updated alan

lib/alan.mjs

@@ -675,10 +675,7 @@ const checkEmbeddingInput = async (input, model) => {
     const getInput = () => arrInput.join(' ');
     const _model = MODELS[model];
     assert(_model, `Invalid model: '${model}'.`);
-    while ((await countTokens(getInput(), { fast: true }) > _model.contextWindow)
-        || (await countTokens(getInput()) > _model.contextWindow)) {
-        arrInput.pop();
-    }
+    await trimPrompt(getInput, arrInput.pop, _model.contextWindow);
     return getInput();
 };
 
@@ -925,16 +922,16 @@ const talk = async (input, options) => {
         default:
             throwError(`Invalid AI engine: '${engine}'.`);
     }
-    while (await countTokens([
+    await trimPrompt(() => [
         ...sys, ...messages, buildVertexMessage(input, { role: user })
-    ]) >= _MODEL.maxInputTokens) {
+    ], () => {
         if (messages.length) {
             session.messages.shift();
             msgBuilder && msgBuilder();
         } else {
             input = trimTailing(trimTailing(input).slice(0, -1)) + '...';
         }
-    }
+    }, _MODEL.maxInputTokens);
     const chat = { request: input };
     const attachments = [];
     (options?.attachments || []).filter(
@@ -1030,6 +1027,63 @@ const distillFile = async (attachments, o) => {
     return packResult(resp);
 };
 
+const prompt = async (input, options) => {
+    const preferredEngines = [
+        { client: OPENAI, func: promptChatGPT },
+        { client: VERTEX, func: promptGemini },
+        { client: GEMINI, func: promptVertex },
+    ];
+    for (const engine of preferredEngines) {
+        if (clients[engine.client]) {
+            const extra = {};
+            if (engine.client === OPENAI && options?.fast) {
+                extra.model = GPT_4O_MINI;
+            }
+            return await engine.func(input, { ...extra, ...options || {} });
+        }
+    }
+    throwError('No AI provider is available.');
+};
+
+const trimPrompt = async (getPrompt, trimFunc, contextWindow, options) => {
+    let i = 0;
+    while ((await countTokens(await getPrompt(), { fast: true }) > contextWindow)
+        || (await countTokens(await getPrompt()) > contextWindow)) {
+        await trimFunc();
+        if (~~options?.maxTry && ++i >= ~~options?.maxTry) { break; }
+    };
+};
+
+const analyzeSessions = async (sessionIds, options) => {
+    chatConfig.sessions = options?.session;
+    const ids = ensureArray(sessionIds);
+    const mdl = MODELS[DEFAULT_MODELS[DEFAULT_CHAT_ENGINE]];
+    const [sses, resp] = [{}, {}];
+    for (const i in ids) { sses[ids[i]] = await getSession(ids[i], options); }
+    const pmt = options?.prompt || (
+        'Help me organize the dialogues in the following JSON into a title '
+        + 'dictionary and return it in JSON format. The returned JSON uses the '
+        + 'key from the original JSON session data as the key and the title as '
+        + 'the value. The following is a conversation data that needs to be '
+        + 'organized: ');
+    const getInput = () => `${pmt}\n\`\`\`JSON\n${JSON.stringify(sses, null, 2)}\n\`\`\``;
+    mdl.contextWindow = 2000
+    await trimPrompt(getInput, () => {
+        if (!Object.values(sses).sort((x, y) =>
+            y.messages.length - x.messages.length)[0].messages.shift()) {
+            delete sses[Object.keys(sses).map(x => [
+                x, JSON.stringify(sses[x]).length,
+            ]).sort((x, y) => y[1] - x[1])[0][0]];
+        }
+    }, mdl.contextWindow);
+    const aiResp = (await prompt(getInput(), {
+        jsonMode: true, fast: true, ...options || {}
+    }))?.response[0].json;
+    assert(aiResp, 'Unable to analyze sessions.');
+    ids.map(x => resp[x] = aiResp[x] || null)
+    return Array.isArray(sessionIds) ? resp : resp[sessionIds[0]];
+};
+
 export default init;
 export {
     _NEED,
@@ -1050,6 +1104,7 @@ export {
     MODELS,
     RETRIEVAL,
     TEXT_EMBEDDING_3_SMALL,
+    analyzeSessions,
     buildGptTrainingCase,
     buildGptTrainingCases,
     cancelGptFineTuningJob,
@@ -1074,6 +1129,7 @@ export {
     getLatestMessage,
     getMaxChatPromptLimit,
     getRun,
+    getSession,
     getThread,
     init,
     initChat,
@@ -1085,6 +1141,7 @@ export {
     listMessages,
     listOpenAIModels,
     modifyAssistant,
+    prompt,
     promptAssistant,
     promptChatGPT,
     promptGemini,
@@ -1094,6 +1151,7 @@ export {
     run,
     tailGptFineTuningEvents,
     talk,
+    trimPrompt,
     uploadFile,
     uploadFileForAssistants,
     uploadFileForFineTuning,