@blank-utils/llm 0.2.2 → 0.2.3

package/README.md

@@ -1,25 +1,42 @@
-# @blank-utils/llm
+<p align="center">
+  <strong>@blank-utils/llm</strong>
+</p>
 
-> Run LLMs directly in your browser with WebGPU acceleration
+<p align="center">
+  Run LLMs directly in your browser — zero server, zero API keys.
+</p>
 
-A simple, ergonomic library for running local LLMs in the browser. Features:
+<p align="center">
+  <a href="https://www.npmjs.com/package/@blank-utils/llm"><img src="https://img.shields.io/npm/v/@blank-utils/llm?style=flat-square&color=0ea5e9" alt="npm"></a>
+  <a href="https://github.com/kiritocode1/local-llm/blob/main/LICENSE"><img src="https://img.shields.io/npm/l/@blank-utils/llm?style=flat-square" alt="license"></a>
+  <img src="https://img.shields.io/badge/react-%E2%89%A518-61dafb?style=flat-square" alt="react">
+  <img src="https://img.shields.io/badge/webgpu-supported-brightgreen?style=flat-square" alt="webgpu">
+</p>
 
-- 🚀 **WebGPU acceleration** via WebLLM (with WASM fallback)
-- ⚛️ **React hooks** with eager background loading
-- 📝 **Streaming support** with real-time token output
-- 🔄 **Message queueing** - users can type while models load
-- 📦 **Zero config** - works out of the box
+---
+
+## Features
+
+- 🚀 **WebGPU acceleration** via [WebLLM](https://github.com/mlc-ai/web-llm) — falls back to WASM through [Transformers.js](https://github.com/huggingface/transformers.js)
+- ⚛️ **React hooks** — `useChat`, `useStream`, `useCompletion` with eager background loading
+- 🔤 **Type-safe model selection** — full autocomplete for 30+ supported models across both backends
+- 📝 **Streaming support** — real-time token output with abort control
+- 🔄 **Message queueing** — users can type while models download; messages are processed once ready
+- 🧩 **Vanilla JS friendly** — works outside React with DOM helpers and a simple `createLLM()` API
+- 📦 **Zero config** — auto-detects WebGPU/WASM and picks the best backend
 
 ## Installation
 
 ```bash
 pnpm add @blank-utils/llm
 # or
-npm install @blank-utils/llm
-# or
 bun add @blank-utils/llm
 ```
 
+> React is an **optional** peer dependency. The core API works without it.
+
+---
+
 ## Quick Start
 
 ### React (Recommended)
@@ -45,7 +62,10 @@ function Chat() {
     isGenerating,
     isPending,
     streamingText,
-  } = useChat();
+  } = useChat({
+    systemPrompt: "You are a helpful assistant.",
+    queueWhileLoading: true,
+  });
 
   return (
     <div>
@@ -53,7 +73,7 @@ function Chat() {
 
       {messages.map((m, i) => (
         <div key={i}>
-          {m.role}: {m.content}
+          <strong>{m.role}:</strong> {m.content}
         </div>
       ))}
 
@@ -83,148 +103,396 @@ const llm = await createLLM({
 });
 
 // Streaming
-await llm.stream("Tell me a joke", (token) => {
-  process.stdout.write(token);
+await llm.stream("Tell me a joke", (token, fullText) => {
+  document.getElementById("output")!.textContent = fullText;
 });
 
-// Or non-streaming
+// Non-streaming
 const response = await llm.chat("Hello!");
+console.log(response);
+```
+
+### Attach to DOM Elements
+
+```typescript
+import { createLLM } from "@blank-utils/llm";
+
+const llm = await createLLM({ model: "smollm2-360m" });
+
+// Wire up an input + output with one call
+const cleanup = llm.attachToInput("#prompt-input", "#response-output", {
+  triggerOnEnter: true,
+  clearOnSend: true,
+});
+```
+
+---
+
+## Architecture
+
+```
+@blank-utils/llm
+├── src/
+│   ├── index.ts          # Main entry — re-exports everything
+│   ├── core.ts           # createLLM() factory, LocalLLM interface
+│   ├── models.ts         # Centralized model registry (single source of truth)
+│   ├── types.ts          # All TypeScript interfaces & types
+│   ├── detect.ts         # WebGPU / WASM capability detection
+│   ├── helpers.ts        # DOM utilities (attachToElements, createChatUI, etc.)
+│   ├── backends/
+│   │   ├── webllm.ts     # WebLLM backend (WebGPU)
+│   │   └── transformers.ts # Transformers.js backend (WASM / WebGPU)
+│   └── react/
+│       └── index.tsx     # React context, provider, hooks, components
+└── dist/                 # Built output (ESM)
+```
+
+### Dual Backend System
+
+|                   | **WebLLM**                | **Transformers.js** |
+| ----------------- | ------------------------- | ------------------- |
+| **Engine**        | MLC / TVM compiled models | ONNX Runtime        |
+| **Device**        | WebGPU only               | WebGPU or WASM      |
+| **Performance**   | Best (GPU-native)         | Good (CPU fallback) |
+| **Model source**  | MLC prebuilt cache        | HuggingFace Hub     |
+| **Auto-detected** | ✅ when WebGPU present    | ✅ fallback         |
+
+The library auto-selects the best backend via `detectCapabilities()`. You can also force a backend:
+
+```tsx
+<LLMProvider model="llama-3.2-3b" backend="webllm" />
 ```
 
-## React Hooks
+---
 
-### `LLMProvider`
+## React API
 
-Wrap your app with the provider to enable LLM functionality:
+### `<LLMProvider>`
+
+Wrap your app to enable LLM functionality. All hooks must be used inside this provider.
 
 ```tsx
 <LLMProvider
-  model="qwen-2.5-0.5b" // Model to load
+  model="qwen-2.5-0.5b" // Model alias or full ID
   backend="auto" // 'webllm' | 'transformers' | 'auto'
-  autoLoad={true} // Start loading immediately
-  onProgress={(p) => {}} // Loading progress callback
-  onLoad={(llm) => {}} // Called when ready
-  onError={(err) => {}} // Error callback
+  autoLoad={true} // Start loading on mount
+  device="auto" // 'webgpu' | 'wasm' | 'auto'
+  quantization="q4" // 'q4' | 'q8' | 'fp16' | 'fp32'
+  systemPrompt="..." // Default system prompt
+  onProgress={(p) => {}} // Loading progress
+  onLoad={(llm) => {}} // Called when model is ready
+  onError={(err) => {}} // Error handler
 >
   {children}
 </LLMProvider>
 ```
 
+> **Tip:** Use `key={modelId}` on `<LLMProvider>` to force a full re-mount when switching models dynamically.
+
 ### `useLLM()`
 
-Access the LLM instance and loading state:
+Access the raw LLM instance and loading state:
 
 ```tsx
 const {
-  llm, // LLM instance (null while loading)
-  isLoading, // Model is downloading
-  isReady, // Model ready for inference
-  loadProgress, // { progress: number, status: string }
-  error, // Error if failed
-  modelId, // Current model ID
-  backend, // 'webllm' | 'transformers'
-  reload, // Reload the model
-  unload, // Unload and free memory
+  llm, // LocalLLM | null
+  isLoading, // boolean — model is downloading
+  isReady, // boolean — model ready for inference
+  loadProgress, // { progress: number, status: string } | null
+  error, // Error | null
+  modelId, // string | null — current model ID
+  backend, // 'webllm' | 'transformers' | null
+  reload, // () => Promise<void>
+  unload, // () => Promise<void>
 } = useLLM();
 ```
 
-### `useChat(options)`
+### `useChat(options?)`
 
-Full chat conversation management with **eager loading** support:
+Full chat conversation management with **eager loading** — users can send messages while the model downloads. Messages are queued and processed automatically once the model is ready.
 
 ```tsx
 const {
   messages, // ChatMessage[]
-  input, // Current input value
-  setInput, // Update input
-  send, // Send message (queues if loading!)
-  isGenerating, // Currently generating response
-  isPending, // Message queued, waiting for model
-  streamingText, // Current streaming output
-  stop, // Stop generation
-  clear, // Clear conversation
-  append, // Add message without generating
-  reload, // Regenerate last response
+  input, // string — controlled input value
+  setInput, // (value: string) => void
+  send, // (content?: string) => Promise<string>
+  isGenerating, // boolean
+  isPending, // boolean — message queued, waiting for model
+  streamingText, // string — current partial response
+  stop, // () => void
+  clear, // () => void
+  append, // (message: ChatMessage) => void
+  reload, // () => Promise<string> — regenerate last response
 } = useChat({
   systemPrompt: "You are a helpful assistant.",
-  queueWhileLoading: true, // Queue messages while model loads
+  queueWhileLoading: true, // default: true
+  initialMessages: [],
+  generateOptions: { temperature: 0.7, maxTokens: 512 },
+  onStart: () => {},
   onToken: (token, fullText) => {},
   onFinish: (response) => {},
+  onError: (error) => {},
 });
 ```
 
-### `useStream(options)`
+### `useStream(options?)`
 
-Simple streaming generation:
+Simple streaming generation without chat history management:
 
 ```tsx
-const { text, isStreaming, stream, stop, clear } = useStream();
+const { text, isStreaming, stream, stop, clear } = useStream({
+  onToken: (token, fullText) => {},
+  onFinish: (response) => {},
+  onError: (error) => {},
+  generateOptions: { temperature: 0.7 },
+});
 
 await stream("Tell me a story");
+// or with message array:
+await stream([{ role: "user", content: "Tell me a story" }]);
 ```
 
-### `useCompletion(options)`
+### `useCompletion(options?)`
 
-Non-streaming completion:
+Non-streaming, single-shot completion:
 
 ```tsx
-const { completion, isLoading, complete, clear } = useCompletion();
+const { completion, isLoading, complete, clear } = useCompletion({
+  generateOptions: { maxTokens: 256 },
+});
 
-await complete("Summarize this text");
+const result = await complete("Summarize this text");
 ```
 
-## Utility Components
+### `<LLMLoading>` / `<LLMReady>`
 
-### `<LLMLoading>`
-
-Shows content only while loading:
+Conditional rendering components:
 
 ```tsx
-<LLMLoading>
-  <p>Loading model...</p>
+<LLMLoading className="loading-state">
+  <p>Downloading model...</p>
 </LLMLoading>
+
+<LLMReady fallback={<Spinner />}>
+  <ChatInterface />
+</LLMReady>
 ```
 
-### `<LLMReady>`
+---
 
-Shows content only when ready:
+## Vanilla JS API
 
-```tsx
-<LLMReady fallback={<Loading />}>
-  <ChatInterface />
-</LLMReady>
+### `createLLM(config?)`
+
+Factory function that auto-detects capabilities, picks a backend, loads a model, and returns a ready-to-use `LocalLLM` instance:
+
+```typescript
+import { createLLM } from "@blank-utils/llm";
+
+const llm = await createLLM({
+  model: "phi-3.5-mini",
+  backend: "auto",
+  systemPrompt: "You are a helpful assistant.",
+  onLoadProgress: (p) => console.log(`${p.status}: ${p.progress}%`),
+});
+
+// Chat (non-streaming)
+const answer = await llm.chat("What is 2+2?");
+
+// Stream
+await llm.stream("Write a poem", (token, fullText) => {
+  process.stdout.write(token);
+});
+
+// Attach to DOM
+const cleanup = llm.attachToInput("#input", "#output");
+
+// Free resources
+await llm.unload();
 ```
 
-## Available Models
+### DOM Helpers
+
+```typescript
+import {
+  createOutputStreamer,
+  attachToElements,
+  createChatUI,
+  createLoadingIndicator,
+} from "@blank-utils/llm";
+
+// Auto-scroll streaming output into an element
+const streamer = createOutputStreamer("#output", { scrollToBottom: true });
+
+// Create a full chat UI in one call
+const { input, output, sendButton, cleanup } = createChatUI("#container");
+
+// Progress indicator
+const loading = createLoadingIndicator("#loading-container");
+loading.show();
+loading.setProgress(50, "Downloading weights...");
+loading.hide();
+```
+
+### Capability Detection
+
+```typescript
+import {
+  detectCapabilities,
+  logCapabilities,
+  isWebGPUSupported,
+} from "@blank-utils/llm";
+
+const caps = await detectCapabilities();
+// { webgpu: true, wasm: true, recommendedBackend: 'webllm', recommendedDevice: 'webgpu' }
+
+await logCapabilities(); // Pretty-prints to console
 
-### Transformers.js Backend (ONNX)
+if (await isWebGPUSupported()) {
+  console.log("WebGPU is available!");
+}
+```
+
+---
+
+## Available Models
 
-| Alias                 | Model                  | Size   |
-| --------------------- | ---------------------- | ------ |
-| `qwen-2.5-0.5b`       | Qwen 2.5 0.5B Instruct | ~350MB |
-| `qwen-2.5-1.5b`       | Qwen 2.5 1.5B Instruct | ~900MB |
-| `qwen-2.5-coder-0.5b` | Qwen 2.5 Coder 0.5B    | ~350MB |
-| `smollm2-135m`        | SmolLM2 135M           | ~100MB |
-| `smollm2-360m`        | SmolLM2 360M           | ~250MB |
-| `tinyllama`           | TinyLlama 1.1B         | ~700MB |
-| `phi-3-mini`          | Phi-3 Mini 4K          | ~2.3GB |
+All models are defined in `src/models.ts` and exported as `WEBLLM_MODELS` and `TRANSFORMERS_MODELS`.  
+You can use either the **alias** (short name) or the **full model ID** when specifying a model.
 
 ### WebLLM Backend (WebGPU)
 
-| Alias           | Model         | Size   |
-| --------------- | ------------- | ------ |
-| `phi-3-mini`    | Phi-3 Mini 4K | ~2.3GB |
-| `llama-3.2-1b`  | Llama 3.2 1B  | ~1GB   |
-| `llama-3.2-3b`  | Llama 3.2 3B  | ~2GB   |
-| `gemma-2-2b`    | Gemma 2 2B    | ~1.5GB |
-| `qwen-2.5-0.5b` | Qwen 2.5 0.5B | ~350MB |
-| `qwen-2.5-1.5b` | Qwen 2.5 1.5B | ~900MB |
+| Alias                   | Model                        | Notes                       |
+| ----------------------- | ---------------------------- | --------------------------- |
+| `llama-3.2-1b`          | Llama 3.2 1B Instruct        | Compact, great quality      |
+| `llama-3.2-3b`          | Llama 3.2 3B Instruct        | Balanced                    |
+| `llama-3.1-8b`          | Llama 3.1 8B Instruct        | High quality                |
+| `llama-3.1-8b-1k`       | Llama 3.1 8B (1K ctx)        | Lower memory                |
+| `phi-3.5-mini`          | Phi 3.5 Mini Instruct        | **Default** — great balance |
+| `phi-3.5-mini-1k`       | Phi 3.5 Mini (1K ctx)        | Lower memory                |
+| `phi-3.5-vision`        | Phi 3.5 Vision               | Vision model                |
+| `qwen-2.5-0.5b`         | Qwen 2.5 0.5B                | Tiny, fast                  |
+| `qwen-2.5-1.5b`         | Qwen 2.5 1.5B                | Small                       |
+| `qwen-2.5-3b`           | Qwen 2.5 3B                  | Medium                      |
+| `qwen-2.5-7b`           | Qwen 2.5 7B                  | Large                       |
+| `qwen-2.5-coder-0.5b`   | Qwen 2.5 Coder 0.5B          | Code-focused                |
+| `qwen-2.5-coder-1.5b`   | Qwen 2.5 Coder 1.5B          | Code-focused                |
+| `qwen-3-0.6b`           | Qwen 3 0.6B                  | Latest gen                  |
+| `qwen-3-1.7b`           | Qwen 3 1.7B                  | Latest gen                  |
+| `qwen-3-4b`             | Qwen 3 4B                    | Latest gen                  |
+| `qwen-3-8b`             | Qwen 3 8B                    | Latest gen                  |
+| `gemma-2-2b`            | Gemma 2 2B                   | Google, efficient           |
+| `gemma-2-2b-1k`         | Gemma 2 2B (1K ctx)          | Lower memory                |
+| `gemma-2-9b`            | Gemma 2 9B                   | Large                       |
+| `smollm2-135m`          | SmolLM2 135M                 | Ultra lightweight           |
+| `smollm2-360m`          | SmolLM2 360M                 | Lightweight                 |
+| `smollm2-1.7b`          | SmolLM2 1.7B                 | Small                       |
+| `mistral-7b`            | Mistral 7B v0.3              | General purpose             |
+| `deepseek-r1-qwen-7b`   | DeepSeek R1 Distill Qwen 7B  | Reasoning                   |
+| `deepseek-r1-llama-8b`  | DeepSeek R1 Distill Llama 8B | Reasoning                   |
+| `hermes-3-llama-3.2-3b` | Hermes 3 Llama 3.2 3B        | Function calling            |
+| `hermes-3-llama-3.1-8b` | Hermes 3 Llama 3.1 8B        | Function calling            |
+
+### Transformers.js Backend (CPU / WASM)
+
+| Alias                 | HuggingFace Model ID                         | Notes        |
+| --------------------- | -------------------------------------------- | ------------ |
+| `qwen-2.5-0.5b`       | `onnx-community/Qwen2.5-0.5B-Instruct`       | **Default**  |
+| `qwen-2.5-1.5b`       | `onnx-community/Qwen2.5-1.5B-Instruct`       | Good quality |
+| `qwen-2.5-coder-0.5b` | `onnx-community/Qwen2.5-Coder-0.5B-Instruct` | Code         |
+| `qwen-2.5-coder-1.5b` | `onnx-community/Qwen2.5-Coder-1.5B-Instruct` | Code         |
+| `qwen-3-0.6b`         | `onnx-community/Qwen3-0.6B-ONNX`             | Latest gen   |
+| `smollm2-135m`        | `HuggingFaceTB/SmolLM2-135M-Instruct`        | Ultra fast   |
+| `smollm2-360m`        | `HuggingFaceTB/SmolLM2-360M-Instruct`        | Fast         |
+| `smollm2-1.7b`        | `HuggingFaceTB/SmolLM2-1.7B-Instruct`        | Good         |
+| `phi-3-mini`          | `Xenova/Phi-3-mini-4k-instruct`              | Strong       |
+| `tinyllama`           | `Xenova/TinyLlama-1.1B-Chat-v1.0`            | Very fast    |
+
+### Type-Safe Model Selection
+
+The `model` prop accepts any key from `WEBLLM_MODELS` or `TRANSFORMERS_MODELS` with full autocomplete, while still allowing arbitrary strings for custom models:
+
+```typescript
+import type {
+  SupportedModel,
+  WebLLMModelID,
+  TransformersModelID,
+} from "@blank-utils/llm";
+
+// Full autocomplete for known models
+const model: SupportedModel = "qwen-2.5-0.5b"; // ✅ autocomplete
+
+// Custom model IDs still work
+const custom: SupportedModel = "my-org/custom-model-onnx"; // ✅ no error
+
+// Import the model maps for programmatic use
+import { WEBLLM_MODELS, TRANSFORMERS_MODELS } from "@blank-utils/llm";
+
+Object.keys(WEBLLM_MODELS); // all WebLLM aliases
+Object.keys(TRANSFORMERS_MODELS); // all Transformers.js aliases
+```
+
+---
+
+## Build & Development
+
+```bash
+# Install dependencies
+bun install
+
+# Build (clean → bundle → assets → types)
+bun run build
+
+# Type-check only
+bun run typecheck
+
+# Run demo page
+bun run demo
+
+# Run tests
+bun test
+```
+
+### Build Pipeline
+
+| Script        | What it does                                                                                                       |
+| ------------- | ------------------------------------------------------------------------------------------------------------------ |
+| `clean`       | Removes `dist/`                                                                                                    |
+| `build:js`    | Bundles `src/index.ts` → `dist/index.js` and `src/react/index.tsx` → `dist/react/index.js` (ESM, externals: react) |
+| `postbuild`   | Copies WASM + ONNX runtime assets into `dist/` and `dist/react/`                                                   |
+| `build:types` | Generates `.d.ts` declaration files via `tsc`                                                                      |
+| `build`       | Runs all of the above in sequence                                                                                  |
+
+### Package Exports
+
+```jsonc
+{
+  ".": {
+    "types": "./dist/index.d.ts",
+    "import": "./dist/index.js",
+  },
+  "./react": {
+    "types": "./dist/react/index.d.ts",
+    "import": "./dist/react/index.js",
+  },
+}
+```
+
+---
 
 ## Browser Requirements
 
-- **WebGPU** (Chrome 113+, Edge 113+) for best performance
-- Falls back to **WebAssembly** for older browsers
+| Feature               | Minimum                    | Notes                          |
+| --------------------- | -------------------------- | ------------------------------ |
+| **WebGPU**            | Chrome 113+, Edge 113+     | Required for WebLLM backend    |
+| **WebAssembly**       | All modern browsers        | Fallback for Transformers.js   |
+| **SharedArrayBuffer** | Requires COOP/COEP headers | Needed for multi-threaded WASM |
+
+The library automatically detects capabilities and picks the best backend. No manual configuration needed.
+
+---
 
 ## License
 
-MIT License
+MIT © [blank](https://github.com/kiritocode1)

package/dist/backends/transformers.d.ts

@@ -7,26 +7,9 @@ import type { LLMProvider, ChatMessage, GenerateOptions, StreamCallback, LoadPro
  * Default model for Transformers.js backend
  * Using Qwen2.5 0.5B as it's well-tested with ONNX
  */
-export declare const DEFAULT_TRANSFORMERS_MODEL = "onnx-community/Qwen2.5-0.5B-Instruct";
-/**
- * Transformers.js compatible models (must have ONNX weights)
- * These are specifically converted for browser use via transformers.js
- *
- * @see https://huggingface.co/onnx-community for more models
- */
-export declare const TRANSFORMERS_MODELS: {
-    readonly 'qwen-2.5-0.5b': "onnx-community/Qwen2.5-0.5B-Instruct";
-    readonly 'qwen-2.5-1.5b': "onnx-community/Qwen2.5-1.5B-Instruct";
-    readonly 'qwen-2.5-coder-0.5b': "onnx-community/Qwen2.5-Coder-0.5B-Instruct";
-    readonly 'qwen-2.5-coder-1.5b': "onnx-community/Qwen2.5-Coder-1.5B-Instruct";
-    readonly 'qwen-3-0.6b': "onnx-community/Qwen3-0.6B-ONNX";
-    readonly 'smollm2-135m': "HuggingFaceTB/SmolLM2-135M-Instruct";
-    readonly 'smollm2-360m': "HuggingFaceTB/SmolLM2-360M-Instruct";
-    readonly 'smollm2-1.7b': "HuggingFaceTB/SmolLM2-1.7B-Instruct";
-    readonly 'phi-3-mini': "Xenova/Phi-3-mini-4k-instruct";
-    readonly tinyllama: "Xenova/TinyLlama-1.1B-Chat-v1.0";
-};
-export type TransformersModelAlias = keyof typeof TRANSFORMERS_MODELS;
+import { DEFAULT_TRANSFORMERS_MODEL, TRANSFORMERS_MODELS, type TransformersModelID } from '../models';
+export { DEFAULT_TRANSFORMERS_MODEL, TRANSFORMERS_MODELS };
+export type TransformersModelAlias = TransformersModelID;
 /**
  * Model size estimates for UI display
  */

package/dist/backends/transformers.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"transformers.d.ts","sourceRoot":"","sources":["../../src/backends/transformers.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,WAAW,EACX,eAAe,EACf,cAAc,EACd,oBAAoB,EAEpB,OAAO,EACP,MAAM,EACN,YAAY,EACb,MAAM,UAAU,CAAC;AAKlB;;;GAGG;AACH,eAAO,MAAM,0BAA0B,yCAAyC,CAAC;AAEjF;;;;;GAKG;AACH,eAAO,MAAM,mBAAmB;;;;;;;;;;;CAoBtB,CAAC;AAEX,MAAM,MAAM,sBAAsB,GAAG,MAAM,OAAO,mBAAmB,CAAC;AAEtE;;GAEG;AACH,eAAO,MAAM,wBAAwB,EAAE,MAAM,CAAC,sBAAsB,EAAE,MAAM,CAW3E,CAAC;AAkGF;;GAEG;AACH,MAAM,WAAW,0BAA0B;IACzC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,YAAY,CAAC;CAC7B;AAED;;GAEG;AACH,qBAAa,oBAAqB,YAAW,WAAW;IACtD,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAkB;IAE3C,OAAO,CAAC,QAAQ,CAAuC;IACvD,OAAO,CAAC,YAAY,CAAuB;IAC3C,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,YAAY,CAAe;gBAEvB,MAAM,GAAE,0BAA+B;IAKnD,IAAI,OAAO,IAAI,OAAO,CAErB;IAED,IAAI,OAAO,IAAI,MAAM,GAAG,IAAI,CAE3B;IAEK,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,oBAAoB,GAAG,OAAO,CAAC,IAAI,CAAC;IAkDvE,IAAI,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC;IAoBzE,MAAM,CACV,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC;IAgCZ,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;CAI9B;AAED;;GAEG;AACH,wBAAgB,0BAA0B,CAAC,MAAM,CAAC,EAAE,0BAA0B,GAAG,oBAAoB,CAEpG"}
+{"version":3,"file":"transformers.d.ts","sourceRoot":"","sources":["../../src/backends/transformers.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,WAAW,EACX,eAAe,EACf,cAAc,EACd,oBAAoB,EAEpB,OAAO,EACP,MAAM,EACN,YAAY,EACb,MAAM,UAAU,CAAC;AAKlB;;;GAGG;AACH,OAAO,EAAE,0BAA0B,EAAE,mBAAmB,EAAE,KAAK,mBAAmB,EAAE,MAAM,WAAW,CAAC;AACtG,OAAO,EAAE,0BAA0B,EAAE,mBAAmB,EAAE,CAAC;AAE3D,MAAM,MAAM,sBAAsB,GAAG,mBAAmB,CAAC;AAEzD;;GAEG;AACH,eAAO,MAAM,wBAAwB,EAAE,MAAM,CAAC,sBAAsB,EAAE,MAAM,CAW3E,CAAC;AAkGF;;GAEG;AACH,MAAM,WAAW,0BAA0B;IACzC,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,YAAY,CAAC;CAC7B;AAED;;GAEG;AACH,qBAAa,oBAAqB,YAAW,WAAW;IACtD,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAkB;IAE3C,OAAO,CAAC,QAAQ,CAAuC;IACvD,OAAO,CAAC,YAAY,CAAuB;IAC3C,OAAO,CAAC,MAAM,CAAS;IACvB,OAAO,CAAC,YAAY,CAAe;gBAEvB,MAAM,GAAE,0BAA+B;IAKnD,IAAI,OAAO,IAAI,OAAO,CAErB;IAED,IAAI,OAAO,IAAI,MAAM,GAAG,IAAI,CAE3B;IAEK,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,oBAAoB,GAAG,OAAO,CAAC,IAAI,CAAC;IAkDvE,IAAI,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC;IAoBzE,MAAM,CACV,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC;IAgCZ,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;CAI9B;AAED;;GAEG;AACH,wBAAgB,0BAA0B,CAAC,MAAM,CAAC,EAAE,0BAA0B,GAAG,oBAAoB,CAEpG"}

package/dist/backends/webllm.d.ts

@@ -7,44 +7,9 @@ import type { LLMProvider, ChatMessage, GenerateOptions, StreamCallback, LoadPro
  * Default model for WebLLM backend
  * Using Phi 3.5 Mini as it's well-tested and reasonably sized
  */
-export declare const DEFAULT_WEBLLM_MODEL = "Phi-3.5-mini-instruct-q4f16_1-MLC";
-/**
- * Popular WebLLM model options with correct MLC model IDs
- * These IDs must match exactly what's in web-llm's prebuiltAppConfig
- *
- * @see https://github.com/mlc-ai/web-llm/blob/main/src/config.ts
- */
-export declare const WEBLLM_MODELS: {
-    readonly 'llama-3.2-1b': "Llama-3.2-1B-Instruct-q4f16_1-MLC";
-    readonly 'llama-3.2-3b': "Llama-3.2-3B-Instruct-q4f16_1-MLC";
-    readonly 'llama-3.1-8b': "Llama-3.1-8B-Instruct-q4f16_1-MLC";
-    readonly 'llama-3.1-8b-1k': "Llama-3.1-8B-Instruct-q4f16_1-MLC-1k";
-    readonly 'phi-3.5-mini': "Phi-3.5-mini-instruct-q4f16_1-MLC";
-    readonly 'phi-3.5-mini-1k': "Phi-3.5-mini-instruct-q4f16_1-MLC-1k";
-    readonly 'phi-3.5-vision': "Phi-3.5-vision-instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-0.5b': "Qwen2.5-0.5B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-1.5b': "Qwen2.5-1.5B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-3b': "Qwen2.5-3B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-7b': "Qwen2.5-7B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-coder-0.5b': "Qwen2.5-Coder-0.5B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-2.5-coder-1.5b': "Qwen2.5-Coder-1.5B-Instruct-q4f16_1-MLC";
-    readonly 'qwen-3-0.6b': "Qwen3-0.6B-q4f16_1-MLC";
-    readonly 'qwen-3-1.7b': "Qwen3-1.7B-q4f16_1-MLC";
-    readonly 'qwen-3-4b': "Qwen3-4B-q4f16_1-MLC";
-    readonly 'qwen-3-8b': "Qwen3-8B-q4f16_1-MLC";
-    readonly 'gemma-2-2b': "gemma-2-2b-it-q4f16_1-MLC";
-    readonly 'gemma-2-2b-1k': "gemma-2-2b-it-q4f16_1-MLC-1k";
-    readonly 'gemma-2-9b': "gemma-2-9b-it-q4f16_1-MLC";
-    readonly 'smollm2-135m': "SmolLM2-135M-Instruct-q0f16-MLC";
-    readonly 'smollm2-360m': "SmolLM2-360M-Instruct-q4f16_1-MLC";
-    readonly 'smollm2-1.7b': "SmolLM2-1.7B-Instruct-q4f16_1-MLC";
-    readonly 'mistral-7b': "Mistral-7B-Instruct-v0.3-q4f16_1-MLC";
-    readonly 'deepseek-r1-qwen-7b': "DeepSeek-R1-Distill-Qwen-7B-q4f16_1-MLC";
-    readonly 'deepseek-r1-llama-8b': "DeepSeek-R1-Distill-Llama-8B-q4f16_1-MLC";
-    readonly 'hermes-3-llama-3.2-3b': "Hermes-3-Llama-3.2-3B-q4f16_1-MLC";
-    readonly 'hermes-3-llama-3.1-8b': "Hermes-3-Llama-3.1-8B-q4f16_1-MLC";
-};
-export type WebLLMModelAlias = keyof typeof WEBLLM_MODELS;
+import { DEFAULT_WEBLLM_MODEL, WEBLLM_MODELS, type WebLLMModelID } from '../models';
+export { DEFAULT_WEBLLM_MODEL, WEBLLM_MODELS };
+export type WebLLMModelAlias = WebLLMModelID;
 /**
  * Model size estimates for UI display
  */

package/dist/backends/webllm.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"webllm.d.ts","sourceRoot":"","sources":["../../src/backends/webllm.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,WAAW,EACX,eAAe,EACf,cAAc,EACd,oBAAoB,EAEpB,OAAO,EACR,MAAM,UAAU,CAAC;AAMlB;;;GAGG;AACH,eAAO,MAAM,oBAAoB,sCAAsC,CAAC;AAExE;;;;;GAKG;AACH,eAAO,MAAM,aAAa;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAgDhB,CAAC;AAEX,MAAM,MAAM,gBAAgB,GAAG,MAAM,OAAO,aAAa,CAAC;AAE1D;;GAEG;AACH,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,gBAAgB,EAAE,MAAM,CA6B/D,CAAC;AAYF;;GAEG;AACH,qBAAa,cAAe,YAAW,WAAW;IAChD,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAY;IAErC,OAAO,CAAC,MAAM,CAA0B;IACxC,OAAO,CAAC,YAAY,CAAuB;IAE3C,IAAI,OAAO,IAAI,OAAO,CAErB;IAED,IAAI,OAAO,IAAI,MAAM,GAAG,IAAI,CAE3B;IAEK,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,oBAAoB,GAAG,OAAO,CAAC,IAAI,CAAC;IAwBvE,IAAI,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC;IAmBzE,MAAM,CACV,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC;IA8BZ,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;CAO9B;AAED;;GAEG;AACH,wBAAgB,oBAAoB,IAAI,cAAc,CAErD"}
+{"version":3,"file":"webllm.d.ts","sourceRoot":"","sources":["../../src/backends/webllm.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EACV,WAAW,EACX,WAAW,EACX,eAAe,EACf,cAAc,EACd,oBAAoB,EAEpB,OAAO,EACR,MAAM,UAAU,CAAC;AAMlB;;;GAGG;AACH,OAAO,EAAE,oBAAoB,EAAE,aAAa,EAAE,KAAK,aAAa,EAAE,MAAM,WAAW,CAAC;AACpF,OAAO,EAAE,oBAAoB,EAAE,aAAa,EAAE,CAAC;AAE/C,MAAM,MAAM,gBAAgB,GAAG,aAAa,CAAC;AAE7C;;GAEG;AACH,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,gBAAgB,EAAE,MAAM,CA6B/D,CAAC;AAYF;;GAEG;AACH,qBAAa,cAAe,YAAW,WAAW;IAChD,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAY;IAErC,OAAO,CAAC,MAAM,CAA0B;IACxC,OAAO,CAAC,YAAY,CAAuB;IAE3C,IAAI,OAAO,IAAI,OAAO,CAErB;IAED,IAAI,OAAO,IAAI,MAAM,GAAG,IAAI,CAE3B;IAEK,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,oBAAoB,GAAG,OAAO,CAAC,IAAI,CAAC;IAwBvE,IAAI,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,MAAM,CAAC;IAmBzE,MAAM,CACV,QAAQ,EAAE,WAAW,EAAE,EACvB,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC;IA8BZ,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC;CAO9B;AAED;;GAEG;AACH,wBAAgB,oBAAoB,IAAI,cAAc,CAErD"}

package/dist/core.d.ts

@@ -4,7 +4,7 @@
  *
  * @module local-llm/core
  */
-export type { Backend, Device, Quantization, LLMConfig, MessageRole, ChatMessage, StreamCallback, LoadProgress, LoadProgressCallback, GenerateOptions, LLMProvider as LLMProviderInterface, AttachOptions, BrowserCapabilities, } from './types';
+export type { Backend, Device, Quantization, LLMConfig, MessageRole, ChatMessage, StreamCallback, LoadProgress, LoadProgressCallback, GenerateOptions, LLMProvider as LLMProviderInterface, AttachOptions, BrowserCapabilities, SupportedModel, } from './types';
 export { checkWebGPU, checkWasm, detectCapabilities, logCapabilities } from './detect';
 export { WebLLMProvider, createWebLLMProvider, DEFAULT_WEBLLM_MODEL, WEBLLM_MODELS, } from './backends/webllm';
 export { TransformersProvider, createTransformersProvider, DEFAULT_TRANSFORMERS_MODEL, TRANSFORMERS_MODELS, } from './backends/transformers';

package/dist/core.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"core.d.ts","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,YAAY,EACV,OAAO,EACP,MAAM,EACN,YAAY,EACZ,SAAS,EACT,WAAW,EACX,WAAW,EACX,cAAc,EACd,YAAY,EACZ,oBAAoB,EACpB,eAAe,EACf,WAAW,IAAI,oBAAoB,EACnC,aAAa,EACb,mBAAmB,GACpB,MAAM,SAAS,CAAC;AAGjB,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,UAAU,CAAC;AAGvF,OAAO,EACL,cAAc,EACd,oBAAoB,EACpB,oBAAoB,EACpB,aAAa,GACd,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,oBAAoB,EACpB,0BAA0B,EAC1B,0BAA0B,EAC1B,mBAAmB,GACpB,MAAM,yBAAyB,CAAC;AAGjC,OAAO,EACL,oBAAoB,EACpB,gBAAgB,EAChB,YAAY,EACZ,sBAAsB,GACvB,MAAM,WAAW,CAAC;AAGnB,OAAO,KAAK,EACV,SAAS,EACT,WAAW,EACX,eAAe,EACf,cAAc,EACd,aAAa,EACd,MAAM,SAAS,CAAC;AAOjB;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;IAE1B;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAEhC;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,QAAQ,GAAG,cAAc,CAAC;IAE5C;;OAEG;IACH,IAAI,CACF,QAAQ,EAAE,WAAW,EAAE,GAAG,MAAM,EAChC,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC,CAAC;IAEnB;;OAEG;IACH,MAAM,CACJ,QAAQ,EAAE,WAAW,EAAE,GAAG,MAAM,EAChC,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC,CAAC;IAEnB;;OAEG;IACH,aAAa,CACX,aAAa,EAAE,MAAM,GAAG,gBAAgB,GAAG,mBAAmB,EAC9D,cAAc,EAAE,MAAM,GAAG,WAAW,EACpC,OAAO,CAAC,EAAE,aAAa,GACtB,MAAM,IAAI,CAAC;IAEd;;OAEG;IACH,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACzB;AAwBD;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAsB,SAAS,CAAC,MAAM,GAAE,SAAc,GAAG,OAAO,CAAC,QAAQ,CAAC,CAyFzE;AAED;;GAEG;AACH,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,OAAO,CAAC,CAG1D"}
+{"version":3,"file":"core.d.ts","sourceRoot":"","sources":["../src/core.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,YAAY,EACV,OAAO,EACP,MAAM,EACN,YAAY,EACZ,SAAS,EACT,WAAW,EACX,WAAW,EACX,cAAc,EACd,YAAY,EACZ,oBAAoB,EACpB,eAAe,EACf,WAAW,IAAI,oBAAoB,EACnC,aAAa,EACb,mBAAmB,EACnB,cAAc,GACf,MAAM,SAAS,CAAC;AAGjB,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,kBAAkB,EAAE,eAAe,EAAE,MAAM,UAAU,CAAC;AAGvF,OAAO,EACL,cAAc,EACd,oBAAoB,EACpB,oBAAoB,EACpB,aAAa,GACd,MAAM,mBAAmB,CAAC;AAE3B,OAAO,EACL,oBAAoB,EACpB,0BAA0B,EAC1B,0BAA0B,EAC1B,mBAAmB,GACpB,MAAM,yBAAyB,CAAC;AAGjC,OAAO,EACL,oBAAoB,EACpB,gBAAgB,EAChB,YAAY,EACZ,sBAAsB,GACvB,MAAM,WAAW,CAAC;AAGnB,OAAO,KAAK,EACV,SAAS,EACT,WAAW,EACX,eAAe,EACf,cAAc,EACd,aAAa,EACd,MAAM,SAAS,CAAC;AAOjB;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,OAAO,CAAC;IAE1B;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IAEhC;;OAEG;IACH,QAAQ,CAAC,OAAO,EAAE,QAAQ,GAAG,cAAc,CAAC;IAE5C;;OAEG;IACH,IAAI,CACF,QAAQ,EAAE,WAAW,EAAE,GAAG,MAAM,EAChC,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC,CAAC;IAEnB;;OAEG;IACH,MAAM,CACJ,QAAQ,EAAE,WAAW,EAAE,GAAG,MAAM,EAChC,OAAO,EAAE,cAAc,EACvB,OAAO,CAAC,EAAE,eAAe,GACxB,OAAO,CAAC,MAAM,CAAC,CAAC;IAEnB;;OAEG;IACH,aAAa,CACX,aAAa,EAAE,MAAM,GAAG,gBAAgB,GAAG,mBAAmB,EAC9D,cAAc,EAAE,MAAM,GAAG,WAAW,EACpC,OAAO,CAAC,EAAE,aAAa,GACtB,MAAM,IAAI,CAAC;IAEd;;OAEG;IACH,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACzB;AAwBD;;;;;;;;;;;;;;;;;;;GAmBG;AACH,wBAAsB,SAAS,CAAC,MAAM,GAAE,SAAc,GAAG,OAAO,CAAC,QAAQ,CAAC,CAyFzE;AAED;;GAEG;AACH,wBAAsB,iBAAiB,IAAI,OAAO,CAAC,OAAO,CAAC,CAG1D"}