@dexto/core 1.3.0 → 1.5.0

package/dist/llm/providers/local/registry.js

@@ -0,0 +1,312 @@
+import "../../../chunk-PTJYTZNU.js";
+const LOCAL_MODEL_REGISTRY = [
+  // ============================================
+  // RECOMMENDED: Best balance of quality and size
+  // ============================================
+  {
+    id: "llama-3.3-8b-q4",
+    name: "Llama 3.3 8B Instruct",
+    description: "Meta's latest 8B model. Excellent general-purpose performance with 128K context.",
+    huggingfaceId: "bartowski/Llama-3.3-8B-Instruct-GGUF",
+    filename: "Llama-3.3-8B-Instruct-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 502e7,
+    // ~5GB
+    contextLength: 131072,
+    categories: ["general", "coding"],
+    minVRAM: 6,
+    minRAM: 8,
+    recommended: true,
+    author: "Meta",
+    license: "llama3.3",
+    supportsTools: true
+  },
+  {
+    id: "qwen-2.5-coder-7b-q4",
+    name: "Qwen 2.5 Coder 7B Instruct",
+    description: "Alibaba's coding-focused model. Excellent for code generation and review.",
+    huggingfaceId: "Qwen/Qwen2.5-Coder-7B-Instruct-GGUF",
+    filename: "qwen2.5-coder-7b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 468e7,
+    // ~4.7GB
+    contextLength: 131072,
+    categories: ["coding"],
+    minVRAM: 6,
+    minRAM: 8,
+    recommended: true,
+    author: "Alibaba",
+    license: "apache-2.0",
+    supportsTools: true
+  },
+  // ============================================
+  // SMALL: Fast models for quick tasks (< 4GB)
+  // ============================================
+  {
+    id: "phi-3.5-mini-q4",
+    name: "Phi 3.5 Mini Instruct",
+    description: "Microsoft's compact model. Great for simple tasks with minimal resources.",
+    huggingfaceId: "bartowski/Phi-3.5-mini-instruct-GGUF",
+    filename: "Phi-3.5-mini-instruct-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 239e7,
+    // ~2.4GB
+    contextLength: 131072,
+    categories: ["small", "general"],
+    minVRAM: 4,
+    minRAM: 4,
+    recommended: true,
+    author: "Microsoft",
+    license: "mit"
+  },
+  {
+    id: "qwen-2.5-3b-q4",
+    name: "Qwen 2.5 3B Instruct",
+    description: "Compact but capable. Good for basic chat and simple tasks.",
+    huggingfaceId: "Qwen/Qwen2.5-3B-Instruct-GGUF",
+    filename: "qwen2.5-3b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 205e7,
+    // ~2GB
+    contextLength: 32768,
+    categories: ["small", "general"],
+    minVRAM: 3,
+    minRAM: 4,
+    author: "Alibaba",
+    license: "apache-2.0"
+  },
+  {
+    id: "gemma-2-2b-q4",
+    name: "Gemma 2 2B Instruct",
+    description: "Google's efficient small model. Good balance of speed and capability.",
+    huggingfaceId: "bartowski/gemma-2-2b-it-GGUF",
+    filename: "gemma-2-2b-it-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 179e7,
+    // ~1.8GB
+    contextLength: 8192,
+    categories: ["small", "general"],
+    minVRAM: 3,
+    minRAM: 4,
+    author: "Google",
+    license: "gemma"
+  },
+  // ============================================
+  // CODING: Optimized for code generation
+  // ============================================
+  {
+    id: "qwen-2.5-coder-14b-q4",
+    name: "Qwen 2.5 Coder 14B Instruct",
+    description: "Larger coding model for complex tasks. Better code understanding.",
+    huggingfaceId: "Qwen/Qwen2.5-Coder-14B-Instruct-GGUF",
+    filename: "qwen2.5-coder-14b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 89e8,
+    // ~8.9GB
+    contextLength: 131072,
+    categories: ["coding"],
+    minVRAM: 10,
+    minRAM: 12,
+    author: "Alibaba",
+    license: "apache-2.0",
+    supportsTools: true
+  },
+  {
+    id: "deepseek-coder-v2-lite-q4",
+    name: "DeepSeek Coder V2 Lite",
+    description: "DeepSeek's efficient coding model. Great for code completion.",
+    huggingfaceId: "bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF",
+    filename: "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 92e8,
+    // ~9.2GB
+    contextLength: 131072,
+    categories: ["coding"],
+    minVRAM: 12,
+    minRAM: 16,
+    author: "DeepSeek",
+    license: "deepseek"
+  },
+  {
+    id: "codestral-22b-q4",
+    name: "Codestral 22B",
+    description: "Mistral's dedicated coding model. Supports 80+ languages.",
+    huggingfaceId: "bartowski/Codestral-22B-v0.1-GGUF",
+    filename: "Codestral-22B-v0.1-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 135e8,
+    // ~13.5GB
+    contextLength: 32768,
+    categories: ["coding"],
+    minVRAM: 16,
+    minRAM: 20,
+    author: "Mistral AI",
+    license: "mnpl"
+  },
+  // ============================================
+  // GENERAL: Versatile all-purpose models
+  // ============================================
+  {
+    id: "mistral-7b-q4",
+    name: "Mistral 7B Instruct v0.3",
+    description: "Mistral's efficient 7B model. Good balance of speed and quality.",
+    huggingfaceId: "bartowski/Mistral-7B-Instruct-v0.3-GGUF",
+    filename: "Mistral-7B-Instruct-v0.3-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 437e7,
+    // ~4.4GB
+    contextLength: 32768,
+    categories: ["general"],
+    minVRAM: 6,
+    minRAM: 8,
+    author: "Mistral AI",
+    license: "apache-2.0",
+    supportsTools: true
+  },
+  {
+    id: "gemma-2-9b-q4",
+    name: "Gemma 2 9B Instruct",
+    description: "Google's capable 9B model. Strong reasoning and instruction following.",
+    huggingfaceId: "bartowski/gemma-2-9b-it-GGUF",
+    filename: "gemma-2-9b-it-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 576e7,
+    // ~5.8GB
+    contextLength: 8192,
+    categories: ["general"],
+    minVRAM: 8,
+    minRAM: 10,
+    author: "Google",
+    license: "gemma"
+  },
+  {
+    id: "llama-3.1-8b-q4",
+    name: "Llama 3.1 8B Instruct",
+    description: "Meta's Llama 3.1. Solid general-purpose performance.",
+    huggingfaceId: "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF",
+    filename: "Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 492e7,
+    // ~4.9GB
+    contextLength: 131072,
+    categories: ["general"],
+    minVRAM: 6,
+    minRAM: 8,
+    author: "Meta",
+    license: "llama3.1",
+    supportsTools: true
+  },
+  // ============================================
+  // REASONING: Strong reasoning capabilities
+  // ============================================
+  {
+    id: "qwen-2.5-14b-q4",
+    name: "Qwen 2.5 14B Instruct",
+    description: "Alibaba's mid-size model. Strong reasoning and long context.",
+    huggingfaceId: "Qwen/Qwen2.5-14B-Instruct-GGUF",
+    filename: "qwen2.5-14b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 87e8,
+    // ~8.7GB
+    contextLength: 131072,
+    categories: ["reasoning", "general"],
+    minVRAM: 10,
+    minRAM: 12,
+    author: "Alibaba",
+    license: "apache-2.0",
+    supportsTools: true
+  },
+  {
+    id: "qwen-2.5-32b-q4",
+    name: "Qwen 2.5 32B Instruct",
+    description: "Alibaba's large model. Excellent reasoning and complex tasks.",
+    huggingfaceId: "Qwen/Qwen2.5-32B-Instruct-GGUF",
+    filename: "qwen2.5-32b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 193e8,
+    // ~19.3GB
+    contextLength: 131072,
+    categories: ["reasoning", "general"],
+    minVRAM: 24,
+    minRAM: 32,
+    author: "Alibaba",
+    license: "apache-2.0",
+    supportsTools: true
+  },
+  // ============================================
+  // VISION: Multimodal models with image support
+  // ============================================
+  {
+    id: "llava-v1.6-mistral-7b-q4",
+    name: "LLaVA v1.6 Mistral 7B",
+    description: "Vision-language model. Can understand and discuss images.",
+    huggingfaceId: "cjpais/llava-v1.6-mistral-7b-gguf",
+    filename: "llava-v1.6-mistral-7b.Q4_K_M.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 45e8,
+    // ~4.5GB
+    contextLength: 4096,
+    categories: ["vision", "general"],
+    minVRAM: 8,
+    minRAM: 10,
+    author: "Microsoft/LLaVA",
+    license: "llama2",
+    supportsVision: true
+  },
+  {
+    id: "qwen-2-vl-7b-q4",
+    name: "Qwen2 VL 7B Instruct",
+    description: "Alibaba's vision-language model. High-quality image understanding.",
+    huggingfaceId: "Qwen/Qwen2-VL-7B-Instruct-GGUF",
+    filename: "qwen2-vl-7b-instruct-q4_k_m.gguf",
+    quantization: "Q4_K_M",
+    sizeBytes: 51e8,
+    // ~5.1GB
+    contextLength: 32768,
+    categories: ["vision", "general"],
+    minVRAM: 8,
+    minRAM: 10,
+    author: "Alibaba",
+    license: "apache-2.0",
+    supportsVision: true
+  }
+];
+function getAllLocalModels() {
+  return [...LOCAL_MODEL_REGISTRY];
+}
+function getLocalModelById(id) {
+  return LOCAL_MODEL_REGISTRY.find((m) => m.id === id);
+}
+function getLocalModelsByCategory(category) {
+  return LOCAL_MODEL_REGISTRY.filter((m) => m.categories.includes(category));
+}
+function getRecommendedLocalModels() {
+  return LOCAL_MODEL_REGISTRY.filter((m) => m.recommended);
+}
+function getModelsForVRAM(vramGB) {
+  return LOCAL_MODEL_REGISTRY.filter((m) => !m.minVRAM || m.minVRAM <= vramGB);
+}
+function getModelsForRAM(ramGB) {
+  return LOCAL_MODEL_REGISTRY.filter((m) => !m.minRAM || m.minRAM <= ramGB);
+}
+function searchLocalModels(query) {
+  const q = query.toLowerCase();
+  return LOCAL_MODEL_REGISTRY.filter(
+    (m) => m.id.toLowerCase().includes(q) || m.name.toLowerCase().includes(q) || m.description.toLowerCase().includes(q)
+  );
+}
+function getDefaultLocalModelId() {
+  const recommended = getRecommendedLocalModels();
+  return recommended[0]?.id ?? "llama-3.3-8b-q4";
+}
+export {
+  LOCAL_MODEL_REGISTRY,
+  getAllLocalModels,
+  getDefaultLocalModelId,
+  getLocalModelById,
+  getLocalModelsByCategory,
+  getModelsForRAM,
+  getModelsForVRAM,
+  getRecommendedLocalModels,
+  searchLocalModels
+};

package/dist/llm/providers/local/schemas.cjs

@@ -0,0 +1,169 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var schemas_exports = {};
+__export(schemas_exports, {
+  GPUBackendSchema: () => GPUBackendSchema,
+  GPUInfoSchema: () => GPUInfoSchema,
+  InstalledModelSchema: () => InstalledModelSchema,
+  LocalLLMConfigSchema: () => LocalLLMConfigSchema,
+  LocalModelCategorySchema: () => LocalModelCategorySchema,
+  LocalModelInfoSchema: () => LocalModelInfoSchema,
+  ModelDownloadOptionsSchema: () => ModelDownloadOptionsSchema,
+  ModelDownloadProgressSchema: () => ModelDownloadProgressSchema,
+  ModelDownloadStatusSchema: () => ModelDownloadStatusSchema,
+  ModelSourceSchema: () => ModelSourceSchema,
+  ModelStateSchema: () => ModelStateSchema,
+  OllamaModelInfoSchema: () => OllamaModelInfoSchema,
+  OllamaStatusSchema: () => OllamaStatusSchema,
+  QuantizationTypeSchema: () => QuantizationTypeSchema
+});
+module.exports = __toCommonJS(schemas_exports);
+var import_zod = require("zod");
+const GPUBackendSchema = import_zod.z.enum(["metal", "cuda", "vulkan", "cpu"]);
+const QuantizationTypeSchema = import_zod.z.enum([
+  "Q2_K",
+  "Q3_K_S",
+  "Q3_K_M",
+  "Q3_K_L",
+  "Q4_0",
+  "Q4_K_S",
+  "Q4_K_M",
+  "Q5_0",
+  "Q5_K_S",
+  "Q5_K_M",
+  "Q6_K",
+  "Q8_0",
+  "F16",
+  "F32"
+]);
+const LocalModelCategorySchema = import_zod.z.enum([
+  "general",
+  "coding",
+  "reasoning",
+  "small",
+  "vision"
+]);
+const ModelSourceSchema = import_zod.z.enum(["huggingface", "ollama"]);
+const ModelDownloadStatusSchema = import_zod.z.enum([
+  "pending",
+  "downloading",
+  "verifying",
+  "complete",
+  "error"
+]);
+const LocalModelInfoSchema = import_zod.z.object({
+  id: import_zod.z.string().min(1).describe("Unique model identifier"),
+  name: import_zod.z.string().min(1).describe("Human-readable display name"),
+  description: import_zod.z.string().describe("Short description of model capabilities"),
+  huggingfaceId: import_zod.z.string().min(1).describe("HuggingFace repository ID"),
+  filename: import_zod.z.string().min(1).describe("GGUF filename to download"),
+  quantization: QuantizationTypeSchema.describe("Quantization level"),
+  sizeBytes: import_zod.z.number().int().positive().describe("File size in bytes"),
+  contextLength: import_zod.z.number().int().positive().describe("Maximum context window"),
+  categories: import_zod.z.array(LocalModelCategorySchema).describe("Model categories"),
+  minVRAM: import_zod.z.number().positive().optional().describe("Minimum VRAM in GB"),
+  minRAM: import_zod.z.number().positive().optional().describe("Minimum RAM in GB"),
+  recommended: import_zod.z.boolean().optional().describe("Whether model is featured"),
+  author: import_zod.z.string().optional().describe("Model author/organization"),
+  license: import_zod.z.string().optional().describe("License type"),
+  supportsVision: import_zod.z.boolean().optional().describe("Whether model supports images"),
+  supportsTools: import_zod.z.boolean().optional().describe("Whether model supports function calling")
+}).strict();
+const ModelDownloadProgressSchema = import_zod.z.object({
+  modelId: import_zod.z.string().min(1),
+  status: ModelDownloadStatusSchema,
+  bytesDownloaded: import_zod.z.number().int().nonnegative(),
+  totalBytes: import_zod.z.number().int().nonnegative(),
+  percentage: import_zod.z.number().min(0).max(100),
+  speed: import_zod.z.number().nonnegative().optional(),
+  eta: import_zod.z.number().nonnegative().optional(),
+  error: import_zod.z.string().optional()
+}).strict();
+const GPUInfoSchema = import_zod.z.object({
+  backend: GPUBackendSchema,
+  available: import_zod.z.boolean(),
+  deviceName: import_zod.z.string().optional(),
+  vramMB: import_zod.z.number().int().nonnegative().optional(),
+  driverVersion: import_zod.z.string().optional()
+}).strict();
+const LocalLLMConfigSchema = import_zod.z.object({
+  provider: import_zod.z.enum(["local", "ollama"]),
+  model: import_zod.z.string().min(1).describe("Model ID or GGUF path"),
+  gpuLayers: import_zod.z.number().int().optional().describe("GPU layers (-1=auto, 0=CPU)"),
+  contextSize: import_zod.z.number().int().positive().optional().describe("Override context size"),
+  threads: import_zod.z.number().int().positive().optional().describe("CPU threads"),
+  batchSize: import_zod.z.number().int().positive().optional().describe("Inference batch size"),
+  modelPath: import_zod.z.string().optional().describe("Resolved path to model file")
+}).strict();
+const InstalledModelSchema = import_zod.z.object({
+  id: import_zod.z.string().min(1),
+  filePath: import_zod.z.string().min(1),
+  sizeBytes: import_zod.z.number().int().positive(),
+  downloadedAt: import_zod.z.string().datetime(),
+  lastUsedAt: import_zod.z.string().datetime().optional(),
+  sha256: import_zod.z.string().optional(),
+  source: ModelSourceSchema
+}).strict();
+const ModelStateSchema = import_zod.z.object({
+  version: import_zod.z.string().default("1.0"),
+  installed: import_zod.z.record(import_zod.z.string(), InstalledModelSchema).default({}),
+  activeModelId: import_zod.z.string().optional(),
+  downloadQueue: import_zod.z.array(import_zod.z.string()).default([])
+}).strict();
+const ModelDownloadOptionsSchema = import_zod.z.object({
+  modelId: import_zod.z.string().min(1),
+  outputDir: import_zod.z.string().optional(),
+  showProgress: import_zod.z.boolean().default(true),
+  hfToken: import_zod.z.string().optional()
+}).strict();
+const OllamaModelInfoSchema = import_zod.z.object({
+  name: import_zod.z.string().min(1),
+  size: import_zod.z.number().int().nonnegative(),
+  digest: import_zod.z.string(),
+  modifiedAt: import_zod.z.string(),
+  details: import_zod.z.object({
+    family: import_zod.z.string().optional(),
+    parameterSize: import_zod.z.string().optional(),
+    quantizationLevel: import_zod.z.string().optional()
+  }).optional()
+}).strict();
+const OllamaStatusSchema = import_zod.z.object({
+  running: import_zod.z.boolean(),
+  url: import_zod.z.string().url(),
+  version: import_zod.z.string().optional(),
+  models: import_zod.z.array(OllamaModelInfoSchema).optional(),
+  error: import_zod.z.string().optional()
+}).strict();
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  GPUBackendSchema,
+  GPUInfoSchema,
+  InstalledModelSchema,
+  LocalLLMConfigSchema,
+  LocalModelCategorySchema,
+  LocalModelInfoSchema,
+  ModelDownloadOptionsSchema,
+  ModelDownloadProgressSchema,
+  ModelDownloadStatusSchema,
+  ModelSourceSchema,
+  ModelStateSchema,
+  OllamaModelInfoSchema,
+  OllamaStatusSchema,
+  QuantizationTypeSchema
+});