utilitas 1997.1.21 → 1997.1.23

package/lib/alan.mjs

@@ -18,17 +18,18 @@ const _NEED = [
 ];
 
 const [
-    OPENAI, GEMINI, CHATGPT, ASSISTANT, OPENAI_EMBEDDING, GEMINI_EMEDDING,
-    OPENAI_TRAINING, OLLAMA, CLAUDE, GPT_4O_MINI, GPT_4O, GPT_O1, GPT_O1_MINI,
-    GEMINI_20_FLASH, NOVA, EMBEDDING_001, MISTRAL, TEXT_EMBEDDING_3_SMALL,
-    TEXT_EMBEDDING_3_LARGE, CLAUDE_35_SONNET, CLAUDE_35_HAIKU, AUDIO, WAV,
+    OPENAI, GEMINI, CHATGPT, OPENAI_EMBEDDING, GEMINI_EMEDDING, OPENAI_TRAINING,
+    OLLAMA, CLAUDE, GPT_4O_MINI, GPT_4O, GPT_O1, GPT_O1_MINI, GEMINI_20_FLASH,
+    GEMINI_20_FLASH_THINKING, NOVA, EMBEDDING_001, MISTRAL,
+    TEXT_EMBEDDING_3_SMALL, TEXT_EMBEDDING_3_LARGE, CLAUDE_35_SONNET,
+    CLAUDE_35_HAIKU, AUDIO, WAV,
 ] = [
-        'OPENAI', 'GEMINI', 'CHATGPT', 'ASSISTANT', 'OPENAI_EMBEDDING',
-        'GEMINI_EMEDDING', 'OPENAI_TRAINING', 'OLLAMA', 'CLAUDE', 'gpt-4o-mini',
-        'gpt-4o', 'o1-preview', 'o1-mini', 'gemini-2.0-flash-exp', 'nova',
-        'embedding-001', 'mistral', 'text-embedding-3-small',
-        'text-embedding-3-large', 'claude-3-5-sonnet-latest',
-        'claude-3-5-haiku-latest', 'audio', 'wav',
+        'OPENAI', 'GEMINI', 'CHATGPT', 'OPENAI_EMBEDDING', 'GEMINI_EMEDDING',
+        'OPENAI_TRAINING', 'OLLAMA', 'CLAUDE', 'gpt-4o-mini', 'gpt-4o',
+        'o1-preview', 'o1-mini', 'gemini-2.0-flash-exp',
+        'gemini-2.0-flash-thinking-exp', 'nova', 'embedding-001',
+        'mistral', 'text-embedding-3-small', 'text-embedding-3-large',
+        'claude-3-5-sonnet-latest', 'claude-3-5-haiku-latest', 'audio', 'wav',
     ];
 
 const [
@@ -74,7 +75,6 @@ const buildGeminiParts = (text, atcmts) => [{ text }, ...atcmts || []];
 const [DEFAULT_CHAT_ENGINE, OPENAI_DEFAULT_VOICE] = [CHATGPT, NOVA];
 
 const DEFAULT_MODELS = {
-    [ASSISTANT]: GPT_4O,
     [CHATGPT]: GPT_4O,
     [GEMINI_EMEDDING]: EMBEDDING_001,
     [GEMINI]: GEMINI_20_FLASH,
@@ -176,8 +176,8 @@ const MODELS = {
         maxFileSize: 20 * 1024 * 1024, // 20 MB
         maxImagePerPrompt: 3000,
         maxImageSize: Infinity,
-        maxOutputTokens: 8192,
-        maxUrlSize: 2 * 1024 * 1024 * 1024, // 2 GB
+        maxOutputTokens: 1024 * 8,
+        maxUrlSize: 1024 * 1024 * 1024 * 2, // 2 GB
         maxVideoLength: 60 * 50, // 50 minutes
         maxVideoLengthWithAudio: 60 * 50, // 50 minutes
         maxVideoLengthWithoutAudio: 60 * 60, // 1 hour
@@ -193,6 +193,25 @@ const MODELS = {
             flac, mp3, m4a, mpga, opus, pcm, wav, webm, tgpp,
         ],
     },
+    [GEMINI_20_FLASH_THINKING]: {
+        // https://cloud.google.com/vertex-ai/generative-ai/docs/thinking-mode?hl=en
+        contextWindow: 1024 * (8 + 32),
+        imageCostTokens: size8k / (768 * 768) * 258,
+        maxFileSize: 20 * 1024 * 1024, // 20 MB
+        maxImagePerPrompt: 3000,
+        maxImageSize: Infinity,
+        maxOutputTokens: 1024 * 8,
+        maxUrlSize: 1024 * 1024 * 1024 * 2, // 2 GB
+        requestLimitsRPM: 15,
+        requestLimitsRPD: 1500,
+        tokenLimitsTPM: 4 * 1000000,
+        trainingData: 'August 2024',
+        vision: true,
+        json: false,
+        supportedMimeTypes: [
+            png, jpeg,
+        ],
+    },
     [MISTRAL]: {
         contextWindow: 128000,
         requestLimitsRPM: Infinity,
@@ -295,13 +314,12 @@ const init = async (options) => {
             if (options?.apiKey) {
                 const { GoogleGenerativeAI } = await need('@google/generative-ai');
                 const genAi = new GoogleGenerativeAI(options.apiKey);
+                const genModel = options?.model || DEFAULT_MODELS[GEMINI];
                 clients[provider] = {
-                    generative: genAi.getGenerativeModel({
-                        model: options?.model || DEFAULT_MODELS[GEMINI],
-                    }),
+                    generative: genAi.getGenerativeModel({ model: genModel }),
                     embedding: genAi.getGenerativeModel({
                         model: DEFAULT_MODELS[GEMINI_EMEDDING],
-                    }),
+                    }), genModel,
                 };
             }
             break;
@@ -350,25 +368,18 @@ const selectGptAudioModel = options => {
         MODELS[options.model]?.audio,
         `Audio modality is not supported by model: ${options.model}`
     );
-    options.model = MODELS[options.model].audio;
+    return MODELS[options.model]?.audio;
 };
 
 const buildGptMessage = (content, options) => {
     content = content || '';
-    let audioModelSelected = false;
-    if (options?.audioMode) {
-        selectGptAudioModel(options);
-        audioModelSelected = true;
-    }
+    let alterModel = options?.audioMode && selectGptAudioModel(options);
     const attachments = (options?.attachments || []).map(x => {
         assert(MODELS[options?.model], 'Model is required.');
         if (MODELS[options.model]?.supportedMimeTypes?.includes?.(x.mime_type)) {
             return { type: 'image_url', image_url: { url: x.url } };
         } else if (MODELS[options.model]?.supportedAudioTypes?.includes?.(x.mime_type)) {
-            if (!audioModelSelected) {
-                selectGptAudioModel(options);
-                audioModelSelected = true;
-            }
+            alterModel = selectGptAudioModel(options);
             return {
                 type: 'input_audio',
                 input_audio: { data: x.data, format: WAV },
@@ -376,6 +387,7 @@ const buildGptMessage = (content, options) => {
         }
         throwError(`Unsupported mime type: '${x.mime_type}'.`);
     });
+    alterModel && (options.model = alterModel);
     const message = String.isString(content) ? {
         role: options?.role || user,
         content: content.length ? [{ type: 'text', text: content }] : [],
@@ -475,20 +487,26 @@ const promptChatGPT = async (content, options = {}) => {
     options.model = options?.model || DEFAULT_MODELS[CHATGPT];
     const message = buildGptMessage(content, options);
     const modalities = options?.modalities || (
-        options?.audioMode ? ['text', AUDIO] : null
+        options?.audioMode ? ['text', AUDIO] : undefined
     );
+    assert(!(
+        options?.jsonMode && MODELS[options.model]?.json == false
+    ), `This model does not support JSON output: ${options.model}`);
     let format;
     [format, options.audioMimeType, options.suffix]
         = options?.stream ? ['pcm16', pcm16, 'pcm.wav'] : [WAV, wav, WAV];
-    let [resp, resultText, resultAudio, chunk] = [await chatGptClient.chat.completions.create({
-        modalities, audio: options?.audio || (
-            modalities?.find?.(x => x === AUDIO) && {
-                voice: OPENAI_DEFAULT_VOICE, format
-            }
-        ), ...messages([...options?.messages || [], message]),
-        ...options?.jsonMode ? { response_format: { type: JSON_OBJECT } } : {},
-        model: options.model, stream: !!options?.stream,
-    }), '', Buffer.alloc(0), null];
+    let [resp, resultText, resultAudio, chunk] = [
+        await chatGptClient.chat.completions.create({
+            modalities, audio: options?.audio || (
+                modalities?.find?.(x => x === AUDIO) && {
+                    voice: OPENAI_DEFAULT_VOICE, format
+                }
+            ), ...messages([...options?.messages || [], message]),
+            ...options?.jsonMode ? {
+                response_format: { type: JSON_OBJECT }
+            } : {}, model: options.model, stream: !!options?.stream,
+        }), '', Buffer.alloc(0), null
+    ];
     if (!options?.stream) {
         return { response: await packGptResp(resp, options) };
     }
@@ -563,132 +581,6 @@ const promptClaude = async (content, options) => {
     return { response: await packGptResp(options?.stream ? event : resp, options) };
 };
 
-const createAssistant = async (options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    // https://platform.openai.com/docs/api-reference/assistants/createAssistant
-    return await clientBeta.assistants.create({
-        model: DEFAULT_MODELS[ASSISTANT], name, instructions,
-        tools: [CODE_INTERPRETER, RETRIEVAL], // 'FUNCTION'
-        ...options?.params || {},
-        // description: null, file_ids: [], metadata: {},
-    });
-};
-
-const getAssistant = async (assistantId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.assistants.retrieve(assistantId);
-};
-
-const modifyAssistant = async (assistantId, assistant, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.assistants.update(assistantId, assistant);
-};
-
-const deleteAssistant = async (assistantId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    const cleanup = await deleteAllFilesFromAssistant(assistantId, options);
-    const respDel = await clientBeta.assistants.del(assistantId);
-    return { cleanup, delete: respDel };
-};
-
-const listAssistant = async (options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return (await clientBeta.assistants.list({
-        order: 'asc', limit: `${GPT_QUERY_LIMIT}`, ...options?.params || {},
-    })).data;
-};
-
-const ensureAssistant = async opts => {
-    if (opts?.assistantId) { return await getAssistant(opts?.assistantId); }
-    const list = await listAssistant(opts);
-    return list.find(x => x.name === name) || await createAssistant(opts);
-};
-
-const createThread = async (options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    // https://platform.openai.com/docs/api-reference/threads/createThread
-    return await clientBeta.threads.create({ ...options?.params || {} });
-};
-
-const getThread = async (threadId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    // https://platform.openai.com/docs/api-reference/threads/getThread
-    return await clientBeta.threads.retrieve(threadId);
-};
-
-const deleteThread = async (threadId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.threads.del(threadId);
-};
-
-const ensureThread = async options => await (
-    options?.threadId ? getThread(options?.threadId) : createThread(options)
-);
-
-const createMessage = async (threadId, content, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    // https://platform.openai.com/docs/api-reference/messages/createMessage
-    return await clientBeta.threads.messages.create(
-        threadId, buildGptMessage(content, options),
-    );
-};
-
-const listMessages = async (threadId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    // https://platform.openai.com/docs/api-reference/messages/listMessages
-    return await clientBeta.threads.messages.list(
-        threadId, { ...options?.params || {} }
-    );
-};
-
-const getLatestMessage = async (threadId, options) => (await listMessages(
-    threadId, { ...options, params: { ...options?.params || {}, limit: 1 } }
-))?.data?.[0];
-
-const run = async (assistantId, threadId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    let content = '';
-    const stream = async (delta, snapshot) => {
-        content += delta.value;
-        await ignoreErrFunc(async () => await options?.stream?.(
-            options?.raw ? delta : await packResp(content), snapshot
-        ), LOG);
-    };
-    let resp = clientBeta.threads.runs[options?.runOnly ? 'create' : 'stream'](
-        threadId, { assistant_id: assistantId }
-    );
-    if (!options?.runOnly) {
-        resp = await new Promise((resolve, reject) => {
-            // https://github.com/openai/openai-node/blob/master/helpers.md#assistant-events
-            resp.on('textCreated', stream)
-                .on('textDelta', stream)
-                .on('textDone', async (delta, snapshot) => resolve(
-                    options?.raw ? delta : await packResp(delta.value), snapshot
-                ))
-                .on('error', reject)
-        });
-    }
-    return await resp;
-};
-
-const getRun = async (threadId, runId, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.threads.runs.retrieve(threadId, runId);
-};
-
-const promptAssistant = async (content, options) => {
-    const assistant = await ensureAssistant(options);
-    const thread = await ensureThread(options);
-    const messageSent = await createMessage(thread.id, content, options);
-    const runResp = await run(assistant.id, thread.id, options);
-    const threadDeleted = options?.deleteThread
-        ? await deleteThread(thread.id) : null;
-    return {
-        assistant, thread, messageSent, run: options?.runOnly ? runResp : null,
-        threadDeleted, response: options?.runOnly ? null : runResp,
-    };
-};
-
 const uploadFile = async (input, options) => {
     const { client } = await getOpenAIClient(options);
     const { content: file, cleanup } = await convert(input, {
@@ -701,44 +593,10 @@ const uploadFile = async (input, options) => {
     return resp;
 };
 
-const uploadFileForAssistants = async (content, options) => await uploadFile(
-    content, { ...options, params: { purpose: 'assistants' } }
-);
-
 const uploadFileForFineTuning = async (content, options) => await uploadFile(
     content, { suffix: 'jsonl', ...options, params: { purpose: 'fine-tune' } }
 );
 
-const attachFileToAssistant = async (assistantId, file_id, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.assistants.files.create(assistantId, { file_id });
-};
-
-const detachFileFromAssistant = async (assistantId, file_id, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    return await clientBeta.assistants.files.del(assistantId, file_id);
-};
-
-const deleteFileFromAssistant = async (assistantId, file_id, options) => {
-    const detach = await detachFileFromAssistant(assistantId, file_id, options);
-    const respDel = await deleteFile(file_id, options);
-    return { detach, delete: respDel };
-};
-
-const deleteAllFilesFromAssistant = async (assistantId, options) => {
-    const files = await listAssistantFiles(assistantId, options);
-    const delPms = [];
-    for (const file of files) {
-        delPms.push(deleteFileFromAssistant(assistantId, file.id, options));
-    }
-    return await Promise.all(delPms);
-};
-
-const uploadFileForRetrieval = async (assistantId, content, options) => {
-    const file = await uploadFileForAssistants(content, options);
-    return await attachFileToAssistant(assistantId, file.id, options);
-};
-
 const listFiles = async (options) => {
     const { client } = await getOpenAIClient(options);
     const files = [];
@@ -752,14 +610,6 @@ const deleteFile = async (file_id, options) => {
     return await client.files.del(file_id);
 };
 
-const listAssistantFiles = async (assistant_id, options) => {
-    const { clientBeta } = await getOpenAIClient(options);
-    const resp = await clientBeta.assistants.files.list(
-        assistant_id, { limit: GPT_QUERY_LIMIT, ...options?.params }
-    );
-    return options?.raw ? resp : resp.data;
-};
-
 const generationConfig = options => ({
     generationConfig: {
         ...options?.generationConfig || {},
@@ -787,10 +637,14 @@ const handleGeminiResponse = async (resp, options) => {
 };
 
 const promptGemini = async (content, options) => {
-    const { generative } = await getGeminiClient(options);
+    const { generative, genModel } = await getGeminiClient(options);
     // https://github.com/google/generative-ai-js/blob/main/samples/node/advanced-chat.js
     // @todo: check this issue similar to Vertex AI:
     // Google's bug: history is not allowed while using inline_data?
+    console.log(genModel);
+    assert(!(
+        options?.jsonMode && MODELS[genModel]?.json == false
+    ), `This model does not support JSON output: ${genModel}`);
     const chat = generative.startChat({
         history: options?.messages && !options?.attachments?.length
             ? options.messages : [],
@@ -922,29 +776,13 @@ const initChat = async (options) => {
         chatConfig.engines[key] = options.engines[i];
         chatConfig.engines[key].model = chatConfig.engines[key].model || model;
         const mxPmpt = MODELS[chatConfig.engines[key].model].maxInputTokens / 2;
-        let pmptTokens = 0;
-        switch (key) {
-            case ASSISTANT:
-                chatConfig.engines[key].assistantId
-                    = chatConfig.engines[key].assistantId
-                    || (await ensureAssistant({
-                        ...options, params: {
-                            name, model: chatConfig.engines[key].model,
-                            instructions: chatConfig.systemPrompt,
-                            ...chatConfig.engines[key].params,
-                        }
-                    })).id;
-                break;
-            case CHATGPT: case GEMINI: case OLLAMA: case CLAUDE:
-                pmptTokens = await countTokens([buildGeminiHistory(
-                    chatConfig.systemPrompt, { role: system }
-                )]); // Use Gemini instead of ChatGPT because of the longer pack
-                assert(
-                    pmptTokens < mxPmpt,
-                    `System prompt is too long: ${pmptTokens} / ${mxPmpt} tokens.`
-                );
-                break;
-        }
+        const pmptTokens = await countTokens([buildGeminiHistory(
+            chatConfig.systemPrompt, { role: system }
+        )]); // Use Gemini instead of ChatGPT because of the longer pack
+        assert(
+            pmptTokens < mxPmpt,
+            `System prompt is too long: ${pmptTokens} / ${mxPmpt} tokens.`
+        );
     }
     return chatConfig;
 };
@@ -1012,8 +850,6 @@ const talk = async (input, options) => {
             };
             msgBuilder()
             break;
-        case ASSISTANT:
-            break;
         case GEMINI:
             sys.push(buildGeminiHistory(session.systemPrompt, { role: user }));
             msgBuilder = () => {
@@ -1061,13 +897,6 @@ const talk = async (input, options) => {
                 messages, model, ...options, attachments,
             });
             break;
-        case ASSISTANT:
-            resp = await promptAssistant(input, {
-                assistantId: chatConfig.engines[engine].assistantId,
-                threadId: session.threadId, ...options, attachments,
-            });
-            session.threadId = resp.thread.id;
-            break;
         case GEMINI:
             resp = await promptGemini(input, {
                 messages, ...options, attachments,
@@ -1233,50 +1062,29 @@ export {
     buildGptTrainingCases,
     cancelGptFineTuningJob,
     countTokens,
-    createAssistant,
     createGeminiEmbedding,
     createGptFineTuningJob,
-    createMessage,
     createOpenAIEmbedding,
-    deleteAllFilesFromAssistant,
-    deleteAssistant,
     deleteFile,
-    deleteFileFromAssistant,
-    deleteThread,
-    detachFileFromAssistant,
     distillFile,
-    ensureAssistant,
-    ensureThread,
-    getAssistant,
     getGptFineTuningJob,
-    getLatestMessage,
     getMaxChatPromptLimit,
-    getRun,
     getSession,
-    getThread,
     init,
     initChat,
-    listAssistant,
-    listAssistantFiles,
     listFiles,
     listGptFineTuningEvents,
     listGptFineTuningJobs,
-    listMessages,
     listOpenAIModels,
-    modifyAssistant,
     prompt,
-    promptAssistant,
     promptChatGPT,
     promptClaude,
     promptGemini,
     promptOllama,
     resetSession,
-    run,
     tailGptFineTuningEvents,
     talk,
     trimPrompt,
     uploadFile,
-    uploadFileForAssistants,
     uploadFileForFineTuning,
-    uploadFileForRetrieval,
 };

package/lib/manifest.mjs

@@ -1,7 +1,7 @@
 const manifest = {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1997.1.21",
+    "version": "1997.1.23",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -46,10 +46,10 @@ const manifest = {
         "js-tiktoken": "^1.0.16",
         "jsdom": "^25.0.1",
         "lorem-ipsum": "^2.0.8",
-        "mailgun.js": "^10.2.4",
+        "mailgun.js": "^10.3.0",
         "mailparser": "^3.7.2",
         "mime": "^4.0.6",
-        "mysql2": "^3.11.5",
+        "mysql2": "^3.12.0",
         "node-mailjet": "^6.0.6",
         "node-polyfill-webpack-plugin": "^4.1.0",
         "office-text-extractor": "^3.0.3",

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1997.1.21",
+    "version": "1997.1.23",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -57,10 +57,10 @@
         "js-tiktoken": "^1.0.16",
         "jsdom": "^25.0.1",
         "lorem-ipsum": "^2.0.8",
-        "mailgun.js": "^10.2.4",
+        "mailgun.js": "^10.3.0",
         "mailparser": "^3.7.2",
         "mime": "^4.0.6",
-        "mysql2": "^3.11.5",
+        "mysql2": "^3.12.0",
         "node-mailjet": "^6.0.6",
         "node-polyfill-webpack-plugin": "^4.1.0",
         "office-text-extractor": "^3.0.3",