改觸發類型後，現有觸發將被清除。您確定要繼續嗎？",
     "codeMode": "程式碼",
     "blockMode": "區塊",
     "helpBlock": "幫助",
@@ -735,10 +737,10 @@ export const languageChineseTraditional = {
     "betaMobileGUI": "測試版行動介面",
     "menu": "選單",
     "connectionOpen": "已開啟連線",
-    "connectionOpenInfo": "多人聊天室已開啟，你可以將聊天室代碼分享給其他使用者。其他使用者可在 Playground > 加入多人聊天室 > 使用代碼加入。",
+    "connectionOpenInfo": "多人聊天室已開啟，您可以將聊天室代碼分享給其他使用者。其他使用者可在 Playground > 加入多人聊天室 > 使用代碼加入。",
     "createMultiuserRoom": "新建多人聊天室",
-    "connectionHost": "你是聊天室的主持人。",
-    "connectionGuest": "你是聊天室的訪客。",
+    "connectionHost": "您是聊天室的主持人。",
+    "connectionGuest": "您是聊天室的訪客。",
     "otherUserRequesting": "其他使用者正在請求中，請稍後重試。",
     "jsonSchema": "JSON Schema",
     "enableJsonSchema": "Enable Schema",
@@ -770,5 +772,12 @@ export const languageChineseTraditional = {
     "translatorPrompt": "翻譯提示詞",
     "translateBeforeHTMLFormatting": "於 HTML 格式化前翻譯",
     "retranslate": "重新翻譯",
-    "loading": "載入中"
+    "loading": "載入中",
+    "autoTranslateCachedOnly": "僅自動翻譯已快取的內容",
+    "notification": "使用系統通知",
+    "permissionDenied": "權限被您的瀏覽器或操作系統拒絕",
+    "customFlags": "自定義修飾詞（Flags）",
+    "enableCustomFlags": "啟用自定義修飾詞（Flags）",
+    "googleCloudTokenization": "Google Cloud Tokenization",
+    "APIPool": "API 工具"
 }

src/lib/ChatScreens/Chat.svelte

@@ -57,7 +57,7 @@
     }: Props = $props();
 
     let msgDisplay = $state('')
-    let translated = $state(DBState.db.autoTranslate)
+    let translated = $state(false)
     let role = $derived(DBState.db.characters[selIdState.selId].chats[DBState.db.characters[selIdState.selId].chatPage].message[idx]?.role)
     async function rm(e:MouseEvent, rec?:boolean){
         if(e.shiftKey){

src/lib/Playground/PlaygroundEmbedding.svelte

@@ -28,8 +28,15 @@
 
 <span class="text-textcolor text-lg">Model</span>
 <SelectInput bind:value={model}>
-    <OptionInput value="MiniLM">MiniLM L6 v2</OptionInput>
-    <OptionInput value="nomic">Nomic Embed Text v1.5</OptionInput>
+    <OptionInput value="MiniLM">MiniLM L6 v2 (CPU)</OptionInput>
+    <OptionInput value="nomic">Nomic Embed Text v1.5 (CPU)</OptionInput>
+    <OptionInput value="nomicGPU">Nomic Embed Text v1.5 (GPU)</OptionInput>
+    <OptionInput value="bgeSmallEn">BGE Small English (CPU)</OptionInput>
+    <OptionInput value="bgeSmallEnGPU">BGE Small English (GPU)</OptionInput>
+    <OptionInput value="bgem3">BGE Medium 3 (CPU)</OptionInput>
+    <OptionInput value="bgem3GPU">BGE Medium 3 (GPU)</OptionInput>
+    <OptionInput value="openai3small">OpenAI text-embedding-3-small</OptionInput>
+    <OptionInput value="openai3large">OpenAI text-embedding-3-large</OptionInput>
     <OptionInput value="custom">Custom (OpenAI-compatible)</OptionInput>
 </SelectInput>
 

src/lib/Playground/PlaygroundTokenizer.svelte

@@ -6,9 +6,12 @@
     let input = $state("");
     let output = $state("");
     let outputLength = $state(0);
+    let time = $state(0)
     const onInput = async () => {
         try {
+            const start = performance.now();
             const tokenized = await encode(input);
+            time = performance.now() - start;
             const tokenizedNumArray = Array.from(tokenized)
             outputLength = tokenizedNumArray.length;
             output = JSON.stringify(tokenizedNumArray);
@@ -29,3 +32,4 @@
 <TextAreaInput value={output} />
 
 <span class="text-textcolor2 text-lg">{outputLength} {language.tokens}</span>
+<span class="text-textcolor2 text-lg">{time} ms</span>

src/lib/Setting/Pages/AdvancedSettings.svelte

@@ -41,6 +41,10 @@
 <span class="text-textcolor">Kei Server URL</span>
 <TextInput marginBottom={true} size={"sm"} bind:value={DBState.db.keiServerURL} placeholder="Leave it blank to use default"/>
 
+<span class="text-textcolor">{language.presetChain} <Help key="presetChain"/></span>
+<TextInput marginBottom={true} size={"sm"} bind:value={DBState.db.presetChain} placeholder="Leave it blank to not use">
+</TextInput>
+
 <span class="text-textcolor">{language.requestretrys} <Help key="requestretrys"/></span>
 <NumberInput marginBottom={true} size={"sm"} min={0} max={20} bind:value={DBState.db.requestRetrys}/>
 
@@ -78,6 +82,9 @@
 <div class="flex items-center mt-4">
     <Check bind:check={DBState.db.forceProxyAsOpenAI} name={language.forceProxyAsOpenAI}> <Help key="forceProxyAsOpenAI"/></Check>
 </div>
+<div class="flex items-center mt-4">
+    <Check bind:check={DBState.db.legacyMediaFindings} name={language.legacyMediaFindings}> <Help key="legacyMediaFindings"/></Check>
+</div>
 <div class="flex items-center mt-4">
     <Check bind:check={DBState.db.autofillRequestUrl} name={language.autoFillRequestURL}> <Help key="autoFillRequestURL"/></Check>
 </div>
@@ -112,6 +119,11 @@
             <Help key="experimental"/><Help key="oaiRandomUser"/>
         </Check>
     </div>
+    <div class="flex items-center mt-4">
+        <Check bind:check={DBState.db.googleClaudeTokenizing} name={language.googleCloudTokenization}>
+            <Help key="experimental"/>
+        </Check>
+    </div>
 {/if}
 {#if DBState.db.showUnrecommended}
     <div class="flex items-center mt-4">

src/lib/Setting/Pages/OtherBotSettings.svelte

@@ -426,12 +426,6 @@
             {/if}
             <span class="text-textcolor">{language.summarizationPrompt} <Help key="summarizationPrompt" /></span>
             <TextAreaInput size="sm" bind:value={DBState.db.supaMemoryPrompt} placeholder="Leave it blank to use default"/>
-            <span class="text-textcolor">{language.HypaMemory} Model</span>
-            <SelectInput className="mt-2 mb-2" bind:value={DBState.db.hypaModel}>
-                <OptionInput value="MiniLM">MiniLM-L6-v2 (Free / Local)</OptionInput>
-                <OptionInput value="nomic">Nomic (Free / Local)</OptionInput>
-                <OptionInput value="ada">OpenAI Ada (Davinci / Curie Only)</OptionInput>
-            </SelectInput>
             <span class="text-textcolor">{language.hypaChunkSize}</span>
             <NumberInput size="sm" marginBottom bind:value={DBState.db.hypaChunkSize} min={100} />
             <span class="text-textcolor">{language.hypaAllocatedTokens}</span>
@@ -454,17 +448,31 @@
                 <span class="text-textcolor">{language.SuperMemory} Prompt</span>
                 <TextInput size="sm" marginBottom bind:value={DBState.db.supaMemoryPrompt} placeholder="Leave it blank to use default"/>
             {/if}
-            {#if DBState.db.hypaMemory}
-                <span class="text-textcolor">{language.HypaMemory} Model</span>
-                <SelectInput className="mt-2 mb-2" bind:value={DBState.db.hypaModel}>
-                    <OptionInput value="MiniLM" >MiniLM-L6-v2 (Free / Local)</OptionInput>
-                    <OptionInput value="ada" >OpenAI Ada (Davinci / Curie Only)</OptionInput>
-                </SelectInput>
-            {/if}
             <div class="flex">
                 <Check bind:check={DBState.db.hypaMemory} name={language.enable + ' ' + language.HypaMemory}/>
             </div>
         {/if}
 
+        <span class="text-textcolor">{language.embedding}</span>
+        <SelectInput className="mt-2 mb-2" bind:value={DBState.db.hypaModel}>
+            {#if 'gpu' in navigator}
+                <OptionInput value="nomicGPU">Nomic Embed Text v1.5 (GPU)</OptionInput>
+                <OptionInput value="bgeSmallEnGPU">BGE Small English (GPU)</OptionInput>
+                <OptionInput value="bgem3GPU">BGE Medium 3 (GPU)</OptionInput>
+            {/if}
+            <OptionInput value="MiniLM">MiniLM L6 v2 (CPU)</OptionInput>
+            <OptionInput value="nomic">Nomic Embed Text v1.5 (CPU)</OptionInput>
+            <OptionInput value="bgeSmallEn">BGE Small English (CPU)</OptionInput>
+            <OptionInput value="bgem3">BGE Medium 3 (CPU)</OptionInput>
+            <OptionInput value="openai3small">OpenAI text-embedding-3-small</OptionInput>
+            <OptionInput value="openai3large">OpenAI text-embedding-3-large</OptionInput>
+            <OptionInput value="ada">OpenAI Ada</OptionInput>
+        </SelectInput>
+
+        {#if DBState.db.hypaModel === 'openai3small' || DBState.db.hypaModel === 'openai3large' || DBState.db.hypaModel === 'ada'}
+            <span class="text-textcolor">OpenAI API Key</span>
+            <TextInput size="sm" marginBottom bind:value={DBState.db.supaMemoryKey}/>
+        {/if}
+
     </Arcodion>
 {/if}

src/lib/UI/ModelList.svelte

@@ -72,9 +72,6 @@
                     {/each}
                 {/await}
             </Arcodion>
-            {#if DBState.db.plugins.length > 0}
-                <button onclick={() => {changeModel('custom')}} class="hover:bg-selected px-6 py-2 text-lg" >Plugin</button>
-            {/if}
             <div class="text-textcolor2 text-xs">
                 <CheckInput name={language.showUnrecommended}  grayText bind:check={showUnrec}/>
             </div>

src/ts/characterCards.ts

@@ -88,7 +88,7 @@ export async function importCharacterProcess(f:{
             return
         }
         const card:CharacterCardV3 = JSON.parse(cardData)
-        if(CCardLib.character.check(card) !== 'v3'){
+        if(card.spec !== 'chara_card_v3'){
             alertError(language.errors.noData)
             return
         }
@@ -1316,7 +1316,7 @@ export async function exportCharacterCard(char:character, type:'png'|'json'|'cha
                             path = `assets/${type}/${itype}/${name}.${card.data.assets[i].ext}`
                         }
                         card.data.assets[i].uri = 'embeded://' + path
-                        await writer.write(path, rData)
+                        await writer.write(path, Buffer.from(await convertImage(rData)))
                     }
                 }
             }

src/ts/model/modellist.ts

@@ -13,6 +13,7 @@ export enum LLMFlags{
     hasStreaming,
     requiresAlternateRole,
     mustStartWithUserInput,
+    poolSupported
 }
 
 export enum LLMProvider{
@@ -52,6 +53,22 @@ export enum LLMFormat{
     AWSBedrockClaude
 }
 
+export enum LLMTokenizer{
+    Unknown,
+    tiktokenCl100kBase,
+    tiktokenO200Base,
+    Mistral,
+    Llama,
+    NovelAI,
+    Claude,
+    NovelList,
+    Llama3,
+    Gemma,
+    GoogleCloud,
+    Cohere,
+    Local
+}
+
 export interface LLMModel{
     id: string
     name: string
@@ -61,7 +78,8 @@ export interface LLMModel{
     provider: LLMProvider
     flags: LLMFlags[]
     format: LLMFormat
-    parameters: Parameter[]
+    parameters: Parameter[],
+    tokenizer: LLMTokenizer
     recommended?: boolean
 }
 
@@ -92,6 +110,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.OpenAICompatible,
         flags: [LLMFlags.hasFullSystemPrompt, LLMFlags.hasStreaming],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'instructgpt35',
@@ -101,6 +120,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.OpenAILegacyInstruct,
         flags: [LLMFlags.hasFullSystemPrompt, LLMFlags.hasStreaming],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_turbo',
@@ -110,6 +130,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.OpenAICompatible,
         flags: [LLMFlags.hasFullSystemPrompt, LLMFlags.hasStreaming],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4o',
@@ -124,6 +145,7 @@ export const LLMModels: LLMModel[] = [
         ],
         recommended: true,
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4om',
@@ -138,6 +160,7 @@ export const LLMModels: LLMModel[] = [
         ],
         recommended: true,
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4',
@@ -150,6 +173,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_32k',
@@ -162,6 +186,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_16k',
@@ -174,6 +199,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_0314',
@@ -186,6 +212,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_0613',
@@ -198,6 +225,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_32k_0613',
@@ -210,6 +238,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_1106',
@@ -222,6 +251,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_0125',
@@ -234,6 +264,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_1106',
@@ -246,6 +277,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_0613',
@@ -258,6 +290,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_16k_0613',
@@ -270,6 +303,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt35_0301',
@@ -282,6 +316,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_0125',
@@ -294,6 +329,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gptvi4_1106',
@@ -306,6 +342,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4_turbo_20240409',
@@ -318,6 +355,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenCl100kBase
     },
     {
         id: 'gpt4o-2024-05-13',
@@ -331,6 +369,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4o-2024-08-06',
@@ -344,6 +383,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4o-2024-11-20',
@@ -357,6 +397,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4o-chatgpt',
@@ -370,6 +411,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4o1-preview',
@@ -382,6 +424,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         id: 'gpt4o1-mini',
@@ -394,6 +437,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.tiktokenO200Base
     },
     {
         name: "Claude 3.5 Sonnet",
@@ -409,6 +453,7 @@ export const LLMModels: LLMModel[] = [
         ],
         recommended: true,
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: "Claude 3.5 Haiku",
@@ -424,6 +469,7 @@ export const LLMModels: LLMModel[] = [
         ],
         recommended: true,
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3.5 Sonnet (20241022)',
@@ -438,6 +484,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: "Claude 3.5 Haiku (20241022)",
@@ -452,6 +499,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3 Haiku (20240307)',
@@ -466,6 +514,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3.5 Sonnet (20240620)',
@@ -480,6 +529,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3 Opus (20240229)',
@@ -494,6 +544,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3 Sonnet (20240229)',
@@ -508,6 +559,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasStreaming
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 2.1',
@@ -518,6 +570,7 @@ export const LLMModels: LLMModel[] = [
             LLMFlags.hasPrefill,
         ],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 2',
@@ -526,6 +579,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 2 100k',
@@ -534,6 +588,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude v1',
@@ -542,6 +597,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude v1 100k',
@@ -550,6 +606,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude Instant v1',
@@ -558,6 +615,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude Instant v1 100k',
@@ -566,6 +624,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude v1.2',
@@ -574,6 +633,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude v1.0',
@@ -582,6 +642,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AnthropicLegacy,
         flags: [LLMFlags.hasPrefill],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3.5 Sonnet (20241022) v2',
@@ -590,6 +651,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AWSBedrockClaude,
         flags: [LLMFlags.hasPrefill, LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3.5 Sonnet (20240620) v1',
@@ -598,6 +660,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AWSBedrockClaude,
         flags: [LLMFlags.hasPrefill, LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Claude 3 Opus (20240229) v1',
@@ -606,6 +669,7 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.AWSBedrockClaude,
         flags: [LLMFlags.hasPrefill, LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
         parameters: ClaudeParameters,
+        tokenizer: LLMTokenizer.Claude
     },
     {
         name: 'Ooba',
@@ -614,7 +678,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.Ooba,
         flags: [LLMFlags.hasFirstSystemPrompt],
         recommended: true,
-        parameters: []
+        parameters: [],
+        tokenizer: LLMTokenizer.Llama
     },
     {
         name: 'Mancer',
@@ -622,7 +687,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.AsIs,
         format: LLMFormat.OobaLegacy,
         flags: [LLMFlags.hasFirstSystemPrompt],
-        parameters: []
+        parameters: [],
+        tokenizer: LLMTokenizer.Llama
     },
     {
         name: 'OpenRouter',
@@ -631,7 +697,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.OpenAICompatible,
         flags: [LLMFlags.hasFullSystemPrompt, LLMFlags.hasImageInput, LLMFlags.hasStreaming],
         parameters: ['temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'repetition_penalty', 'min_p', 'top_a', 'top_k'],
-        recommended: true
+        recommended: true,
+        tokenizer: LLMTokenizer.Unknown
     },
     {
         name: 'Mistral Small Latest',
@@ -641,7 +708,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.Mistral,
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.mustStartWithUserInput, LLMFlags.requiresAlternateRole],
         recommended: true,
-        parameters: ['temperature', 'presence_penalty', 'frequency_penalty']
+        parameters: ['temperature', 'presence_penalty', 'frequency_penalty'],
+        tokenizer: LLMTokenizer.Mistral
     },
     {
         name: 'Mistral Medium Latest',
@@ -651,7 +719,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.Mistral,
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.mustStartWithUserInput, LLMFlags.requiresAlternateRole],
         recommended: true,
-        parameters: ['temperature', 'presence_penalty', 'frequency_penalty']
+        parameters: ['temperature', 'presence_penalty', 'frequency_penalty'],
+        tokenizer: LLMTokenizer.Mistral
     },
     {
         name: 'Mistral Large 2411',
@@ -660,7 +729,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.Mistral,
         format: LLMFormat.Mistral,
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.mustStartWithUserInput, LLMFlags.requiresAlternateRole],
-        parameters: ['temperature', 'presence_penalty', 'frequency_penalty']
+        parameters: ['temperature', 'presence_penalty', 'frequency_penalty'],
+        tokenizer: LLMTokenizer.Mistral
     },
     {
         name: 'Mistral Nemo',
@@ -669,7 +739,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.Mistral,
         format: LLMFormat.Mistral,
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.mustStartWithUserInput, LLMFlags.requiresAlternateRole],
-        parameters: ['temperature', 'presence_penalty', 'frequency_penalty']
+        parameters: ['temperature', 'presence_penalty', 'frequency_penalty'],
+        tokenizer: LLMTokenizer.Mistral
     },
     {
         name: 'Mistral Large Latest',
@@ -679,7 +750,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.Mistral,
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.mustStartWithUserInput, LLMFlags.requiresAlternateRole],
         parameters: ['temperature', 'presence_penalty', 'frequency_penalty'],
-        recommended: true
+        recommended: true,
+        tokenizer: LLMTokenizer.Mistral
     },
     {
         name: "Gemini Pro 1.5 0827",
@@ -687,16 +759,27 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters:  ['temperature', 'top_k', 'top_p']
+        parameters:  ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Exp 1121",
         id: 'gemini-exp-1121',
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
-        flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
+        flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported],
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud,
+    },
+    {
+        name: "Gemini Exp 1206",
+        id: 'gemini-exp-1206',
+        provider: LLMProvider.GoogleCloud,
+        format: LLMFormat.GoogleCloud,
+        flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt, LLMFlags.poolSupported],
         recommended: true,
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Pro 1.5",
@@ -705,7 +788,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
         recommended: true,
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Flash 1.5",
@@ -714,7 +798,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
         recommended: true,
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Exp 1121",
@@ -723,7 +808,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.VertexAI,
         format: LLMFormat.VertexAIGemini,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.Gemma
     },
     {
         name: "Gemini Pro 1.5",
@@ -732,7 +818,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.VertexAI,
         format: LLMFormat.VertexAIGemini,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.Gemma
     },
     {
         name: "Gemini Flash 1.5",
@@ -741,7 +828,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.VertexAI,
         format: LLMFormat.VertexAIGemini,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.Gemma
     },
     {
         name: "Gemini Exp 1114",
@@ -749,7 +837,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Pro 1.5 002",
@@ -757,7 +846,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Flash 1.5 002",
@@ -765,7 +855,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Pro",
@@ -773,7 +864,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Pro Vision",
@@ -781,7 +873,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Ultra",
@@ -789,7 +882,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: "Gemini Ultra Vision",
@@ -797,7 +891,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.GoogleCloud,
         format: LLMFormat.GoogleCloud,
         flags: [LLMFlags.hasImageInput, LLMFlags.hasFirstSystemPrompt],
-        parameters: ['temperature', 'top_k', 'top_p']
+        parameters: ['temperature', 'top_k', 'top_p'],
+        tokenizer: LLMTokenizer.GoogleCloud
     },
     {
         name: 'Kobold',
@@ -812,7 +907,8 @@ export const LLMModels: LLMModel[] = [
             'repetition_penalty',
             'top_k',
             'top_a'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Unknown
     },
     {
         name: "SuperTrin",
@@ -820,7 +916,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.NovelList,
         format: LLMFormat.NovelList,
         flags: [],
-        parameters: []
+        parameters: [],
+        tokenizer: LLMTokenizer.NovelList
     },
     {
         name: "Damsel",
@@ -828,7 +925,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.NovelList,
         format: LLMFormat.NovelList,
         flags: [],
-        parameters: []
+        parameters: [],
+        tokenizer: LLMTokenizer.NovelList
     },
     {
         name: "Command R",
@@ -840,7 +938,8 @@ export const LLMModels: LLMModel[] = [
         recommended: true,
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Command R Plus",
@@ -852,7 +951,8 @@ export const LLMModels: LLMModel[] = [
         recommended: true,
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Command R 08-2024",
@@ -863,7 +963,8 @@ export const LLMModels: LLMModel[] = [
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole, LLMFlags.mustStartWithUserInput],
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Command R 03-2024",
@@ -874,7 +975,8 @@ export const LLMModels: LLMModel[] = [
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole, LLMFlags.mustStartWithUserInput],
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Command R Plus 08-2024",
@@ -885,7 +987,8 @@ export const LLMModels: LLMModel[] = [
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole, LLMFlags.mustStartWithUserInput],
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Command R Plus 04-2024",
@@ -896,7 +999,8 @@ export const LLMModels: LLMModel[] = [
         flags: [LLMFlags.hasFirstSystemPrompt, LLMFlags.requiresAlternateRole, LLMFlags.mustStartWithUserInput],
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.Cohere
     },
     {
         name: "Clio",
@@ -907,7 +1011,8 @@ export const LLMModels: LLMModel[] = [
         recommended: true,
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.NovelAI
     },
     {
         name: "Kayra",
@@ -918,7 +1023,8 @@ export const LLMModels: LLMModel[] = [
         recommended: true,
         parameters: [
             'temperature', 'top_k', 'top_p', 'presence_penalty', 'frequency_penalty'
-        ]
+        ],
+        tokenizer: LLMTokenizer.NovelAI
     },
     {
         id: 'ollama-hosted',
@@ -926,7 +1032,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.AsIs,
         format: LLMFormat.Ollama,
         flags: [LLMFlags.hasFullSystemPrompt],
-        parameters: OpenAIParameters
+        parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.Unknown
     },
     {
         id: 'hf:::Xenova/opt-350m',
@@ -934,7 +1041,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.WebLLM,
         format: LLMFormat.WebLLM,
         flags: [LLMFlags.hasFullSystemPrompt],
-        parameters: OpenAIParameters
+        parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.Local
     },
     {
         id: 'hf:::Xenova/tiny-random-mistral',
@@ -942,7 +1050,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.WebLLM,
         format: LLMFormat.WebLLM,
         flags: [LLMFlags.hasFullSystemPrompt],
-        parameters: OpenAIParameters
+        parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.Local
     },
     {
         id: 'hf:::Xenova/gpt2-large-conversational',
@@ -950,7 +1059,8 @@ export const LLMModels: LLMModel[] = [
         provider: LLMProvider.WebLLM,
         format: LLMFormat.WebLLM,
         flags: [LLMFlags.hasFullSystemPrompt],
-        parameters: OpenAIParameters
+        parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.Local
     },
     {
         id: 'custom',
@@ -959,7 +1069,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.Plugin,
         flags: [LLMFlags.hasFullSystemPrompt],
         recommended: true,
-        parameters: ['temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'repetition_penalty', 'min_p', 'top_a', 'top_k']
+        parameters: ['temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'repetition_penalty', 'min_p', 'top_a', 'top_k'],
+        tokenizer: LLMTokenizer.Unknown
     },
     {
         id: 'reverse_proxy',
@@ -968,7 +1079,8 @@ export const LLMModels: LLMModel[] = [
         format: LLMFormat.OpenAICompatible,
         flags: [LLMFlags.hasFullSystemPrompt, LLMFlags.hasStreaming],
         recommended: true,
-        parameters: ['temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'repetition_penalty', 'min_p', 'top_a', 'top_k']
+        parameters: ['temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'repetition_penalty', 'min_p', 'top_a', 'top_k'],
+        tokenizer: LLMTokenizer.Unknown
     }
 ]
 
@@ -1002,7 +1114,8 @@ export function getModelInfo(id: string): LLMModel{
             provider: LLMProvider.WebLLM,
             format: LLMFormat.WebLLM,
             flags: [],
-            parameters: OpenAIParameters
+            parameters: OpenAIParameters,
+            tokenizer: LLMTokenizer.Local
         }
     }
     if(id.startsWith('horde:::')){
@@ -1016,7 +1129,8 @@ export function getModelInfo(id: string): LLMModel{
             provider: LLMProvider.Horde,
             format: LLMFormat.Horde,
             flags: [],
-            parameters: OpenAIParameters
+            parameters: OpenAIParameters,
+            tokenizer: LLMTokenizer.Unknown
         }
     }
 
@@ -1029,7 +1143,8 @@ export function getModelInfo(id: string): LLMModel{
         provider: LLMProvider.AsIs,
         format: LLMFormat.OpenAICompatible,
         flags: [],
-        parameters: OpenAIParameters
+        parameters: OpenAIParameters,
+        tokenizer: LLMTokenizer.Unknown
     }
 }
 

src/ts/parser.svelte.ts

@@ -318,9 +318,17 @@ async function parseAdditionalAssets(data:string, char:simpleCharacterArgument|c
                 }
             }
         }
-        const path = assetPaths[name]
+        let path = assetPaths[name]
         if(!path){
-            return ''
+            if(DBState.db.legacyMediaFindings){
+                return ''
+            }
+
+            path = getClosestMatch(name, assetPaths)
+
+            if(!path){
+                return ''
+            }
         }
         switch(type){
             case 'raw':
@@ -366,6 +374,50 @@ async function parseAdditionalAssets(data:string, char:simpleCharacterArgument|c
     return data
 }
 
+function getClosestMatch(name:string, assetPaths:{[key:string]:{path:string, ext?:string}}){
+    if(Object.keys(assetPaths).length === 0){
+        return null
+    }
+
+    const dest = (a:string, b:string) => {
+        let d:Int16Array[] = []
+
+        for(let i=0;i<a.length+1;i++){
+            d.push(new Int16Array(b.length+1))
+        }
+    
+        for(let i=0;i<=a.length;i++){
+            d[i][0] = i
+        }
+
+        for(let i=0;i<=b.length;i++){
+            d[0][i] = i
+        }
+    
+        for(let i=1; i<=a.length; i++){
+            for(let j=1;j<=b.length;j++){
+                d[i][j] = Math.min(
+                    d[i-1][j-1] + (a.charAt(i-1)===b.charAt(j-1) ? 0 : 1),
+                    d[i-1][j]+1, d[i][j-1]+1
+                )
+            }
+        }
+    
+        return d[a.length][b.length];
+    }
+
+    let closest = ''
+    let closestDist = 999999
+    for(const key in assetPaths){
+        const dist = dest(name.trim().replace(/[_ ]/g, ''), key.trim().replace(/[_ ]/g, ''))
+        if(dist < closestDist){
+            closest = key
+            closestDist = dist
+        }
+    }
+    return assetPaths[closest]
+}
+
 async function parseInlayImages(data:string){
     const inlayMatch = data.match(/{{inlay::(.+?)}}/g)
     if(inlayMatch){

src/ts/process/embedding/addinfo.ts

@@ -3,7 +3,7 @@ import { HypaProcesser } from '../memory/hypamemory'
 import { getUserName } from "src/ts/util";
 
 export async function additionalInformations(char: character,chats:Chat,){
-    const processer = new HypaProcesser('MiniLM')
+    const processer = new HypaProcesser()
     const db = getDatabase()
 
     const info = char.additionalText

src/ts/process/files/image.ts

@@ -84,7 +84,7 @@ export async function getInlayImage(id: string){
 
 export function supportsInlayImage(){
     const db = getDatabase()
-    return db.aiModel.startsWith('gptv') || db.aiModel === 'gemini-pro-vision' || db.aiModel.startsWith('claude-3') || db.aiModel.startsWith('gpt4_turbo') || db.aiModel.startsWith('gpt5') || db.aiModel.startsWith('gpt4o') ||
+    return db.aiModel.startsWith('gptv') || db.aiModel === 'gemini-pro-vision' || db.aiModel.startsWith('gemini-exp') || db.aiModel.startsWith('claude-3') || db.aiModel.startsWith('gpt4_turbo') || db.aiModel.startsWith('gpt5') || db.aiModel.startsWith('gpt4o') ||
         (db.aiModel === 'reverse_proxy' && (
             db.proxyRequestModel?.startsWith('gptv') || db.proxyRequestModel === 'gemini-pro-vision' || db.proxyRequestModel?.startsWith('claude-3') || db.proxyRequestModel.startsWith('gpt4_turbo') ||
             db.proxyRequestModel?.startsWith('gpt5') || db.proxyRequestModel?.startsWith('gpt4o') ||

src/ts/process/files/multisend.ts

@@ -124,7 +124,7 @@ async function sendPDFFile(arg:sendFileArg) {
         }
     }
     console.log(texts)
-    const hypa = new HypaProcesser('MiniLM')
+    const hypa = new HypaProcesser()
     hypa.addText(texts)
     const result = await hypa.similaritySearch(arg.query)
     let message = ''
@@ -142,7 +142,7 @@ async function sendTxtFile(arg:sendFileArg) {
     const lines = arg.file.split('\n').filter((a) => {
         return a !== ''
     })
-    const hypa = new HypaProcesser('MiniLM')
+    const hypa = new HypaProcesser()
     hypa.addText(lines)
     const result = await hypa.similaritySearch(arg.query)
     let message = ''
@@ -157,7 +157,7 @@ async function sendTxtFile(arg:sendFileArg) {
 }
 
 async function sendXMLFile(arg:sendFileArg) {
-    const hypa = new HypaProcesser('MiniLM')
+    const hypa = new HypaProcesser()
     let nodeTexts:string[] = []
     const parser = new DOMParser();
     const xmlDoc = parser.parseFromString(arg.file, "text/xml");

src/ts/process/index.svelte.ts

@@ -1,10 +1,10 @@
 import { get, writable } from "svelte/store";
-import { type character, type MessageGenerationInfo, type Chat } from "../storage/database.svelte";
+import { type character, type MessageGenerationInfo, type Chat, changeToPreset } from "../storage/database.svelte";
 import { DBState } from '../stores.svelte';
 import { CharEmotion, selectedCharID } from "../stores.svelte";
 import { ChatTokenizer, tokenize, tokenizeNum } from "../tokenizer";
 import { language } from "../../lang";
-import { alertError } from "../alert";
+import { alertError, alertToast } from "../alert";
 import { loadLoreBookV3Prompt } from "./lorebook.svelte";
 import { findCharacterbyId, getAuthorNoteDefaultText, getPersonaPrompt, getUserName, isLastCharPunctuation, trimUntilPunctuation } from "../util";
 import { requestChatData } from "./request";
@@ -109,6 +109,23 @@ export async function sendChat(chatProcessIndex = -1,arg:{
     }
     doingChat.set(true)
 
+    if(chatProcessIndex === -1 && DBState.db.presetChain){
+        const names = DBState.db.presetChain.split(',').map((v) => v.trim())
+        const randomSelect = Math.floor(Math.random() * names.length)
+        const ele = names[randomSelect]
+
+        const findId = DBState.db.botPresets.findIndex((v) => {
+            return v.name === ele
+        })
+
+        if(findId === -1){
+            alertToast(`Cannot find preset: ${ele}`)
+        }
+        else{
+            changeToPreset(findId, true)
+        }
+    }
+
     if(connectionOpen){
         chatProcessStage.set(4)
         const peerSafe = await peerSafeCheck()
@@ -1396,7 +1413,7 @@ export async function sendChat(chatProcessIndex = -1,arg:{
             }
 
             if(DBState.db.emotionProcesser === 'embedding'){
-                const hypaProcesser = new HypaProcesser('MiniLM')
+                const hypaProcesser = new HypaProcesser()
                 await hypaProcesser.addText(emotionList.map((v) => 'emotion:' + v))
                 let searched = (await hypaProcesser.similaritySearchScored(result)).map((v) => {
                     v[0] = v[0].replace("emotion:",'')

src/ts/process/lua.ts

@@ -213,7 +213,7 @@ export async function runLua(code:string, arg:{
                 if(!LuaLowLevelIds.has(id)){
                     return
                 }
-                const processer = new HypaProcesser('MiniLM')
+                const processer = new HypaProcesser()
                 await processer.addText(value)
                 return await processer.similaritySearch(source)
             })

src/ts/process/memory/hanuraiMemory.ts

@@ -13,7 +13,7 @@ export async function hanuraiMemory(chats:OpenAIChat[],arg:{
 }){
     const db = getDatabase()
     const tokenizer = arg.tokenizer
-    const processer = new HypaProcesser('MiniLM')
+    const processer = new HypaProcesser()
     let addTexts:string[] = []
     const queryStartIndex=chats.length-maxRecentChatQuery
     console.log(chats.length,maxRecentChatQuery,queryStartIndex)

src/ts/process/memory/hypamemory.ts

@@ -1,22 +1,49 @@
 import localforage from "localforage";
-import {globalFetch} from "src/ts/globalApi.svelte";
-import {runEmbedding} from "../transformers";
-import {appendLastPath} from "src/ts/util";
+import { globalFetch } from "src/ts/globalApi.svelte";
+import { runEmbedding } from "../transformers";
+import { alertError } from "src/ts/alert";
+import { appendLastPath } from "src/ts/util";
+import { getDatabase } from "src/ts/storage/database.svelte";
 
 
+export type HypaModel = 'ada'|'MiniLM'|'nomic'|'custom'|'nomicGPU'|'bgeSmallEn'|'bgeSmallEnGPU'|'bgem3'|'bgem3GPU'|'openai3small'|'openai3large'
+
+const localModels = {
+    models: {
+        'MiniLM':'Xenova/all-MiniLM-L6-v2',
+        'nomic':'nomic-ai/nomic-embed-text-v1.5',
+        'nomicGPU':'nomic-ai/nomic-embed-text-v1.5',
+        'bgeSmallEn': 'BAAI/bge-small-en-v1.5',
+        'bgeSmallEnGPU': 'BAAI/bge-small-en-v1.5',
+        'bgem3': 'BAAI/bge-m3',
+        'bgem3GPU': 'BAAI/bge-m3',
+    },
+    gpuModels:[
+        'nomicGPU',
+        'bgeSmallEnGPU',
+        'bgem3GPU'
+    ]
+}
+
 export class HypaProcesser{
     oaikey:string
     vectors:memoryVector[]
     forage:LocalForage
-    model:'ada'|'MiniLM'|'nomic'|'custom'
+    model:HypaModel
     customEmbeddingUrl:string
 
-    constructor(model:'ada'|'MiniLM'|'nomic'|'custom',customEmbeddingUrl?:string){
+    constructor(model:HypaModel|'auto' = 'auto',customEmbeddingUrl?:string){
         this.forage = localforage.createInstance({
             name: "hypaVector"
         })
         this.vectors = []
-        this.model = model
+        if(model === 'auto'){
+            const db = getDatabase()
+            this.model = db.hypaModel || 'MiniLM'
+        }
+        else{
+            this.model = model
+        }
         this.customEmbeddingUrl = customEmbeddingUrl
     }
 
@@ -38,9 +65,9 @@ export class HypaProcesser{
     
     
     async getEmbeds(input:string[]|string):Promise<VectorArray[]> {
-        if(this.model === 'MiniLM' || this.model === 'nomic'){
+        if(Object.keys(localModels.models).includes(this.model)){
             const inputs:string[] = Array.isArray(input) ? input : [input]
-            let results:Float32Array[] = await runEmbedding(inputs, this.model === 'nomic' ? 'nomic-ai/nomic-embed-text-v1.5' : 'Xenova/all-MiniLM-L6-v2')
+            let results:Float32Array[] = await runEmbedding(inputs, localModels.models[this.model], localModels.gpuModels.includes(this.model) ? 'webgpu' : 'wasm')
             return results
         }
         let gf = null;
@@ -57,14 +84,21 @@ export class HypaProcesser{
                 },
             })
         }
-        if(this.model === 'ada'){
+        if(this.model === 'ada' || this.model === 'openai3small' || this.model === 'openai3large'){
+            const db = getDatabase()
+            const models = {
+                'ada':'text-embedding-ada-002',
+                'openai3small':'text-embedding-3-small',
+                'openai3large':'text-embedding-3-large'
+            }
+
             gf = await globalFetch("https://api.openai.com/v1/embeddings", {
                 headers: {
-                "Authorization": "Bearer " + this.oaikey
+                    "Authorization": "Bearer " + db.supaMemoryKey || this.oaikey
                 },
                 body: {
-                "input": input,
-                "model": "text-embedding-ada-002"
+                    "input": input,
+                    "model": models[this.model]
                 }
             })
         }
@@ -138,7 +172,7 @@ export class HypaProcesser{
     }
 
     async similaritySearchScored(query: string) {
-        return await this.similaritySearchVectorWithScore((await this.getEmbeds(query))[0],)
+        return await this.similaritySearchVectorWithScore((await this.getEmbeds(query))[0],);
     }
 
     private async similaritySearchVectorWithScore(

src/ts/process/memory/hypav2.ts

@@ -30,6 +30,7 @@ async function summary(
     stringlizedChat: string
 ): Promise<{ success: boolean; data: string }> {
     const db = getDatabase();
+    console.log("Summarizing");
 
     if (db.supaModelType === "distilbart") {
         try {
@@ -101,35 +102,27 @@ async function summary(
             supaPrompt.replaceAll("{{slot}}", stringlizedChat)
         );
 
-        const promptbody: OpenAIChat[] = parsedPrompt ?? [
+        const promptbody: OpenAIChat[] = (parsedPrompt ?? [
             {
                 role: "user",
                 content: stringlizedChat,
             },
             {
                 role: "system",
-                content: supaPrompt,
-            },
-        ];
-        console.log(
-            "Using submodel: ",
-            db.subModel,
-            "for supaMemory model"
-        );
-        const da = await requestChatData(
-            {
-                formated: promptbody,
-                bias: {},
-                useStreaming: false,
-                noMultiGen: true,
-            },
-            "memory"
-        );
-        if (
-            da.type === "fail" ||
-            da.type === "streaming" ||
-            da.type === "multiline"
-        ) {
+                content: supaPrompt
+            }
+        ]).map(message => ({
+            ...message,
+            memo: "supaPrompt"
+        }));
+        console.log("Using submodel: ", db.subModel, "for supaMemory model");
+        const da = await requestChatData({
+            formated: promptbody,
+            bias: {},
+            useStreaming: false,
+            noMultiGen: true
+        }, 'memory');
+        if (da.type === 'fail' || da.type === 'streaming' || da.type === 'multiline') {
             return {
                 success: false,
                 data: "SupaMemory: HTTP: " + da.result,
@@ -179,7 +172,7 @@ export async function regenerateSummary(
     mainChunkIndex: number
 ) : Promise<void> {
     const targetMainChunk = data.mainChunks[mainChunkIndex];
-    
+
 }
 export async function hypaMemoryV2(
     chats: OpenAIChat[],
@@ -209,6 +202,7 @@ export async function hypaMemoryV2(
     currentTokens += allocatedTokens + chats.length * 4; // ChatML token counting from official openai documentation
     let mainPrompt = "";
     const lastTwoChats = chats.slice(-2);
+    // Error handling for infinite summarization attempts
     let summarizationFailures = 0;
     const maxSummarizationFailures = 3;
 

src/ts/process/models/modelString.ts

@@ -4,7 +4,7 @@ export function getGenerationModelString(){
     const db = getDatabase()
     switch (db.aiModel){
         case 'reverse_proxy':
-            return 'reverse_proxy-' + (db.reverseProxyOobaMode ? 'ooba' : db.proxyRequestModel)
+            return 'custom-' + (db.reverseProxyOobaMode ? 'ooba' : db.customProxyRequestModel)
         case 'openrouter':
             return 'openrouter-' + db.openrouterRequestModel
         default:

src/ts/process/request.ts

@@ -1364,7 +1364,6 @@ async function requestGoogleCloudVertex(arg:RequestDataArgumentExtended):Promise
 
 
     let reformatedChat:GeminiChat[] = []
-    let pendingImage = ''
     let systemPrompt = ''
 
     if(formated[0].role === 'system'){
@@ -1374,10 +1373,7 @@ async function requestGoogleCloudVertex(arg:RequestDataArgumentExtended):Promise
 
     for(let i=0;i<formated.length;i++){
         const chat = formated[i]
-        if(chat.memo && chat.memo.startsWith('inlayImage')){
-            pendingImage = chat.content
-            continue
-        }
+  
         if(i === 0){
             if(chat.role === 'user' || chat.role === 'assistant'){
                 reformatedChat.push({
@@ -1403,7 +1399,34 @@ async function requestGoogleCloudVertex(arg:RequestDataArgumentExtended):Promise
                 chat.role === 'assistant' ? 'MODEL' :
                 chat.role
 
-            if(prevChat.role === qRole){
+            if (chat.multimodals && chat.multimodals.length > 0 && chat.role === "user") {
+                let geminiParts: GeminiPart[] = [];
+                
+                geminiParts.push({
+                    text: chat.content,
+                });
+                
+                for (const modal of chat.multimodals) {
+                    if (modal.type === "image") {
+                        const dataurl = modal.base64;
+                        const base64 = dataurl.split(",")[1];
+                        const mediaType = dataurl.split(";")[0].split(":")[1];
+            
+                        geminiParts.push({
+                            inlineData: {
+                                mimeType: mediaType,
+                                data: base64,
+                            }
+                        });
+                    }
+                }
+        
+                reformatedChat.push({
+                    role: "USER",
+                    parts: geminiParts,
+                });
+        
+            } else if (prevChat.role === qRole) {
                 reformatedChat[reformatedChat.length-1].parts[0].text += '\n' + chat.content
                 continue
             }
@@ -1420,36 +1443,7 @@ async function requestGoogleCloudVertex(arg:RequestDataArgumentExtended):Promise
                     })
                 }
             }
-            else if(chat.role === 'user' && pendingImage !== ''){
-                //conver image to jpeg so it can be inlined
-                const canv = document.createElement('canvas')
-                const img = new Image()
-                img.src = pendingImage  
-                await img.decode()
-                canv.width = img.width
-                canv.height = img.height
-                const ctx = canv.getContext('2d')
-                ctx.drawImage(img, 0, 0)
-                const base64 = canv.toDataURL('image/jpeg').replace(/^data:image\/jpeg;base64,/, "")
-                const mimeType = 'image/jpeg'
-                pendingImage = ''
-                canv.remove()
-                img.remove()
 
-                reformatedChat.push({
-                    role: "USER",
-                    parts: [
-                    {
-                        text: chat.content,
-                    },
-                    {
-                        inlineData: {
-                            mimeType: mimeType,
-                            data: base64
-                        }
-                    }]
-                })
-            }
             else if(chat.role === 'assistant' || chat.role === 'user'){
                 reformatedChat.push({
                     role: chat.role === 'user' ? 'USER' : 'MODEL',
@@ -1578,14 +1572,24 @@ async function requestGoogleCloudVertex(arg:RequestDataArgumentExtended):Promise
         }
     }
 
-    const url = arg.customURL ?? (arg.modelInfo.format === LLMFormat.VertexAIGemini ?
-        `https://${REGION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/us-central1/publishers/google/models/${arg.modelInfo.internalID}:streamGenerateContent`
-        : `https://generativelanguage.googleapis.com/v1beta/models/${arg.modelInfo.internalID}:generateContent?key=${db.google.accessToken}`)
+    let url = ''
+    
+    if(arg.customURL){
+        const u = new URL(arg.customURL)
+        u.searchParams.set('key', db.proxyKey)
+        url = u.toString()
+    }
+    else if(arg.modelInfo.format === LLMFormat.VertexAIGemini){
+        url =`https://${REGION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/us-central1/publishers/google/models/${arg.modelInfo.internalID}:streamGenerateContent`
+    }
+    else{
+        url = `https://generativelanguage.googleapis.com/v1beta/models/${arg.modelInfo.internalID}:generateContent?key=${db.google.accessToken}`
+    }
     const res = await globalFetch(url, {
         headers: headers,
         body: body,
         chatId: arg.chatId,
-        abortSignal: arg.abortSignal
+        abortSignal: arg.abortSignal,
     })
 
     if(!res.ok){
@@ -2533,7 +2537,7 @@ async function requestWebLLM(arg:RequestDataArgumentExtended):Promise<requestDat
         top_p: db.ooba.top_p,
         repetition_penalty: db.ooba.repetition_penalty,
         typical_p: db.ooba.typical_p,
-    })
+    } as any)
     return {
         type: 'success',
         result: unstringlizeChat(v.generated_text as string, formated, currentChar?.name ?? '')

src/ts/process/scripts.ts

@@ -321,7 +321,7 @@ export async function processScriptFull(char:character|groupChat|simpleCharacter
             }
         }
 
-        const processer = new HypaProcesser('MiniLM')
+        const processer = new HypaProcesser()
         await processer.addText(assetNames)
         const matches = data.matchAll(assetRegex)
 

src/ts/process/transformers.ts

@@ -1,4 +1,4 @@
-import {env, AutoTokenizer, pipeline, type SummarizationOutput, type TextGenerationConfig, type TextGenerationOutput, FeatureExtractionPipeline, TextToAudioPipeline, type ImageToTextOutput } from '@xenova/transformers';
+import {env, AutoTokenizer, pipeline, type SummarizationOutput, type TextGenerationConfig, type TextGenerationOutput, FeatureExtractionPipeline, TextToAudioPipeline, type ImageToTextOutput } from '@huggingface/transformers';
 import { unzip } from 'fflate';
 import { globalFetch, loadAsset, saveAsset } from 'src/ts/globalApi.svelte';
 import { selectSingleFile } from 'src/ts/util';
@@ -15,6 +15,7 @@ async function initTransformers(){
     env.useBrowserCache = false
     env.useFSCache = false
     env.useCustomCache = true
+    env.allowLocalModels = true
     env.customCache = {
         put: async (url:URL|string, response:Response) => {
             await tfCache.put(url, response)
@@ -33,10 +34,12 @@ async function initTransformers(){
     console.log('transformers loaded')
 }
 
-export const runTransformers = async (baseText:string, model:string,config:TextGenerationConfig = {}) => {
+export const runTransformers = async (baseText:string, model:string,config:TextGenerationConfig, device:'webgpu'|'wasm' = 'wasm') => {
     await initTransformers()
     let text = baseText
-    let generator = await pipeline('text-generation', model);
+    let generator = await pipeline('text-generation', model, {
+        device
+    });
     let output = await generator(text, config) as TextGenerationOutput
     const outputOne = output[0]
     return outputOne
@@ -50,16 +53,25 @@ export const runSummarizer = async (text: string) => {
 }
 
 let extractor:FeatureExtractionPipeline = null
+let lastEmbeddingModelQuery:string = ''
 type EmbeddingModel = 'Xenova/all-MiniLM-L6-v2'|'nomic-ai/nomic-embed-text-v1.5'
-export const runEmbedding = async (texts: string[], model:EmbeddingModel = 'Xenova/all-MiniLM-L6-v2'):Promise<Float32Array[]> => {
+export const runEmbedding = async (texts: string[], model:EmbeddingModel = 'Xenova/all-MiniLM-L6-v2', device:'webgpu'|'wasm'):Promise<Float32Array[]> => {
     await initTransformers()
-    if(!extractor){
-        extractor = await pipeline('feature-extraction', model);
+    console.log('running embedding')
+    let embeddingModelQuery = model + device
+    if(!extractor || embeddingModelQuery !== lastEmbeddingModelQuery){
+        extractor = await pipeline('feature-extraction', model, {
+            device: device,
+            progress_callback: (progress) => {
+                console.log(progress)
+            }
+        });
+        console.log('extractor loaded')
     }
     let result = await extractor(texts, { pooling: 'mean', normalize: true });
     console.log(texts, result)
     const data = result.data as Float32Array
-
+    console.log(data)
     const lenPerText = data.length / texts.length
     let res:Float32Array[] = []
     for(let i = 0; i < texts.length; i++){

src/ts/process/triggers.ts

@@ -459,7 +459,7 @@ export async function runTrigger(char:character,mode:triggerMode, arg:{
                         break
                     }
 
-                    const processer = new HypaProcesser('MiniLM')
+                    const processer = new HypaProcesser()
                     const effectValue = risuChatParser(effect.value,{chara:char})
                     const source = risuChatParser(effect.source,{chara:char})
                     await processer.addText(effectValue.split('§'))

src/ts/storage/database.svelte.ts

@@ -12,7 +12,7 @@ import { defaultColorScheme, type ColorScheme } from '../gui/colorscheme';
 import type { PromptItem, PromptSettings } from '../process/prompt';
 import type { OobaChatCompletionRequestParams } from '../model/ooba';
 
-export let appVer = "141.2.0"
+export let appVer = "143.0.1"
 export let webAppSubVer = ''
 
 
@@ -703,7 +703,7 @@ export interface Database{
     colorSchemeName:string
     promptTemplate?:PromptItem[]
     forceProxyAsOpenAI?:boolean
-    hypaModel:'ada'|'MiniLM'
+    hypaModel:HypaModel
     saveTime?:number
     mancerHeader:string
     emotionProcesser:'submodel'|'embedding',
@@ -857,6 +857,9 @@ export interface Database{
     notification: boolean
     customFlags: LLMFlags[]
     enableCustomFlags: boolean
+    googleClaudeTokenizing: boolean
+    presetChain: string
+    legacyMediaFindings?:boolean
 }
 
 interface SeparateParameters{
@@ -1597,6 +1600,7 @@ import { decodeRPack, encodeRPack } from '../rpack/rpack_bg';
 import { DBState, selectedCharID } from '../stores.svelte';
 import { LLMFlags, LLMFormat } from '../model/modellist';
 import type { Parameter } from '../process/request';
+import type { HypaModel } from '../process/memory/hypamemory';
 
 export async function downloadPreset(id:number, type:'json'|'risupreset'|'return' = 'json'){
     saveCurrentPreset()

src/ts/tokenizer.ts

@@ -6,6 +6,7 @@ import { supportsInlayImage } from "./process/files/image";
 import { risuChatParser } from "./parser.svelte";
 import { tokenizeGGUFModel } from "./process/models/local";
 import { globalFetch } from "./globalApi.svelte";
+import { getModelInfo, LLMTokenizer } from "./model/modellist";
 
 
 export const tokenizerList = [
@@ -44,58 +45,86 @@ export async function encode(data:string):Promise<(number[]|Uint32Array|Int32Arr
                 return await tikJS(data, 'o200k_base')
         }
     }
-    if(db.aiModel.startsWith('novellist')){
+    const modelInfo = getModelInfo(db.aiModel)
+
+    if(modelInfo.tokenizer === LLMTokenizer.NovelList){
         const nv= await tokenizeWebTokenizers(data, 'novellist')
         return nv
     }
-    if(db.aiModel.startsWith('claude')){
+    if(modelInfo.tokenizer === LLMTokenizer.Claude){
         return await tokenizeWebTokenizers(data, 'claude')
     }
-    if(db.aiModel.startsWith('novelai')){
+    if(modelInfo.tokenizer === LLMTokenizer.NovelAI){
         return await tokenizeWebTokenizers(data, 'novelai')
     }
-    if(db.aiModel.startsWith('mistral')){
+    if(modelInfo.tokenizer === LLMTokenizer.Mistral){
         return await tokenizeWebTokenizers(data, 'mistral')
     }
-    if(db.aiModel === 'mancer' ||
-        db.aiModel === 'textgen_webui' ||
-        (db.aiModel === 'reverse_proxy' && db.reverseProxyOobaMode)){
+    if(modelInfo.tokenizer === LLMTokenizer.Llama){
         return await tokenizeWebTokenizers(data, 'llama')
     }
-    if(db.aiModel.startsWith('local_')){
+    if(modelInfo.tokenizer === LLMTokenizer.Local){
         return await tokenizeGGUFModel(data)
     }
-    if(db.aiModel === 'ooba'){
-        if(db.reverseProxyOobaArgs.tokenizer === 'mixtral' || db.reverseProxyOobaArgs.tokenizer === 'mistral'){
-            return await tokenizeWebTokenizers(data, 'mistral')
-        }
-        else if(db.reverseProxyOobaArgs.tokenizer === 'llama'){
-            return await tokenizeWebTokenizers(data, 'llama')
-        }
-        else{
-            return await tokenizeWebTokenizers(data, 'llama')
-        }
-    }
-    if(db.aiModel.startsWith('gpt4o')){
+    if(modelInfo.tokenizer === LLMTokenizer.tiktokenO200Base){
         return await tikJS(data, 'o200k_base')
     }
-    if(db.aiModel.startsWith('gemini')){
+    if(modelInfo.tokenizer === LLMTokenizer.GoogleCloud && db.googleClaudeTokenizing){
+        return await tokenizeGoogleCloud(data)
+    }
+    if(modelInfo.tokenizer === LLMTokenizer.Gemma || modelInfo.tokenizer === LLMTokenizer.GoogleCloud){
         return await tokenizeWebTokenizers(data, 'gemma')
     }
-    if(db.aiModel.startsWith('cohere')){
+    if(modelInfo.tokenizer === LLMTokenizer.Cohere){
         return await tokenizeWebTokenizers(data, 'cohere')
     }
 
     return await tikJS(data)
 }
 
-type tokenizerType = 'novellist'|'claude'|'novelai'|'llama'|'mistral'|'llama3'|'gemma'|'cohere'
+type tokenizerType = 'novellist'|'claude'|'novelai'|'llama'|'mistral'|'llama3'|'gemma'|'cohere'|'googleCloud'
 
 let tikParser:Tiktoken = null
 let tokenizersTokenizer:Tokenizer = null
 let tokenizersType:tokenizerType = null
 let lastTikModel = 'cl100k_base'
 
+let googleCloudTokenizedCache = new Map<string, number>()
+
+async function tokenizeGoogleCloud(text:string) {
+    const db = getDatabase()
+    const model = getModelInfo(db.aiModel)
+
+    if(googleCloudTokenizedCache.has(text + model.internalID)){
+        const count = googleCloudTokenizedCache.get(text)
+        return new Uint32Array(count)
+    }
+
+    const res = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${model.internalID}:countTokens?key=${db.google?.accessToken}`, {
+        method: 'POST',
+        headers: {
+            "Content-Type": "application/json",
+        },
+        body: JSON.stringify({
+            contents: [{
+                parts:[{
+                    text: text
+                }]
+            }]
+        }),
+    })
+
+    if(res.status !== 200){
+        return await tokenizeWebTokenizers(text, 'gemma')
+    }
+
+    const json = await res.json()
+    googleCloudTokenizedCache.set(text + model.internalID, json.totalTokens as number)
+    const count = json.totalTokens as number
+
+    return new Uint32Array(count)
+}
+
 async function tikJS(text:string, model='cl100k_base') {
     if(!tikParser || lastTikModel !== model){
         if(model === 'cl100k_base'){