utilitas 1998.2.18 → 1998.2.19

package/lib/alan.mjs

@@ -1,37 +1,42 @@
+import { fileTypeFromBuffer } from 'file-type';
+import { end, loop } from './event.mjs';
+import { createWavHeader } from './media.mjs';
 import { BASE64, BUFFER, DATAURL, MIME_BINARY, STREAM, convert } from './storage.mjs';
 import { create as createUoid } from './uoid.mjs';
-import { createWavHeader } from './media.mjs';
-import { end, loop } from './event.mjs';
-import { fileTypeFromBuffer } from 'file-type';
 
 import {
+    log as _log,
+    renderText as _renderText,
     base64Encode, ensureArray, ensureString, extract, ignoreErrFunc,
-    log as _log, need, parseJson, renderText as _renderText, throwError,
+    need, parseJson,
+    throwError,
 } from './utilitas.mjs';
 
 const _NEED = [
-    '@anthropic-ai/sdk', '@google/generative-ai', 'js-tiktoken', 'ollama',
-    'OpenAI',
+    '@anthropic-ai/sdk', '@anthropic-ai/vertex-sdk', '@google/generative-ai',
+    'js-tiktoken', 'ollama', 'OpenAI',
 ];
 
 const [
     OPENAI, GEMINI, CHATGPT, OPENAI_EMBEDDING, GEMINI_EMEDDING, OPENAI_TRAINING,
     OLLAMA, CLAUDE, GPT_4O_MINI, GPT_4O, GPT_O1, GPT_O3_MINI, GEMINI_20_FLASH,
     GEMINI_20_FLASH_THINKING, GEMINI_20_PRO, NOVA, EMBEDDING_001, DEEPSEEK_R1,
-    DEEPSEEK_R1_32B, MD_CODE, CHATGPT_REASONING, TEXT_EMBEDDING_3_SMALL,
-    TEXT_EMBEDDING_3_LARGE, CLAUDE_35_SONNET, CLAUDE_35_HAIKU, AUDIO, WAV,
-    CHATGPT_MINI, ATTACHMENTS, CHAT, OPENAI_VOICE, MEDIUM, LOW, HIGH,
-    GPT_REASONING_EFFORT, THINK, THINK_STR, THINK_END,
+    DEEPSEEK_R1_32B, DEEPSEEK_R1_70B, MD_CODE, CHATGPT_REASONING,
+    TEXT_EMBEDDING_3_SMALL, TEXT_EMBEDDING_3_LARGE, CLAUDE_35_SONNET,
+    CLAUDE_35_HAIKU, CLOUD_37_SONNET, AUDIO, WAV, CHATGPT_MINI, ATTACHMENTS,
+    CHAT, OPENAI_VOICE, MEDIUM, LOW, HIGH, GPT_REASONING_EFFORT, THINK,
+    THINK_STR, THINK_END, AZURE,
 ] = [
         'OPENAI', 'GEMINI', 'CHATGPT', 'OPENAI_EMBEDDING', 'GEMINI_EMEDDING',
         'OPENAI_TRAINING', 'OLLAMA', 'CLAUDE', 'gpt-4o-mini', 'gpt-4o', 'o1',
         'o3-mini', 'gemini-2.0-flash', 'gemini-2.0-flash-thinking-exp',
         'gemini-2.0-pro-exp', 'nova', 'embedding-001', 'deepseek-r1',
-        'deepseek-r1:32b', '```', 'CHATGPT_REASONING', 'text-embedding-3-small',
-        'text-embedding-3-large', 'claude-3-5-sonnet-latest',
-        'claude-3-5-haiku-latest', 'audio', 'wav', 'CHATGPT_MINI',
+        'deepseek-r1:32b', 'deepseek-r1:70b', '```', 'CHATGPT_REASONING',
+        'text-embedding-3-small', 'text-embedding-3-large',
+        'claude-3-5-sonnet-latest', 'claude-3-5-haiku-latest',
+        'claude-3-7-sonnet@20250219', 'audio', 'wav', 'CHATGPT_MINI',
         '[ATTACHMENTS]', 'CHAT', 'OPENAI_VOICE', 'medium', 'low', 'high',
-        'medium', 'think', '<think>', '</think>',
+        'medium', 'think', '<think>', '</think>', 'AZURE',
     ];
 
 const [
@@ -81,10 +86,11 @@ const DEFAULT_MODELS = {
     [CHATGPT_MINI]: GPT_4O_MINI,
     [CHATGPT_REASONING]: GPT_O3_MINI,
     [CHATGPT]: GPT_4O,
-    [CLAUDE]: CLAUDE_35_SONNET,
+    [CLAUDE]: CLOUD_37_SONNET,
     [GEMINI_EMEDDING]: EMBEDDING_001,
     [GEMINI]: GEMINI_20_FLASH,
     [OLLAMA]: DEEPSEEK_R1,
+    [AZURE]: DEEPSEEK_R1,
     [OPENAI_EMBEDDING]: TEXT_EMBEDDING_3_SMALL,
     [OPENAI_TRAINING]: GPT_4O_MINI, // https://platform.openai.com/docs/guides/fine-tuning
     [OPENAI_VOICE]: NOVA,
@@ -252,7 +258,7 @@ const MODELS = {
         ],
     },
     [DEEPSEEK_R1]: {
-        contextWindow: 128000,
+        contextWindow: 128 * 1000,
         maxOutputTokens: 32768,
         requestLimitsRPM: Infinity,
         tokenLimitsTPM: Infinity,
@@ -298,10 +304,29 @@ const MODELS = {
             png, jpeg, gif, webp, pdf,
         ],
     },
+    // https://console.cloud.google.com/vertex-ai/publishers/anthropic/model-garden/claude-3-7-sonnet?authuser=5&inv=1&invt=Abqftg&project=backend-alpha-97077
+    [CLOUD_37_SONNET]: {
+        contextWindow: 200 * 1000,
+        maxOutputTokens: 64 * 1000, // Should be 128 * 1000, but Anthropic SDK limits it to 64 * 1000
+        imageCostTokens: size8k / 750,
+        documentCostTokens: 3000 * 100, // 100 pages: https://docs.anthropic.com/en/docs/build-with-claude/pdf-support
+        maxImagePerPrompt: 5, // https://docs.anthropic.com/en/docs/build-with-claude/vision
+        maxImageSize: 1092, // by pixels
+        maxDocumentPages: 100,
+        maxDocumentFile: 1024 * 1024 * 32, // 32MB
+        requestLimitsRPM: 50,
+        tokenLimitsITPM: 40000,
+        tokenLimitsOTPM: 8000,
+        trainingData: 'Apr 2024', // ?
+        supportedMimeTypes: [
+            png, jpeg, gif, webp, pdf,
+        ],
+    },
 };
 
 MODELS[CLAUDE_35_HAIKU] = MODELS[CLAUDE_35_SONNET];
 MODELS[DEEPSEEK_R1_32B] = MODELS[DEEPSEEK_R1];
+MODELS[DEEPSEEK_R1_70B] = MODELS[DEEPSEEK_R1];
 
 for (const n in MODELS) {
     MODELS[n]['name'] = n;
@@ -340,10 +365,14 @@ const unifyType = (type, name) => {
 const init = async (options) => {
     const provider = unifyProvider(options);
     switch (provider) {
-        case OPENAI:
+        case OPENAI: case AZURE:
             if (options?.apiKey) {
-                const OpenAI = await need('openai');
-                const openai = new OpenAI(options);
+                provider === AZURE && assert(
+                    options?.baseURL, 'Azure api endpoint is required.'
+                );
+                const libOpenAI = await need('openai', { raw: true });
+                const openai = new (options?.endpoint && options?.deployment
+                    ? libOpenAI.AzureOpenAI : libOpenAI.OpenAI)(options);
                 clients[provider] = { client: openai, clientBeta: openai.beta };
             }
             break;
@@ -373,9 +402,20 @@ const init = async (options) => {
             }
             break;
         case CLAUDE:
-            if (options?.apiKey) {
-                const Anthropic = await need('@anthropic-ai/sdk');
-                const anthropic = new Anthropic({ apiKey: options?.apiKey });
+            if (options?.apiKey || (options?.credentials && options?.projectId)) {
+                // https://github.com/anthropics/anthropic-sdk-typescript/tree/main/packages/vertex-sdk
+                const Anthropic = (await need(options?.credentials
+                    ? '@anthropic-ai/vertex-sdk' : '@anthropic-ai/sdk', { raw: true }))[
+                    options?.credentials ? 'AnthropicVertex' : 'Anthropic'
+                ];
+                if (options?.credentials) {
+                    process.env['GOOGLE_APPLICATION_CREDENTIALS'] = options.credentials;
+                    process.env['ANTHROPIC_VERTEX_PROJECT_ID'] = options.projectId;
+                }
+                const anthropic = new Anthropic({
+                    ...options?.apiKey ? { apiKey: options.apiKey } : {},
+                    ...options?.credentials ? { region: options?.region || 'us-east5' } : {},
+                });
                 clients[provider] = { client: anthropic };
             }
             break;
@@ -611,7 +651,9 @@ const promptChatGPT = async (content, options = {}) => {
     // Structured Outputs: https://openai.com/index/introducing-structured-outputs-in-the-api/
     client.baseURL !== OPENAI_BASE_URL
         && options?.attachments?.length && (options.attachments = []);
-    if (options?.model) { } else if (options?.reasoning) {
+    if (options?.model) { } else if (options?.provider === AZURE) {
+        options.model = DEFAULT_MODELS[AZURE];
+    } else if (options?.reasoning) {
         options.model = DEFAULT_MODELS[CHATGPT_REASONING];
     } else {
         options.model = DEFAULT_MODELS[CHATGPT];
@@ -671,6 +713,10 @@ const promptChatGPT = async (content, options = {}) => {
     return await packGptResp(chunk, options);
 };
 
+const promptAzure = async (content, options = {}) => await promptChatGPT(
+    content, { ...options, provider: AZURE }
+);
+
 const promptOllama = async (content, options = {}) => {
     const { client, model } = await getOllamaClient(options);
     // https://github.com/ollama/ollama-js
@@ -939,7 +985,6 @@ const initChat = async (options) => {
         assert(model, `Invalid chat model: '${i}'.`);
         chatConfig.engines[key] = options.engines[i];
         chatConfig.engines[key].model = chatConfig.engines[key].model || model;
-        print(chatConfig.engines[key].model);
         const mxPmpt = MODELS[chatConfig.engines[key].model].maxInputTokens / 2;
         const pmptTokens = await countTokens([buildGeminiHistory(
             chatConfig.systemPrompt, { role: system }
@@ -1006,7 +1051,7 @@ const talk = async (input, options) => {
     const session = await getSession(sessionId, { engine, ...options });
     let [resp, sys, messages, msgBuilder] = [null, [], [], null];
     switch (engine) {
-        case CHATGPT:
+        case CHATGPT: case AZURE:
             sys.push(buildGptMessage(session.systemPrompt, { role: system }));
             msgBuilder = () => {
                 messages = [];
@@ -1088,6 +1133,9 @@ const talk = async (input, options) => {
         case OLLAMA:
             resp = await promptOllama(input, { messages, model, ...options });
             break;
+        case AZURE:
+            resp = await promptAzure(input, { messages, model, ...options });
+            break;
     }
     chat.response = resp.text;
     chat?.request && chat?.response && session.messages.push(chat);
@@ -1208,39 +1256,27 @@ const analyzeSessions = async (sessionIds, options) => {
 };
 
 const PREFERRED_ENGINES = [
-    { client: OPENAI, func: promptChatGPT, multimodal: 1 },
-    { client: GEMINI, func: promptGemini, multimodal: 0 },
+    { client: OPENAI, func: promptChatGPT, multimodal: 0 },
+    { client: GEMINI, func: promptGemini, multimodal: 1 },
     { client: CLAUDE, func: promptClaude, multimodal: 2 },
+    { client: AZURE, func: promptAzure, multimodal: 3 },
     { client: OLLAMA, func: promptOllama, multimodal: 99 },
 ]; // keep gpt first to avoid gemini grounding by default
 
 export default init;
 export {
-    _NEED,
-    ATTACHMENT_TOKEN_COST,
-    CODE_INTERPRETER,
-    DEFAULT_MODELS,
+    ATTACHMENT_TOKEN_COST, CLOUD_37_SONNET, CODE_INTERPRETER, DEEPSEEK_R1,
+    DEEPSEEK_R1_32B, DEEPSEEK_R1_70B, DEFAULT_MODELS,
     EMBEDDING_001,
-    FUNCTION,
-    GEMINI_20_FLASH_THINKING,
-    GEMINI_20_FLASH,
-    GPT_4O_MINI,
-    GPT_4O,
-    GPT_O3_MINI,
-    GPT_O1,
-    DEEPSEEK_R1,
-    DEEPSEEK_R1_32B,
-    MODELS,
+    FUNCTION, GEMINI_20_FLASH, GEMINI_20_FLASH_THINKING, GPT_4O, GPT_4O_MINI, GPT_O1, GPT_O3_MINI, MODELS,
     OPENAI_VOICE,
     RETRIEVAL,
-    TEXT_EMBEDDING_3_SMALL,
-    analyzeSessions,
+    TEXT_EMBEDDING_3_SMALL, _NEED, analyzeSessions,
     buildGptTrainingCase,
     buildGptTrainingCases,
     cancelGptFineTuningJob,
     countTokens,
-    createGeminiEmbedding,
-    createGptFineTuningJob,
+    createGeminiEmbedding, createGptFineTuningJob,
     createOpenAIEmbedding,
     deleteFile,
     distillFile,
@@ -1255,8 +1291,7 @@ export {
     listGptFineTuningJobs,
     listOpenAIModels,
     ogg,
-    prompt,
-    promptChatGPT,
+    prompt, promptAzure, promptChatGPT,
     promptClaude,
     promptGemini,
     promptOllama,
@@ -1266,5 +1301,5 @@ export {
     trimPrompt,
     uploadFile,
     uploadFileForFineTuning,
-    wav,
+    wav
 };

package/lib/manifest.mjs

@@ -1,7 +1,7 @@
 const manifest = {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.18",
+    "version": "1998.2.19",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -25,6 +25,7 @@ const manifest = {
     },
     "devDependencies": {
         "@anthropic-ai/sdk": "^0.36.3",
+        "@anthropic-ai/vertex-sdk": "^0.7.0",
         "@ffmpeg-installer/ffmpeg": "^1.1.0",
         "@ffprobe-installer/ffprobe": "^2.1.2",
         "@google-cloud/speech": "^6.7.0",

package/lib/storage.mjs

@@ -1,16 +1,17 @@
 import {
+    log as _log,
     base64Decode, base64Encode, ensureString, extract, ignoreErrFunc,
-    log as _log, mergeAtoB, need, throwError, trim, voidFunc, which,
+    mergeAtoB, need, throwError, trim, voidFunc, which,
 } from './utilitas.mjs';
 
-import { basename, extname, join } from 'path';
-import { constants as consts, createReadStream, promises as fs, readSync } from 'fs';
-import { defaultAlgorithm, hash } from './encryption.mjs';
-import { deflate as __zip, unzip as __unzip } from 'zlib';
 import { fileTypeFromBuffer } from 'file-type';
+import { constants as consts, createReadStream, promises as fs, readSync } from 'fs';
 import { homedir, tmpdir } from 'os';
+import { basename, extname, join } from 'path';
 import { promisify } from 'util';
 import { v4 as uuidv4 } from 'uuid';
+import { unzip as __unzip, deflate as __zip } from 'zlib';
+import { defaultAlgorithm, hash } from './encryption.mjs';
 
 const _NEED = ['file-type', 'mime-types', '@google-cloud/storage'];
 const errorMessage = 'Invalid file.';
@@ -467,26 +468,15 @@ const deleteOnCloud = async (path, options) => {
 };
 
 export {
-    _NEED,
-    BASE64,
-    BUFFER,
-    DATAURL,
-    FILE,
-    MIME_BINARY,
-    STREAM,
-    analyzeFile,
-    assertPath,
-    blobToBuffer,
-    convert,
-    decodeBase64DataURL,
+    _NEED, analyzeFile,
+    assertPath, BASE64, blobToBuffer, BUFFER, convert, DATAURL, decodeBase64DataURL,
     deleteFileOnCloud,
     deleteOnCloud,
     downloadFileFromCloud,
     downloadFromCloud,
     encodeBase64DataURL,
     exists,
-    existsOnCloud,
-    getConfig,
+    existsOnCloud, FILE, getConfig,
     getConfigFilename,
     getGcUrlByBucket,
     getIdByGs,
@@ -497,18 +487,16 @@ export {
     legalFilename,
     lsOnCloud,
     mapFilename,
-    mergeFile,
-    readFile,
+    mergeFile, MIME_BINARY, readFile,
     readJson,
     sanitizeFilename,
     setConfig,
-    sliceFile,
-    touchPath,
+    sliceFile, STREAM, touchPath,
     tryRm,
     unzip,
     uploadToCloud,
     writeFile,
     writeJson,
     writeTempFile,
-    zip,
+    zip
 };

package/lib/vision.mjs

@@ -3,14 +3,16 @@ import {
 } from './storage.mjs';
 
 import {
-    ensureArray, ensureString, ignoreErrFunc, log as _log, need, throwError,
+    log as _log,
+    ensureArray, ensureString, ignoreErrFunc,
+    need, throwError,
     trim,
 } from './utilitas.mjs';
 
-import { v4 as uuidv4 } from 'uuid';
-import { getApiKeyCredentials } from './encryption.mjs';
 import fs from 'node:fs';
 import path from 'node:path';
+import { v4 as uuidv4 } from 'uuid';
+import { getApiKeyCredentials } from './encryption.mjs';
 
 const _NEED = [
     '@google-cloud/vision', 'office-text-extractor', 'pdfjs-dist',
@@ -268,5 +270,5 @@ export {
     parseOfficeFile,
     read,
     readAll,
-    see,
+    see
 };

package/package.json

@@ -1,7 +1,7 @@
 {
     "name": "utilitas",
     "description": "Just another common utility for JavaScript.",
-    "version": "1998.2.18",
+    "version": "1998.2.19",
     "private": false,
     "homepage": "https://github.com/Leask/utilitas",
     "main": "index.mjs",
@@ -36,6 +36,7 @@
     },
     "devDependencies": {
         "@anthropic-ai/sdk": "^0.36.3",
+        "@anthropic-ai/vertex-sdk": "^0.7.0",
         "@ffmpeg-installer/ffmpeg": "^1.1.0",
         "@ffprobe-installer/ffprobe": "^2.1.2",
         "@google-cloud/speech": "^6.7.0",