ruvector 0.2.16 → 0.2.18

package/dist/core/onnx/pkg/ruvector_onnx_embeddings_wasm_cjs.js

@@ -0,0 +1,127 @@
+/**
+ * CommonJS-compatible WASM loader for Node.js
+ *
+ * This file provides a way to load the WASM module without requiring
+ * the --experimental-wasm-modules flag by manually loading the WASM bytes.
+ *
+ * Usage:
+ *   const wasm = require('./ruvector_onnx_embeddings_wasm_cjs.js');
+ *   await wasm.init(); // or wasm.initSync(wasmBytes)
+ */
+
+const fs = require('fs');
+const path = require('path');
+
+// Re-export everything from the JS bindings
+const bindings = require('./ruvector_onnx_embeddings_wasm_bg.js');
+
+// Track initialization state
+let initialized = false;
+let initPromise = null;
+
+/**
+ * Initialize the WASM module asynchronously
+ * Automatically loads the .wasm file from the same directory
+ */
+async function init(wasmInput) {
+  if (initialized) return bindings;
+
+  if (initPromise) {
+    await initPromise;
+    return bindings;
+  }
+
+  initPromise = (async () => {
+    let wasmBytes;
+
+    if (wasmInput instanceof WebAssembly.Module) {
+      // Already compiled module
+      const instance = await WebAssembly.instantiate(wasmInput, getImports());
+      bindings.__wbg_set_wasm(instance.exports);
+      finishInit();
+      return;
+    } else if (wasmInput instanceof ArrayBuffer || wasmInput instanceof Uint8Array) {
+      // Raw bytes provided
+      wasmBytes = wasmInput;
+    } else if (typeof wasmInput === 'string') {
+      // Path to WASM file
+      wasmBytes = fs.readFileSync(wasmInput);
+    } else {
+      // Auto-detect WASM file location
+      const wasmPath = path.join(__dirname, 'ruvector_onnx_embeddings_wasm_bg.wasm');
+      wasmBytes = fs.readFileSync(wasmPath);
+    }
+
+    const wasmModule = await WebAssembly.compile(wasmBytes);
+    const instance = await WebAssembly.instantiate(wasmModule, getImports());
+
+    bindings.__wbg_set_wasm(instance.exports);
+    finishInit();
+  })();
+
+  await initPromise;
+  return bindings;
+}
+
+/**
+ * Initialize the WASM module synchronously
+ * Requires the WASM bytes to be provided
+ */
+function initSync(wasmBytes) {
+  if (initialized) return bindings;
+
+  if (!wasmBytes) {
+    const wasmPath = path.join(__dirname, 'ruvector_onnx_embeddings_wasm_bg.wasm');
+    wasmBytes = fs.readFileSync(wasmPath);
+  }
+
+  const wasmModule = new WebAssembly.Module(wasmBytes);
+  const instance = new WebAssembly.Instance(wasmModule, getImports());
+
+  bindings.__wbg_set_wasm(instance.exports);
+  finishInit();
+
+  return bindings;
+}
+
+/**
+ * Get the WASM import object
+ */
+function getImports() {
+  return {
+    './ruvector_onnx_embeddings_wasm_bg.js': bindings,
+  };
+}
+
+/**
+ * Finalize initialization
+ */
+function finishInit() {
+  if (typeof bindings.__wbindgen_init_externref_table === 'function') {
+    bindings.__wbindgen_init_externref_table();
+  }
+  initialized = true;
+}
+
+/**
+ * Check if initialized
+ */
+function isInitialized() {
+  return initialized;
+}
+
+// Export init functions and all bindings
+module.exports = {
+  init,
+  initSync,
+  isInitialized,
+  default: init,
+  // Re-export all bindings
+  WasmEmbedder: bindings.WasmEmbedder,
+  WasmEmbedderConfig: bindings.WasmEmbedderConfig,
+  PoolingStrategy: bindings.PoolingStrategy,
+  cosineSimilarity: bindings.cosineSimilarity,
+  normalizeL2: bindings.normalizeL2,
+  simd_available: bindings.simd_available,
+  version: bindings.version,
+};

package/dist/core/onnx-llm.d.ts

@@ -0,0 +1,206 @@
+/**
+ * ONNX LLM Text Generation for RuVector
+ *
+ * Provides real local LLM inference using ONNX Runtime via transformers.js
+ * Supports small models that run efficiently on CPU:
+ * - SmolLM 135M - Smallest, fast (~135MB)
+ * - SmolLM 360M - Better quality (~360MB)
+ * - TinyLlama 1.1B - Best small model quality (~1GB quantized)
+ * - Qwen2.5 0.5B - Good balance (~500MB)
+ *
+ * Features:
+ * - Automatic model downloading and caching
+ * - Quantized INT4/INT8 models for efficiency
+ * - Streaming generation support
+ * - Temperature, top-k, top-p sampling
+ * - KV cache for efficient multi-turn conversations
+ */
+export interface OnnxLLMConfig {
+    /** Model ID (default: 'Xenova/smollm-135m-instruct') */
+    modelId?: string;
+    /** Cache directory for models */
+    cacheDir?: string;
+    /** Use quantized model (default: true) */
+    quantized?: boolean;
+    /** Device: 'cpu' | 'webgpu' (default: 'cpu') */
+    device?: 'cpu' | 'webgpu';
+    /** Maximum context length */
+    maxLength?: number;
+}
+export interface GenerationConfig {
+    /** Maximum new tokens to generate (default: 128) */
+    maxNewTokens?: number;
+    /** Temperature for sampling (default: 0.7) */
+    temperature?: number;
+    /** Top-p nucleus sampling (default: 0.9) */
+    topP?: number;
+    /** Top-k sampling (default: 50) */
+    topK?: number;
+    /** Repetition penalty (default: 1.1) */
+    repetitionPenalty?: number;
+    /** Stop sequences */
+    stopSequences?: string[];
+    /** System prompt for chat models */
+    systemPrompt?: string;
+    /** Enable streaming (callback for each token) */
+    onToken?: (token: string) => void;
+}
+export interface GenerationResult {
+    /** Generated text */
+    text: string;
+    /** Number of tokens generated */
+    tokensGenerated: number;
+    /** Time taken in milliseconds */
+    timeMs: number;
+    /** Tokens per second */
+    tokensPerSecond: number;
+    /** Model used */
+    model: string;
+    /** Whether model was loaded from cache */
+    cached: boolean;
+}
+export declare const AVAILABLE_MODELS: {
+    readonly 'trm-tinystories': {
+        readonly id: "Xenova/TinyStories-33M";
+        readonly name: "TinyStories 33M (TRM)";
+        readonly size: "~65MB";
+        readonly description: "Ultra-tiny model for stories and basic generation";
+        readonly contextLength: 512;
+    };
+    readonly 'trm-gpt2-tiny': {
+        readonly id: "Xenova/gpt2";
+        readonly name: "GPT-2 124M (TRM)";
+        readonly size: "~250MB";
+        readonly description: "Classic GPT-2 tiny for general text";
+        readonly contextLength: 1024;
+    };
+    readonly 'trm-distilgpt2': {
+        readonly id: "Xenova/distilgpt2";
+        readonly name: "DistilGPT-2 (TRM)";
+        readonly size: "~82MB";
+        readonly description: "Distilled GPT-2, fastest general model";
+        readonly contextLength: 1024;
+    };
+    readonly 'smollm-135m': {
+        readonly id: "HuggingFaceTB/SmolLM-135M-Instruct";
+        readonly name: "SmolLM 135M";
+        readonly size: "~135MB";
+        readonly description: "Smallest instruct model, very fast";
+        readonly contextLength: 2048;
+    };
+    readonly 'smollm-360m': {
+        readonly id: "HuggingFaceTB/SmolLM-360M-Instruct";
+        readonly name: "SmolLM 360M";
+        readonly size: "~360MB";
+        readonly description: "Small model, fast, better quality";
+        readonly contextLength: 2048;
+    };
+    readonly 'smollm2-135m': {
+        readonly id: "HuggingFaceTB/SmolLM2-135M-Instruct";
+        readonly name: "SmolLM2 135M";
+        readonly size: "~135MB";
+        readonly description: "Latest SmolLM v2, improved capabilities";
+        readonly contextLength: 2048;
+    };
+    readonly 'smollm2-360m': {
+        readonly id: "HuggingFaceTB/SmolLM2-360M-Instruct";
+        readonly name: "SmolLM2 360M";
+        readonly size: "~360MB";
+        readonly description: "Latest SmolLM v2, better quality";
+        readonly contextLength: 2048;
+    };
+    readonly 'qwen2.5-0.5b': {
+        readonly id: "Qwen/Qwen2.5-0.5B-Instruct";
+        readonly name: "Qwen2.5 0.5B";
+        readonly size: "~300MB quantized";
+        readonly description: "Good balance of speed and quality, multilingual";
+        readonly contextLength: 4096;
+    };
+    readonly tinyllama: {
+        readonly id: "TinyLlama/TinyLlama-1.1B-Chat-v1.0";
+        readonly name: "TinyLlama 1.1B";
+        readonly size: "~600MB quantized";
+        readonly description: "Best small model quality, slower";
+        readonly contextLength: 2048;
+    };
+    readonly 'codegemma-2b': {
+        readonly id: "google/codegemma-2b";
+        readonly name: "CodeGemma 2B";
+        readonly size: "~1GB quantized";
+        readonly description: "Code generation specialist";
+        readonly contextLength: 8192;
+    };
+    readonly 'deepseek-coder-1.3b': {
+        readonly id: "deepseek-ai/deepseek-coder-1.3b-instruct";
+        readonly name: "DeepSeek Coder 1.3B";
+        readonly size: "~700MB quantized";
+        readonly description: "Excellent for code tasks";
+        readonly contextLength: 4096;
+    };
+    readonly 'phi-2': {
+        readonly id: "microsoft/phi-2";
+        readonly name: "Phi-2 2.7B";
+        readonly size: "~1.5GB quantized";
+        readonly description: "High quality small model";
+        readonly contextLength: 2048;
+    };
+    readonly 'phi-3-mini': {
+        readonly id: "microsoft/Phi-3-mini-4k-instruct";
+        readonly name: "Phi-3 Mini";
+        readonly size: "~2GB quantized";
+        readonly description: "Best quality tiny model";
+        readonly contextLength: 4096;
+    };
+};
+export type ModelKey = keyof typeof AVAILABLE_MODELS;
+/**
+ * Check if transformers.js is available
+ */
+export declare function isTransformersAvailable(): Promise<boolean>;
+/**
+ * Initialize the ONNX LLM with specified model
+ */
+export declare function initOnnxLLM(config?: OnnxLLMConfig): Promise<boolean>;
+/**
+ * Generate text using ONNX LLM
+ */
+export declare function generate(prompt: string, config?: GenerationConfig): Promise<GenerationResult>;
+/**
+ * Generate with streaming (token by token)
+ */
+export declare function generateStream(prompt: string, config?: GenerationConfig): Promise<AsyncGenerator<string, GenerationResult, undefined>>;
+/**
+ * Chat completion with conversation history
+ */
+export declare function chat(messages: Array<{
+    role: 'system' | 'user' | 'assistant';
+    content: string;
+}>, config?: GenerationConfig): Promise<GenerationResult>;
+/**
+ * Get model information
+ */
+export declare function getModelInfo(): {
+    model: string | null;
+    ready: boolean;
+    availableModels: typeof AVAILABLE_MODELS;
+};
+/**
+ * Unload the current model to free memory
+ */
+export declare function unload(): Promise<void>;
+export declare class OnnxLLM {
+    private config;
+    private initialized;
+    constructor(config?: OnnxLLMConfig);
+    init(): Promise<boolean>;
+    generate(prompt: string, config?: GenerationConfig): Promise<GenerationResult>;
+    chat(messages: Array<{
+        role: 'system' | 'user' | 'assistant';
+        content: string;
+    }>, config?: GenerationConfig): Promise<GenerationResult>;
+    unload(): Promise<void>;
+    get ready(): boolean;
+    get model(): string | null;
+}
+export default OnnxLLM;
+//# sourceMappingURL=onnx-llm.d.ts.map

package/dist/core/onnx-llm.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"onnx-llm.d.ts","sourceRoot":"","sources":["../../src/core/onnx-llm.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAaH,MAAM,WAAW,aAAa;IAC5B,wDAAwD;IACxD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,iCAAiC;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,0CAA0C;IAC1C,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gDAAgD;IAChD,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,CAAC;IAC1B,6BAA6B;IAC7B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,gBAAgB;IAC/B,oDAAoD;IACpD,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,8CAA8C;IAC9C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,4CAA4C;IAC5C,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,mCAAmC;IACnC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,wCAAwC;IACxC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,qBAAqB;IACrB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,oCAAoC;IACpC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,iDAAiD;IACjD,OAAO,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;CACnC;AAED,MAAM,WAAW,gBAAgB;IAC/B,qBAAqB;IACrB,IAAI,EAAE,MAAM,CAAC;IACb,iCAAiC;IACjC,eAAe,EAAE,MAAM,CAAC;IACxB,iCAAiC;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,wBAAwB;IACxB,eAAe,EAAE,MAAM,CAAC;IACxB,iBAAiB;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,0CAA0C;IAC1C,MAAM,EAAE,OAAO,CAAC;CACjB;AAMD,eAAO,MAAM,gBAAgB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAmHnB,CAAC;AAEX,MAAM,MAAM,QAAQ,GAAG,MAAM,OAAO,gBAAgB,CAAC;AAYrD;;GAEG;AACH,wBAAsB,uBAAuB,IAAI,OAAO,CAAC,OAAO,CAAC,CAOhE;AAED;;GAEG;AACH,wBAAsB,WAAW,CAAC,MAAM,GAAE,aAAkB,GAAG,OAAO,CAAC,OAAO,CAAC,CAqD9E;AAED;;GAEG;AACH,wBAAsB,QAAQ,CAC5B,MAAM,EAAE,MAAM,EACd,MAAM,GAAE,gBAAqB,GAC5B,OAAO,CAAC,gBAAgB,CAAC,CA0C3B;AAED;;GAEG;AACH,wBAAsB,cAAc,CAClC,MAAM,EAAE,MAAM,EACd,MAAM,GAAE,gBAAqB,GAC5B,OAAO,CAAC,cAAc,CAAC,MAAM,EAAE,gBAAgB,EAAE,SAAS,CAAC,CAAC,CA0D9D;AAED;;GAEG;AACH,wBAAsB,IAAI,CACxB,QAAQ,EAAE,KAAK,CAAC;IAAE,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC,EAC3E,MAAM,GAAE,gBAAqB,GAC5B,OAAO,CAAC,gBAAgB,CAAC,CAsB3B;AAED;;GAEG;AACH,wBAAgB,YAAY,IAAI;IAC9B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,KAAK,EAAE,OAAO,CAAC;IACf,eAAe,EAAE,OAAO,gBAAgB,CAAC;CAC1C,CAMA;AAED;;GAEG;AACH,wBAAsB,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAQ5C;AAMD,qBAAa,OAAO;IAClB,OAAO,CAAC,MAAM,CAAgB;IAC9B,OAAO,CAAC,WAAW,CAAS;gBAEhB,MAAM,GAAE,aAAkB;IAIhC,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC;IAMxB,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAAC,gBAAgB,CAAC;IAK9E,IAAI,CACR,QAAQ,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;QAAC,OAAO,EAAE,MAAM,CAAA;KAAE,CAAC,EAC3E,MAAM,CAAC,EAAE,gBAAgB,GACxB,OAAO,CAAC,gBAAgB,CAAC;IAKtB,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;IAK7B,IAAI,KAAK,IAAI,OAAO,CAEnB;IAED,IAAI,KAAK,IAAI,MAAM,GAAG,IAAI,CAEzB;CACF;AAED,eAAe,OAAO,CAAC"}