Temperature
diff --git a/public/locales/ar-sa.json b/public/locales/ar-sa.json
index 90ef674cf..1716ffdef 100644
--- a/public/locales/ar-sa.json
+++ b/public/locales/ar-sa.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "الحد الأقصى لطول الاستجابة (الرموز,الحرف)",
"Multiple swipes per generation": "الضربات الشديدة المتعددة لكل جيل",
"Enable OpenAI completion streaming": "تمكين بث الاكتمال من OpenAI",
- "Enable Cohere web-search connector": "تمكين موصل بحث الويب Cohere",
- "Web-search": "البحث في الويب",
- "Allow the model to use the web-search connector.": "اسمح للنموذج باستخدام موصل بحث الويب.",
"Frequency Penalty": "عقوبة التكرار",
"Presence Penalty": "عقوبة الوجود",
"Count Penalty": "عد ضربة جزاء",
diff --git a/public/locales/de-de.json b/public/locales/de-de.json
index ac4e63298..19d048378 100644
--- a/public/locales/de-de.json
+++ b/public/locales/de-de.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Maximale Antwortlänge (Tokens)",
"Multiple swipes per generation": "Mehrere Swipes pro Generation",
"Enable OpenAI completion streaming": "OpenAI-Vervollständigungsstreaming aktivieren",
- "Enable Cohere web-search connector": "Cohere-Websuch-Connector aktivieren",
- "Web-search": "Web-Suche",
- "Allow the model to use the web-search connector.": "Erlauben Sie dem Modell, den Websuch-Connector zu verwenden.",
"Frequency Penalty": "Frequenzstrafe",
"Presence Penalty": "Präsenzstrafe",
"Count Penalty": "Strafe zählen",
diff --git a/public/locales/es-es.json b/public/locales/es-es.json
index b7b746e5d..722595459 100644
--- a/public/locales/es-es.json
+++ b/public/locales/es-es.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Longitud máxima de respuesta (tokens)",
"Multiple swipes per generation": "Múltiples golpes por generación",
"Enable OpenAI completion streaming": "Activar streaming de completado de OpenAI",
- "Enable Cohere web-search connector": "Habilitar el conector de búsqueda web de Cohere",
- "Web-search": "Búsqueda Web",
- "Allow the model to use the web-search connector.": "Permita que el modelo utilice el conector de búsqueda web.",
"Frequency Penalty": "Penalización de frecuencia",
"Presence Penalty": "Penalización de presencia",
"Count Penalty": "Penalización de conteo",
diff --git a/public/locales/fr-fr.json b/public/locales/fr-fr.json
index e069895b6..551e484c4 100644
--- a/public/locales/fr-fr.json
+++ b/public/locales/fr-fr.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Longueur maximale de la réponse (tokens)",
"Multiple swipes per generation": "Plusieurs balayages par génération",
"Enable OpenAI completion streaming": "Activer le streaming de complétion OpenAI",
- "Enable Cohere web-search connector": "Activer le connecteur de recherche Web Cohere",
- "Web-search": "Recherche Internet",
- "Allow the model to use the web-search connector.": "Autorisez le modèle à utiliser le connecteur de recherche Web.",
"Frequency Penalty": "Pénalité de fréquence",
"Presence Penalty": "Pénalité de présence",
"Count Penalty": "Pénalité de décompte",
diff --git a/public/locales/is-is.json b/public/locales/is-is.json
index 634a7c2c2..36cb62965 100644
--- a/public/locales/is-is.json
+++ b/public/locales/is-is.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Hámarks lengd svörunar (í táknum)",
"Multiple swipes per generation": "Mörg högg á hverja kynslóð",
"Enable OpenAI completion streaming": "Virkja OpenAI klárastreymi",
- "Enable Cohere web-search connector": "Virkja Cohere vefleitartengi",
- "Web-search": "Vefleit",
- "Allow the model to use the web-search connector.": "Leyfðu líkaninu að nota vefleitartengið.",
"Frequency Penalty": "Tíðnarefning",
"Presence Penalty": "Tilkoma refning",
"Count Penalty": "Telja víti",
diff --git a/public/locales/it-it.json b/public/locales/it-it.json
index bab851988..3bdfef772 100644
--- a/public/locales/it-it.json
+++ b/public/locales/it-it.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Lunghezza massima della risposta (token)",
"Multiple swipes per generation": "Più passaggi per generazione",
"Enable OpenAI completion streaming": "Abilita lo streaming di completamento OpenAI",
- "Enable Cohere web-search connector": "Abilita il connettore di ricerca web Cohere",
- "Web-search": "Ricerca sul web",
- "Allow the model to use the web-search connector.": "Consenti al modello di utilizzare il connettore di ricerca web.",
"Frequency Penalty": "Penalità di frequenza",
"Presence Penalty": "Penalità di presenza",
"Count Penalty": "Conte Penalità",
diff --git a/public/locales/ja-jp.json b/public/locales/ja-jp.json
index 88453ad0c..f7652d2d8 100644
--- a/public/locales/ja-jp.json
+++ b/public/locales/ja-jp.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "最大応答長（トークン数）",
"Multiple swipes per generation": "世代ごとに複数のスワイプ",
"Enable OpenAI completion streaming": "OpenAIの完了ストリーミングを有効にする",
- "Enable Cohere web-search connector": "Cohereウェブ検索コネクタを有効にする",
- "Web-search": "ウェブ検索",
- "Allow the model to use the web-search connector.": "モデルが Web 検索コネクタを使用できるようにします。",
"Frequency Penalty": "頻度ペナルティ",
"Presence Penalty": "存在ペナルティ",
"Count Penalty": "カウントペナルティ",
diff --git a/public/locales/ko-kr.json b/public/locales/ko-kr.json
index 919209f92..5c5a73f7c 100644
--- a/public/locales/ko-kr.json
+++ b/public/locales/ko-kr.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "최대 응답 길이 (토큰)",
"Multiple swipes per generation": "세대당 다중 스와이프",
"Enable OpenAI completion streaming": "OpenAI 완성 스트리밍 활성화",
- "Enable Cohere web-search connector": "Cohere 웹 검색 커넥터 활성화",
- "Web-search": "웹 서핑",
- "Allow the model to use the web-search connector.": "모델이 웹 검색 커넥터를 사용하도록 허용합니다.",
"Frequency Penalty": "빈도 패널티",
"Presence Penalty": "존재 패널티",
"Count Penalty": "카운트 페널티",
diff --git a/public/locales/nl-nl.json b/public/locales/nl-nl.json
index a0e7e8f9b..e0a43e09f 100644
--- a/public/locales/nl-nl.json
+++ b/public/locales/nl-nl.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Maximale lengte van het antwoord (tokens)",
"Multiple swipes per generation": "Meerdere swipes per generatie",
"Enable OpenAI completion streaming": "OpenAI voltooiingsstreaming inschakelen",
- "Enable Cohere web-search connector": "Schakel de Cohere-webzoekconnector in",
- "Web-search": "Web-zoeken",
- "Allow the model to use the web-search connector.": "Sta toe dat het model de webzoekconnector gebruikt.",
"Frequency Penalty": "Frequentieboete",
"Presence Penalty": "Aanwezigheidsboete",
"Count Penalty": "Tel straf",
diff --git a/public/locales/pt-pt.json b/public/locales/pt-pt.json
index 1ecfd5222..6087e5db3 100644
--- a/public/locales/pt-pt.json
+++ b/public/locales/pt-pt.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Comprimento Máximo da Resposta (tokens)",
"Multiple swipes per generation": "Vários furtos por geração",
"Enable OpenAI completion streaming": "Ativar streaming de conclusão do OpenAI",
- "Enable Cohere web-search connector": "Ativar o conector de pesquisa na Web Cohere",
- "Web-search": "Pesquisa na internet",
- "Allow the model to use the web-search connector.": "Permita que o modelo use o conector de pesquisa na Web.",
"Frequency Penalty": "Pena de Frequência",
"Presence Penalty": "Pena de Presença",
"Count Penalty": "Contar penalidade",
diff --git a/public/locales/ru-ru.json b/public/locales/ru-ru.json
index cd8068a98..38289cb4a 100644
--- a/public/locales/ru-ru.json
+++ b/public/locales/ru-ru.json
@@ -772,7 +772,6 @@
"Type a message, or /? for help": "Введите сообщение, или /? для получения справки",
"Welcome to SillyTavern!": "Добро пожаловать в SillyTavern!",
"Won't be shared with the character card on export.": "Не попадут в карточку персонажа при экспорте.",
- "Web-search": "Веб-поиск",
"Persona Name:": "Имя персоны:",
"User first message": "Первое сообщение пользователя",
"extension_token_counter": "Токенов:",
@@ -1200,8 +1199,6 @@
"Streaming_desc": "Выводить текст последовательно по мере его генерации.\rЕсли параметр выключен, ответы будут отображаться сразу целиком, и только после полного завершения генерации.",
"Max prompt cost:": "Max prompt cost:",
"TFS": "TFS",
- "Enable Cohere web-search connector": "Enable Cohere web-search connector",
- "Allow the model to use the web-search connector.": "Allow the model to use the web-search connector.",
"Count Penalty": "Count Penalty",
"Min P": "Min P",
"NSFW": "NSFW",
diff --git a/public/locales/uk-ua.json b/public/locales/uk-ua.json
index 7fa6e2ea3..5152fa611 100644
--- a/public/locales/uk-ua.json
+++ b/public/locales/uk-ua.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Довжина відповіді (токени)",
"Multiple swipes per generation": "Кілька свайпів за покоління",
"Enable OpenAI completion streaming": "Увімкнути потокове завершення OpenAI",
- "Enable Cohere web-search connector": "Увімкнути конектор веб-пошуку Cohere",
- "Web-search": "Веб-пошук",
- "Allow the model to use the web-search connector.": "Дозвольте моделі використовувати конектор веб-пошуку.",
"Frequency Penalty": "Штраф за частоту",
"Presence Penalty": "Штраф за наявність",
"Count Penalty": "Рахувати пенальті",
diff --git a/public/locales/vi-vn.json b/public/locales/vi-vn.json
index 268fdce77..79e2594eb 100644
--- a/public/locales/vi-vn.json
+++ b/public/locales/vi-vn.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "Độ dài phản hồi tối đa (token)",
"Multiple swipes per generation": "Vuốt nhiều lần trong một lần tạo",
"Enable OpenAI completion streaming": "Bật streaming của OpenAI",
- "Enable Cohere web-search connector": "Bật web tìm kiếm của Cohere",
- "Web-search": "Tìm kiếm trên web",
- "Allow the model to use the web-search connector.": "Cho phép model sử dụng trình kết nối tìm kiếm trên web.",
"Frequency Penalty": "Frequency Penalty",
"Presence Penalty": "Presence Penalty",
"Count Penalty": "Count Penalty",
diff --git a/public/locales/zh-cn.json b/public/locales/zh-cn.json
index df8b7f831..fc6fae29c 100644
--- a/public/locales/zh-cn.json
+++ b/public/locales/zh-cn.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "最大回复长度（以词符数计）",
"Multiple swipes per generation": "每次生成多个备选回复",
"Enable OpenAI completion streaming": "启用OpenAI文本补全流式传输",
- "Enable Cohere web-search connector": "启用 Cohere 网络搜索连接器",
- "Web-search": "联网搜索",
- "Allow the model to use the web-search connector.": "允许模型使用联网搜索。",
"Frequency Penalty": "频率惩罚",
"Presence Penalty": "存在惩罚",
"Count Penalty": "计数惩罚",
diff --git a/public/locales/zh-tw.json b/public/locales/zh-tw.json
index 59270882f..6e853966a 100644
--- a/public/locales/zh-tw.json
+++ b/public/locales/zh-tw.json
@@ -56,9 +56,6 @@
"Max Response Length (tokens)": "最大回應長度(符記數)",
"Multiple swipes per generation": "每次生成多次滑動",
"Enable OpenAI completion streaming": "啟用 OpenAI 補充串流",
- "Enable Cohere web-search connector": "啟用 Cohere 網頁搜尋連接器",
- "Web-search": "網頁搜尋",
- "Allow the model to use the web-search connector.": "允許模型使用網頁搜尋連接器",
"Frequency Penalty": "頻率懲罰",
"Presence Penalty": "存在懲罰",
"Count Penalty": "計數懲罰",
diff --git a/public/script.js b/public/script.js
index cc3504b60..f5d4bd480 100644
--- a/public/script.js
+++ b/public/script.js
@@ -5436,7 +5436,7 @@ function extractMessageFromData(data) {
case 'novel':
return data.output;
case 'openai':
- return data?.choices?.[0]?.message?.content ?? data?.choices?.[0]?.text ?? data?.text ?? '';
+ return data?.choices?.[0]?.message?.content ?? data?.choices?.[0]?.text ?? data?.text ?? data?.message?.content?.[0]?.text ?? data?.message?.tool_plan ?? '';
default:
return '';
}
diff --git a/public/scripts/openai.js b/public/scripts/openai.js
index 90334f299..7fb23e711 100644
--- a/public/scripts/openai.js
+++ b/public/scripts/openai.js
@@ -225,7 +225,6 @@ const default_settings = {
top_a_openai: 0,
repetition_penalty_openai: 1,
stream_openai: false,
- websearch_cohere: false,
openai_max_context: max_4k,
openai_max_tokens: 300,
wrap_in_quotes: false,
@@ -302,7 +301,6 @@ const oai_settings = {
top_a_openai: 0,
repetition_penalty_openai: 1,
stream_openai: false,
- websearch_cohere: false,
openai_max_context: max_4k,
openai_max_tokens: 300,
wrap_in_quotes: false,
@@ -1847,7 +1845,6 @@ async function sendOpenAIRequest(type, messages, signal) {
generate_data['frequency_penalty'] = Math.min(Math.max(Number(oai_settings.freq_pen_openai), 0), 1);
generate_data['presence_penalty'] = Math.min(Math.max(Number(oai_settings.pres_pen_openai), 0), 1);
generate_data['stop'] = getCustomStoppingStrings(5);
- generate_data['websearch'] = oai_settings.websearch_cohere;
}
if (isPerplexity) {
@@ -1976,12 +1973,14 @@ async function sendOpenAIRequest(type, messages, signal) {
}
function getStreamingReply(data) {
- if (oai_settings.chat_completion_source == chat_completion_sources.CLAUDE) {
+ if (oai_settings.chat_completion_source === chat_completion_sources.CLAUDE) {
return data?.delta?.text || '';
- } else if (oai_settings.chat_completion_source == chat_completion_sources.MAKERSUITE) {
+ } else if (oai_settings.chat_completion_source === chat_completion_sources.MAKERSUITE) {
return data?.candidates?.[0]?.content?.parts?.[0]?.text || '';
+ } else if (oai_settings.chat_completion_source === chat_completion_sources.COHERE) {
+ return data?.delta?.message?.content?.text || data?.delta?.message?.tool_plan || '';
} else {
- return data.choices[0]?.delta?.content ?? data.choices[0]?.message?.content ?? data.choices[0]?.text ?? '';
+ return data.choices?.[0]?.delta?.content ?? data.choices?.[0]?.message?.content ?? data.choices?.[0]?.text ?? '';
}
}
@@ -2857,7 +2856,6 @@ function loadOpenAISettings(data, settings) {
oai_settings.min_p_openai = settings.min_p_openai ?? default_settings.min_p_openai;
oai_settings.repetition_penalty_openai = settings.repetition_penalty_openai ?? default_settings.repetition_penalty_openai;
oai_settings.stream_openai = settings.stream_openai ?? default_settings.stream_openai;
- oai_settings.websearch_cohere = settings.websearch_cohere ?? default_settings.websearch_cohere;
oai_settings.openai_max_context = settings.openai_max_context ?? default_settings.openai_max_context;
oai_settings.openai_max_tokens = settings.openai_max_tokens ?? default_settings.openai_max_tokens;
oai_settings.bias_preset_selected = settings.bias_preset_selected ?? default_settings.bias_preset_selected;
@@ -2931,7 +2929,6 @@ function loadOpenAISettings(data, settings) {
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(); }
$('#stream_toggle').prop('checked', oai_settings.stream_openai);
- $('#websearch_toggle').prop('checked', oai_settings.websearch_cohere);
$('#api_url_scale').val(oai_settings.api_url_scale);
$('#openai_proxy_password').val(oai_settings.proxy_password);
$('#claude_assistant_prefill').val(oai_settings.assistant_prefill);
@@ -3258,7 +3255,6 @@ async function saveOpenAIPreset(name, settings, triggerUi = true) {
personality_format: settings.personality_format,
group_nudge_prompt: settings.group_nudge_prompt,
stream_openai: settings.stream_openai,
- websearch_cohere: settings.websearch_cohere,
prompts: settings.prompts,
prompt_order: settings.prompt_order,
api_url_scale: settings.api_url_scale,
@@ -3682,7 +3678,6 @@ function onSettingsPresetChange() {
personality_format: ['#personality_format_textarea', 'personality_format', false],
group_nudge_prompt: ['#group_nudge_prompt_textarea', 'group_nudge_prompt', false],
stream_openai: ['#stream_toggle', 'stream_openai', true],
- websearch_cohere: ['#websearch_toggle', 'websearch_cohere', true],
prompts: ['', 'prompts', false],
prompt_order: ['', 'prompt_order', false],
api_url_scale: ['#api_url_scale', 'api_url_scale', false],
@@ -4846,11 +4841,6 @@ export function initOpenAI() {
saveSettingsDebounced();
});
- $('#websearch_toggle').on('change', function () {
- oai_settings.websearch_cohere = !!$('#websearch_toggle').prop('checked');
- saveSettingsDebounced();
- });
-
$('#wrap_in_quotes').on('change', function () {
oai_settings.wrap_in_quotes = !!$('#wrap_in_quotes').prop('checked');
saveSettingsDebounced();
diff --git a/public/scripts/sse-stream.js b/public/scripts/sse-stream.js
index 620caf3c5..dbe481e78 100644
--- a/public/scripts/sse-stream.js
+++ b/public/scripts/sse-stream.js
@@ -108,9 +108,21 @@ function getDelay(s) {
* @returns {AsyncGenerator<{data: object, chunk: string}>} The parsed data and the chunk to be sent.
*/
async function* parseStreamData(json) {
+ // Cohere
+ if (typeof json.delta.message === 'object' && ['tool-plan-delta', 'content-delta'].includes(json.type)) {
+ const text = json?.delta?.message?.content?.text ?? '';
+ for (let i = 0; i < text.length; i++) {
+ const str = json.delta.message.content.text[i];
+ yield {
+ data: { ...json, delta: { message: { content: { text: str } } } },
+ chunk: str,
+ };
+ }
+ return;
+ }
// Claude
- if (typeof json.delta === 'object') {
- if (typeof json.delta.text === 'string' && json.delta.text.length > 0) {
+ if (typeof json.delta === 'object' && typeof json.delta.text === 'string') {
+ if (json.delta.text.length > 0) {
for (let i = 0; i < json.delta.text.length; i++) {
const str = json.delta.text[i];
yield {
diff --git a/public/scripts/tool-calling.js b/public/scripts/tool-calling.js
index b43d56da7..41f27964e 100644
--- a/public/scripts/tool-calling.js
+++ b/public/scripts/tool-calling.js
@@ -401,6 +401,22 @@ export class ToolManager {
}
}
}
+ const cohereToolEvents = ['message-start', 'tool-call-start', 'tool-call-delta', 'tool-call-end'];
+ if (cohereToolEvents.includes(parsed?.type) && typeof parsed?.delta?.message === 'object') {
+ const choiceIndex = 0;
+ const toolCallIndex = parsed?.index ?? 0;
+
+ if (!Array.isArray(toolCalls[choiceIndex])) {
+ toolCalls[choiceIndex] = [];
+ }
+
+ if (toolCalls[choiceIndex][toolCallIndex] === undefined) {
+ toolCalls[choiceIndex][toolCallIndex] = {};
+ }
+
+ const targetToolCall = toolCalls[choiceIndex][toolCallIndex];
+ ToolManager.#applyToolCallDelta(targetToolCall, parsed.delta.message);
+ }
if (typeof parsed?.content_block === 'object') {
const choiceIndex = 0;
const toolCallIndex = parsed?.index ?? 0;
@@ -503,6 +519,7 @@ export class ToolManager {
chat_completion_sources.CLAUDE,
chat_completion_sources.OPENROUTER,
chat_completion_sources.GROQ,
+ chat_completion_sources.COHERE,
];
return supportedSources.includes(oai_settings.chat_completion_source);
}
@@ -529,7 +546,15 @@ export class ToolManager {
// Parsed tool calls from streaming data
if (Array.isArray(data) && data.length > 0 && Array.isArray(data[0])) {
- return isClaudeToolCall(data[0]) ? data[0].filter(x => x).map(convertClaudeToolCall) : data[0];
+ if (isClaudeToolCall(data[0])) {
+ return data[0].filter(x => x).map(convertClaudeToolCall);
+ }
+
+ if (typeof data[0]?.[0]?.tool_calls === 'object') {
+ return Array.isArray(data[0]?.[0]?.tool_calls) ? data[0][0].tool_calls : [data[0][0].tool_calls];
+ }
+
+ return data[0];
}
// Parsed tool calls from non-streaming data
@@ -550,6 +575,11 @@ export class ToolManager {
return content;
}
}
+
+ // Cohere tool calls
+ if (typeof data?.message?.tool_calls === 'object') {
+ return Array.isArray(data?.message?.tool_calls) ? data.message.tool_calls : [data.message.tool_calls];
+ }
}
/**
diff --git a/src/cohere-stream.js b/src/cohere-stream.js
deleted file mode 100644
index d59ecad5e..000000000
--- a/src/cohere-stream.js
+++ /dev/null
@@ -1,126 +0,0 @@
-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;
diff --git a/src/endpoints/backends/chat-completions.js b/src/endpoints/backends/chat-completions.js
index d2f963007..c190e6f71 100644
--- a/src/endpoints/backends/chat-completions.js
+++ b/src/endpoints/backends/chat-completions.js
@@ -5,7 +5,6 @@ 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, convertAI21Messages, mergeMessages } = require('../../prompt-converters');
-const CohereStream = require('../../cohere-stream');
const { readSecret, SECRET_KEYS } = require('../secrets');
const { getTokenizerModel, getSentencepiceTokenizer, getTiktokenTokenizer, sentencepieceTokenizers, TEXT_COMPLETION_MODELS } = require('../tokenizers');
@@ -13,7 +12,8 @@ const { getTokenizerModel, getSentencepiceTokenizer, getTiktokenTokenizer, sente
const API_OPENAI = 'https://api.openai.com/v1';
const API_CLAUDE = 'https://api.anthropic.com/v1';
const API_MISTRAL = 'https://api.mistral.ai/v1';
-const API_COHERE = 'https://api.cohere.ai/v1';
+const API_COHERE_V1 = 'https://api.cohere.ai/v1';
+const API_COHERE_V2 = 'https://api.cohere.ai/v2';
const API_PERPLEXITY = 'https://api.perplexity.ai';
const API_GROQ = 'https://api.groq.com/openai/v1';
const API_MAKERSUITE = 'https://generativelanguage.googleapis.com';
@@ -553,13 +553,14 @@ async function sendCohereRequest(request, response) {
try {
const convertedHistory = convertCohereMessages(request.body.messages, request.body.char_name, request.body.user_name);
- const connectors = [];
const tools = [];
- const canDoWebSearch = !String(request.body.model).includes('c4ai-aya');
- if (request.body.websearch && canDoWebSearch) {
- connectors.push({
- id: 'web-search',
+ if (Array.isArray(request.body.tools) && request.body.tools.length > 0) {
+ tools.push(...request.body.tools);
+ tools.forEach(tool => {
+ if (tool?.function?.parameters?.$schema) {
+ delete tool.function.parameters.$schema;
+ }
});
}
@@ -567,9 +568,7 @@ async function sendCohereRequest(request, response) {
const requestBody = {
stream: Boolean(request.body.stream),
model: request.body.model,
- message: convertedHistory.userPrompt,
- preamble: convertedHistory.systemPrompt,
- chat_history: convertedHistory.chatHistory,
+ messages: convertedHistory.chatHistory,
temperature: request.body.temperature,
max_tokens: request.body.max_tokens,
k: request.body.top_k,
@@ -578,16 +577,13 @@ async function sendCohereRequest(request, response) {
stop_sequences: request.body.stop,
frequency_penalty: request.body.frequency_penalty,
presence_penalty: request.body.presence_penalty,
- prompt_truncation: 'AUTO_PRESERVE_ORDER',
- connectors: connectors,
documents: [],
tools: tools,
- search_queries_only: false,
};
const canDoSafetyMode = String(request.body.model).endsWith('08-2024');
if (canDoSafetyMode) {
- requestBody.safety_mode = 'NONE';
+ requestBody.safety_mode = 'OFF';
}
console.log('Cohere request:', requestBody);
@@ -603,11 +599,11 @@ async function sendCohereRequest(request, response) {
timeout: 0,
};
- const apiUrl = API_COHERE + '/chat';
+ const apiUrl = API_COHERE_V2 + '/chat';
if (request.body.stream) {
- const stream = await global.fetch(apiUrl, config);
- parseCohereStream(stream, request, response);
+ const stream = await fetch(apiUrl, config);
+ forwardFetchResponse(stream, response);
} else {
const generateResponse = await fetch(apiUrl, config);
if (!generateResponse.ok) {
@@ -658,7 +654,7 @@ router.post('/status', jsonParser, async function (request, response_getstatus_o
headers = {};
mergeObjectWithYaml(headers, request.body.custom_include_headers);
} else if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.COHERE) {
- api_url = API_COHERE;
+ api_url = API_COHERE_V1;
api_key_openai = readSecret(request.user.directories, SECRET_KEYS.COHERE);
headers = {};
} else if (request.body.chat_completion_source === CHAT_COMPLETION_SOURCES.ZEROONEAI) {
diff --git a/src/prompt-converters.js b/src/prompt-converters.js
index 23dcd05c9..47cd9e368 100644
--- a/src/prompt-converters.js
+++ b/src/prompt-converters.js
@@ -277,38 +277,9 @@ function convertClaudeMessages(messages, prefillString, useSysPrompt, useTools,
* @param {object[]} messages Array of messages
* @param {string} charName Character name
* @param {string} userName User name
- * @returns {{systemPrompt: string, chatHistory: object[], userPrompt: string}} Prompt for Cohere
+ * @returns {{chatHistory: object[]}} Prompt for Cohere
*/
function convertCohereMessages(messages, charName = '', userName = '') {
- const roleMap = {
- 'system': 'SYSTEM',
- 'user': 'USER',
- 'assistant': 'CHATBOT',
- };
- let systemPrompt = '';
-
- // 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;
- 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);
-
if (messages.length === 0) {
messages.unshift({
role: 'user',
@@ -316,17 +287,41 @@ function convertCohereMessages(messages, charName = '', userName = '') {
});
}
- const lastNonSystemMessageIndex = messages.findLastIndex(msg => msg.role === 'user' || msg.role === 'assistant');
- const userPrompt = messages.slice(lastNonSystemMessageIndex).map(msg => msg.content).join('\n\n') || PROMPT_PLACEHOLDER;
-
- const chatHistory = messages.slice(0, lastNonSystemMessageIndex).map(msg => {
- return {
- role: roleMap[msg.role] || 'USER',
- message: msg.content,
- };
+ messages.forEach((msg, index) => {
+ // Tool calls require an assistent primer
+ if (Array.isArray(msg.tool_calls)) {
+ if (index > 0 && messages[index - 1].role === 'assistant') {
+ msg.content = messages[index - 1].content;
+ messages.splice(index - 1, 1);
+ } else {
+ msg.content = `I'm going to call a tool for that: ${msg.tool_calls.map(tc => tc?.function?.name).join(', ')}`;
+ }
+ }
+ // No names support (who would've thought)
+ if (msg.name) {
+ if (msg.role == 'system' && msg.name == 'example_assistant') {
+ if (charName && !msg.content.startsWith(`${charName}: `)) {
+ msg.content = `${charName}: ${msg.content}`;
+ }
+ }
+ if (msg.role == 'system' && msg.name == 'example_user') {
+ if (userName && !msg.content.startsWith(`${userName}: `)) {
+ msg.content = `${userName}: ${msg.content}`;
+ }
+ }
+ if (msg.role !== 'system' && !msg.content.startsWith(`${msg.name}: `)) {
+ msg.content = `${msg.name}: ${msg.content}`;
+ }
+ delete msg.name;
+ }
});
- return { systemPrompt: systemPrompt.trim(), chatHistory, userPrompt };
+ // A prompt should end with a user/tool message
+ if (messages.length && !['user', 'tool'].includes(messages[messages.length - 1].role)) {
+ messages[messages.length - 1].role = 'user';
+ }
+
+ return { chatHistory: messages };
}
/**
diff --git a/src/vectors/cohere-vectors.js b/src/vectors/cohere-vectors.js
index 1ec01130c..96ce5bff5 100644
--- a/src/vectors/cohere-vectors.js
+++ b/src/vectors/cohere-vectors.js
@@ -17,7 +17,7 @@ async function getCohereBatchVector(texts, isQuery, directories, model) {
throw new Error('No API key found');
}
- const response = await fetch('https://api.cohere.ai/v1/embed', {
+ const response = await fetch('https://api.cohere.ai/v2/embed', {
method: 'POST',
headers: {
'Content-Type': 'application/json',
@@ -26,6 +26,7 @@ async function getCohereBatchVector(texts, isQuery, directories, model) {
body: JSON.stringify({
texts: texts,
model: model,
+ embedding_types: ['float'],
input_type: isQuery ? 'search_query' : 'search_document',
truncate: 'END',
}),
@@ -38,12 +39,12 @@ async function getCohereBatchVector(texts, isQuery, directories, model) {
}
const data = await response.json();
- if (!Array.isArray(data?.embeddings)) {
+ if (!Array.isArray(data?.embeddings?.float)) {
console.log('API response was not an array');
throw new Error('API response was not an array');
}
- return data.embeddings;
+ return data.embeddings.float;
}
/**