@dexto/core 1.4.0 → 1.5.0

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

@@ -0,0 +1,343 @@
+"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 registry_exports = {};
+__export(registry_exports, {
+  LOCAL_MODEL_REGISTRY: () => LOCAL_MODEL_REGISTRY,
+  getAllLocalModels: () => getAllLocalModels,
+  getDefaultLocalModelId: () => getDefaultLocalModelId,
+  getLocalModelById: () => getLocalModelById,
+  getLocalModelsByCategory: () => getLocalModelsByCategory,
+  getModelsForRAM: () => getModelsForRAM,
+  getModelsForVRAM: () => getModelsForVRAM,
+  getRecommendedLocalModels: () => getRecommendedLocalModels,
+  searchLocalModels: () => searchLocalModels
+});
+module.exports = __toCommonJS(registry_exports);
+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";
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  LOCAL_MODEL_REGISTRY,
+  getAllLocalModels,
+  getDefaultLocalModelId,
+  getLocalModelById,
+  getLocalModelsByCategory,
+  getModelsForRAM,
+  getModelsForVRAM,
+  getRecommendedLocalModels,
+  searchLocalModels
+});

package/dist/llm/providers/local/registry.d.ts

@@ -0,0 +1,51 @@
+/**
+ * Curated registry of local GGUF models.
+ *
+ * This registry contains vetted models from HuggingFace that are known to work
+ * well with node-llama-cpp. Models are organized by size and use case.
+ *
+ * Model selection criteria:
+ * - Well-maintained quantizations (bartowski, TheBloke, official repos)
+ * - Good performance/size trade-off (Q4_K_M as default)
+ * - Clear licensing for commercial use where possible
+ * - Tested with node-llama-cpp
+ */
+import type { LocalModelInfo } from './types.js';
+/**
+ * Curated list of recommended local models.
+ * Sorted by category and size for easy selection.
+ */
+export declare const LOCAL_MODEL_REGISTRY: LocalModelInfo[];
+/**
+ * Get all models from the registry.
+ */
+export declare function getAllLocalModels(): LocalModelInfo[];
+/**
+ * Get a model by ID.
+ */
+export declare function getLocalModelById(id: string): LocalModelInfo | undefined;
+/**
+ * Get models by category.
+ */
+export declare function getLocalModelsByCategory(category: string): LocalModelInfo[];
+/**
+ * Get recommended models (featured in UI).
+ */
+export declare function getRecommendedLocalModels(): LocalModelInfo[];
+/**
+ * Get models that fit within VRAM constraints.
+ */
+export declare function getModelsForVRAM(vramGB: number): LocalModelInfo[];
+/**
+ * Get models that fit within RAM constraints (CPU inference).
+ */
+export declare function getModelsForRAM(ramGB: number): LocalModelInfo[];
+/**
+ * Search models by name or description.
+ */
+export declare function searchLocalModels(query: string): LocalModelInfo[];
+/**
+ * Get the default model ID for first-time setup.
+ */
+export declare function getDefaultLocalModelId(): string;
+//# sourceMappingURL=registry.d.ts.map

package/dist/llm/providers/local/registry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../../../../src/llm/providers/local/registry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjD;;;GAGG;AACH,eAAO,MAAM,oBAAoB,EAAE,cAAc,EAsQhD,CAAC;AAEF;;GAEG;AACH,wBAAgB,iBAAiB,IAAI,cAAc,EAAE,CAEpD;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,EAAE,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAExE;AAED;;GAEG;AACH,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,GAAG,cAAc,EAAE,CAE3E;AAED;;GAEG;AACH,wBAAgB,yBAAyB,IAAI,cAAc,EAAE,CAE5D;AAED;;GAEG;AACH,wBAAgB,gBAAgB,CAAC,MAAM,EAAE,MAAM,GAAG,cAAc,EAAE,CAEjE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,cAAc,EAAE,CAE/D;AAED;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,MAAM,GAAG,cAAc,EAAE,CAQjE;AAED;;GAEG;AACH,wBAAgB,sBAAsB,IAAI,MAAM,CAI/C"}