@blank-utils/llm 0.5.2 → 0.5.4

package/dist/{chunk-QUWLNQIN.js → chunk-OJCFVVNG.js}

@@ -56,43 +56,86 @@ async function logCapabilities() {
 // src/models.ts
 var DEFAULT_WEBLLM_MODEL = "Phi-3.5-mini-instruct-q4f16_1-MLC";
 var WEBLLM_MODELS = {
-  // === Llama 3.2 Models (Meta) - Excellent quality, reasonable size ===
+  // === Llama 3.2 Models ===
   "llama-3.2-1b": "Llama-3.2-1B-Instruct-q4f16_1-MLC",
   "llama-3.2-3b": "Llama-3.2-3B-Instruct-q4f16_1-MLC",
-  // === Llama 3.1 Models (Meta) - Larger, higher quality ===
+  // === Llama 3.1 & 3 Models ===
   "llama-3.1-8b": "Llama-3.1-8B-Instruct-q4f16_1-MLC",
   "llama-3.1-8b-1k": "Llama-3.1-8B-Instruct-q4f16_1-MLC-1k",
-  // Smaller context for lower memory
-  // === Phi Models (Microsoft) - Great balance of size/quality ===
+  "llama-3.1-70b": "Llama-3.1-70B-Instruct-q3f16_1-MLC",
+  "llama-3-8b": "Llama-3-8B-Instruct-q4f16_1-MLC",
+  "llama-3-8b-1k": "Llama-3-8B-Instruct-q4f16_1-MLC-1k",
+  "llama-3-70b": "Llama-3-70B-Instruct-q3f16_1-MLC",
+  // === Llama 2 Models ===
+  "llama-2-7b": "Llama-2-7b-chat-hf-q4f16_1-MLC",
+  "llama-2-7b-1k": "Llama-2-7b-chat-hf-q4f16_1-MLC-1k",
+  "llama-2-13b": "Llama-2-13b-chat-hf-q4f16_1-MLC",
+  // === Phi Models ===
   "phi-3.5-mini": "Phi-3.5-mini-instruct-q4f16_1-MLC",
   "phi-3.5-mini-1k": "Phi-3.5-mini-instruct-q4f16_1-MLC-1k",
-  // Smaller context for lower memory
   "phi-3.5-vision": "Phi-3.5-vision-instruct-q4f16_1-MLC",
   // Vision model
-  // === Qwen 2.5 Models (Alibaba) - Good multilingual support ===
+  "phi-3-mini-4k": "Phi-3-mini-4k-instruct-q4f16_1-MLC",
+  "phi-3-mini-4k-1k": "Phi-3-mini-4k-instruct-q4f16_1-MLC-1k",
+  "phi-2": "phi-2-q4f16_1-MLC",
+  "phi-1.5": "phi-1_5-q4f16_1-MLC",
+  // === Qwen 3 Models ===
+  "qwen3-0.6b": "Qwen3-0.6B-q4f16_1-MLC",
+  "qwen3-1.7b": "Qwen3-1.7B-q4f16_1-MLC",
+  "qwen3-4b": "Qwen3-4B-q4f16_1-MLC",
+  "qwen3-8b": "Qwen3-8B-q4f16_1-MLC",
+  // === Qwen 2.5 Models ===
   "qwen-2.5-0.5b": "Qwen2.5-0.5B-Instruct-q4f16_1-MLC",
   "qwen-2.5-1.5b": "Qwen2.5-1.5B-Instruct-q4f16_1-MLC",
   "qwen-2.5-3b": "Qwen2.5-3B-Instruct-q4f16_1-MLC",
   "qwen-2.5-7b": "Qwen2.5-7B-Instruct-q4f16_1-MLC",
   "qwen-2.5-coder-0.5b": "Qwen2.5-Coder-0.5B-Instruct-q4f16_1-MLC",
   "qwen-2.5-coder-1.5b": "Qwen2.5-Coder-1.5B-Instruct-q4f16_1-MLC",
-  // === Gemma 2 Models (Google) - Efficient and capable ===
+  "qwen-2.5-coder-3b": "Qwen2.5-Coder-3B-Instruct-q4f16_1-MLC",
+  "qwen-2.5-coder-7b": "Qwen2.5-Coder-7B-Instruct-q4f16_1-MLC",
+  "qwen-2.5-math-1.5b": "Qwen2.5-Math-1.5B-Instruct-q4f16_1-MLC",
+  // === Qwen 2 Models ===
+  "qwen2-math-1.5b": "Qwen2-Math-1.5B-Instruct-q4f16_1-MLC",
+  "qwen2-math-7b": "Qwen2-Math-7B-Instruct-q4f16_1-MLC",
+  // === Gemma 2 Models ===
   "gemma-2-2b": "gemma-2-2b-it-q4f16_1-MLC",
   "gemma-2-2b-1k": "gemma-2-2b-it-q4f16_1-MLC-1k",
-  // Smaller context for lower memory
   "gemma-2-9b": "gemma-2-9b-it-q4f16_1-MLC",
-  // === SmolLM2 Models (HuggingFace) - Ultra lightweight ===
+  "gemma-2-2b-jpn": "gemma-2-2b-jpn-it-q4f16_1-MLC",
+  // === Gemma 1 Models ===
+  "gemma-2b": "gemma-2b-it-q4f16_1-MLC",
+  "gemma-2b-1k": "gemma-2b-it-q4f16_1-MLC-1k",
+  // === SmolLM2 Models ===
   "smollm2-135m": "SmolLM2-135M-Instruct-q0f16-MLC",
   "smollm2-360m": "SmolLM2-360M-Instruct-q4f16_1-MLC",
   "smollm2-1.7b": "SmolLM2-1.7B-Instruct-q4f16_1-MLC",
-  // === Mistral Models - Good general purpose ===
+  // === Mistral & Ministral Models ===
   "mistral-7b": "Mistral-7B-Instruct-v0.3-q4f16_1-MLC",
-  // === DeepSeek R1 Distill Models - Reasoning focused ===
+  "mistral-7b-v0.2": "Mistral-7B-Instruct-v0.2-q4f16_1-MLC",
+  "ministral-3-3b-base": "Ministral-3-3B-Base-2512-q4f16_1-MLC",
+  "ministral-3-3b-reasoning": "Ministral-3-3B-Reasoning-2512-q4f16_1-MLC",
+  "ministral-3-3b-instruct": "Ministral-3-3B-Instruct-2512-BF16-q4f16_1-MLC",
+  // === DeepSeek R1 Distill Models ===
   "deepseek-r1-qwen-7b": "DeepSeek-R1-Distill-Qwen-7B-q4f16_1-MLC",
   "deepseek-r1-llama-8b": "DeepSeek-R1-Distill-Llama-8B-q4f16_1-MLC",
-  // === Hermes Models - Function calling capable ===
+  // === Hermes Models ===
   "hermes-3-llama-3.2-3b": "Hermes-3-Llama-3.2-3B-q4f16_1-MLC",
-  "hermes-3-llama-3.1-8b": "Hermes-3-Llama-3.1-8B-q4f16_1-MLC"
+  "hermes-3-llama-3.1-8b": "Hermes-3-Llama-3.1-8B-q4f16_1-MLC",
+  "hermes-2-theta-llama-3-8b": "Hermes-2-Theta-Llama-3-8B-q4f16_1-MLC",
+  "hermes-2-pro-llama-3-8b": "Hermes-2-Pro-Llama-3-8B-q4f16_1-MLC",
+  "hermes-2-pro-mistral-7b": "Hermes-2-Pro-Mistral-7B-q4f16_1-MLC",
+  "openhermes-2.5-mistral-7b": "OpenHermes-2.5-Mistral-7B-q4f16_1-MLC",
+  "neuralhermes-2.5-mistral-7b": "NeuralHermes-2.5-Mistral-7B-q4f16_1-MLC",
+  // === Other Models ===
+  "tinyllama-1.1b": "TinyLlama-1.1B-Chat-v1.0-q4f16_1-MLC",
+  "tinyllama-1.1b-1k": "TinyLlama-1.1B-Chat-v1.0-q4f16_1-MLC-1k",
+  "tinyllama-1.1b-v0.4": "TinyLlama-1.1B-Chat-v0.4-q4f16_1-MLC",
+  "tinyllama-1.1b-v0.4-1k": "TinyLlama-1.1B-Chat-v0.4-q4f16_1-MLC-1k",
+  "redpajama-3b": "RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC",
+  "redpajama-3b-1k": "RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC-1k",
+  "stablelm-2-zephyr-1.6b": "stablelm-2-zephyr-1_6b-q4f16_1-MLC",
+  "stablelm-2-zephyr-1.6b-1k": "stablelm-2-zephyr-1_6b-q4f16_1-MLC-1k",
+  "wizardmath-7b": "WizardMath-7B-V1.1-q4f16_1-MLC"
 };
 
 // src/backends/webllm.ts
@@ -1022,6 +1065,31 @@ import { jsx as jsx3, jsxs as jsxs3 } from "react/jsx-runtime";
 function cn(...inputs) {
   return twMerge(clsx(inputs));
 }
+async function resizeImageForWebLLM(dataUrl) {
+  return new Promise((resolve, reject) => {
+    const img = new Image();
+    img.onload = () => {
+      const TARGET_W = 1344;
+      const TARGET_H = 1008;
+      const canvas = document.createElement("canvas");
+      canvas.width = TARGET_W;
+      canvas.height = TARGET_H;
+      const ctx = canvas.getContext("2d");
+      if (!ctx) return resolve(dataUrl);
+      ctx.fillStyle = "black";
+      ctx.fillRect(0, 0, TARGET_W, TARGET_H);
+      const scale = Math.min(TARGET_W / img.width, TARGET_H / img.height);
+      const w = img.width * scale;
+      const h = img.height * scale;
+      const x = (TARGET_W - w) / 2;
+      const y = (TARGET_H - h) / 2;
+      ctx.drawImage(img, x, y, w, h);
+      resolve(canvas.toDataURL("image/jpeg", 0.9));
+    };
+    img.onerror = () => resolve(dataUrl);
+    img.src = dataUrl;
+  });
+}
 function sanitizeMarkdownLanguageBlocks(markdown) {
   let cleanMarkdown = markdown;
   cleanMarkdown = cleanMarkdown.replace(/```([a-zA-Z0-9+#-]*)[_|█▋]+[ \t]*(?:\n|$)/gi, "```$1\n");
@@ -1213,21 +1281,27 @@ Additional instructions:
 ${systemPrompt}` : systemPrompt;
       apiMessages.push({ role: "system", content: finalSystemPrompt });
     }
-    currentMessages.forEach((m) => {
+    for (const m of currentMessages) {
       let content = m.content;
       if (m.role === "user" && m.images && m.images.length > 0 && isVisionModel(modelId || "")) {
+        const processedImages = await Promise.all(
+          m.images.map((img) => resizeImageForWebLLM(img.dataUrl))
+        );
         content = [
           { type: "text", text: m.content },
-          ...m.images.map((img) => ({ type: "image_url", image_url: { url: img.dataUrl } }))
+          ...processedImages.map((url) => ({ type: "image_url", image_url: { url } }))
         ];
       }
       apiMessages.push({ role: m.role, content });
-    });
+    }
     let finalUserContent = userContent;
     if (attachedImages.length > 0 && isVisionModel(modelId || "")) {
+      const processedImages = await Promise.all(
+        attachedImages.map((img) => resizeImageForWebLLM(img.dataUrl))
+      );
       finalUserContent = [
         { type: "text", text: userContent },
-        ...attachedImages.map((img) => ({ type: "image_url", image_url: { url: img.dataUrl } }))
+        ...processedImages.map((url) => ({ type: "image_url", image_url: { url } }))
       ];
     }
     apiMessages.push({ role: "user", content: finalUserContent });

package/dist/{index-DJmiDzY3.d.ts → index-3qtk9myB.d.ts}

@@ -20,26 +20,67 @@ declare const WEBLLM_MODELS: {
     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 'llama-3.1-70b': "Llama-3.1-70B-Instruct-q3f16_1-MLC";
+    readonly 'llama-3-8b': "Llama-3-8B-Instruct-q4f16_1-MLC";
+    readonly 'llama-3-8b-1k': "Llama-3-8B-Instruct-q4f16_1-MLC-1k";
+    readonly 'llama-3-70b': "Llama-3-70B-Instruct-q3f16_1-MLC";
+    readonly 'llama-2-7b': "Llama-2-7b-chat-hf-q4f16_1-MLC";
+    readonly 'llama-2-7b-1k': "Llama-2-7b-chat-hf-q4f16_1-MLC-1k";
+    readonly 'llama-2-13b': "Llama-2-13b-chat-hf-q4f16_1-MLC";
     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 'phi-3-mini-4k': "Phi-3-mini-4k-instruct-q4f16_1-MLC";
+    readonly 'phi-3-mini-4k-1k': "Phi-3-mini-4k-instruct-q4f16_1-MLC-1k";
+    readonly 'phi-2': "phi-2-q4f16_1-MLC";
+    readonly 'phi-1.5': "phi-1_5-q4f16_1-MLC";
+    readonly 'qwen3-0.6b': "Qwen3-0.6B-q4f16_1-MLC";
+    readonly 'qwen3-1.7b': "Qwen3-1.7B-q4f16_1-MLC";
+    readonly 'qwen3-4b': "Qwen3-4B-q4f16_1-MLC";
+    readonly 'qwen3-8b': "Qwen3-8B-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-2.5-coder-3b': "Qwen2.5-Coder-3B-Instruct-q4f16_1-MLC";
+    readonly 'qwen-2.5-coder-7b': "Qwen2.5-Coder-7B-Instruct-q4f16_1-MLC";
+    readonly 'qwen-2.5-math-1.5b': "Qwen2.5-Math-1.5B-Instruct-q4f16_1-MLC";
+    readonly 'qwen2-math-1.5b': "Qwen2-Math-1.5B-Instruct-q4f16_1-MLC";
+    readonly 'qwen2-math-7b': "Qwen2-Math-7B-Instruct-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 'gemma-2-2b-jpn': "gemma-2-2b-jpn-it-q4f16_1-MLC";
+    readonly 'gemma-2b': "gemma-2b-it-q4f16_1-MLC";
+    readonly 'gemma-2b-1k': "gemma-2b-it-q4f16_1-MLC-1k";
     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 'mistral-7b-v0.2': "Mistral-7B-Instruct-v0.2-q4f16_1-MLC";
+    readonly 'ministral-3-3b-base': "Ministral-3-3B-Base-2512-q4f16_1-MLC";
+    readonly 'ministral-3-3b-reasoning': "Ministral-3-3B-Reasoning-2512-q4f16_1-MLC";
+    readonly 'ministral-3-3b-instruct': "Ministral-3-3B-Instruct-2512-BF16-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";
+    readonly 'hermes-2-theta-llama-3-8b': "Hermes-2-Theta-Llama-3-8B-q4f16_1-MLC";
+    readonly 'hermes-2-pro-llama-3-8b': "Hermes-2-Pro-Llama-3-8B-q4f16_1-MLC";
+    readonly 'hermes-2-pro-mistral-7b': "Hermes-2-Pro-Mistral-7B-q4f16_1-MLC";
+    readonly 'openhermes-2.5-mistral-7b': "OpenHermes-2.5-Mistral-7B-q4f16_1-MLC";
+    readonly 'neuralhermes-2.5-mistral-7b': "NeuralHermes-2.5-Mistral-7B-q4f16_1-MLC";
+    readonly 'tinyllama-1.1b': "TinyLlama-1.1B-Chat-v1.0-q4f16_1-MLC";
+    readonly 'tinyllama-1.1b-1k': "TinyLlama-1.1B-Chat-v1.0-q4f16_1-MLC-1k";
+    readonly 'tinyllama-1.1b-v0.4': "TinyLlama-1.1B-Chat-v0.4-q4f16_1-MLC";
+    readonly 'tinyllama-1.1b-v0.4-1k': "TinyLlama-1.1B-Chat-v0.4-q4f16_1-MLC-1k";
+    readonly 'redpajama-3b': "RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC";
+    readonly 'redpajama-3b-1k': "RedPajama-INCITE-Chat-3B-v1-q4f16_1-MLC-1k";
+    readonly 'stablelm-2-zephyr-1.6b': "stablelm-2-zephyr-1_6b-q4f16_1-MLC";
+    readonly 'stablelm-2-zephyr-1.6b-1k': "stablelm-2-zephyr-1_6b-q4f16_1-MLC-1k";
+    readonly 'wizardmath-7b': "WizardMath-7B-V1.1-q4f16_1-MLC";
 };
 type WebLLMModelID = keyof typeof WEBLLM_MODELS;
 /**

package/dist/index.d.ts

@@ -1,5 +1,5 @@
-import { B as BrowserCapabilities, L as LLMProvider, a as Backend, b as LoadProgressCallback, C as ChatMessage, G as GenerateOptions, S as StreamCallback, A as AttachOptions } from './index-DJmiDzY3.js';
-export { c as Chat, d as ChatInput, e as ChatInputProps, f as ChatProps, D as DEFAULT_WEBLLM_MODEL, g as Device, h as LLMConfig, i as LLMContextValue, j as LLMLoading, k as LLMLoadingProps, l as LLMProvider, m as LLMProviderProps, n as LLMReady, o as LLMReadyProps, p as LoadProgress, q as LocalLLM, M as MessageRole, Q as Quantization, U as UseChatOptions, r as UseChatReturn, s as UseCompletionOptions, t as UseCompletionReturn, u as UseStreamOptions, v as UseStreamReturn, W as WEBLLM_MODELS, w as createLLM, w as default, x as isWebGPUSupported, y as useChat, z as useCompletion, E as useLLM, F as useStream } from './index-DJmiDzY3.js';
+import { B as BrowserCapabilities, L as LLMProvider, a as Backend, b as LoadProgressCallback, C as ChatMessage, G as GenerateOptions, S as StreamCallback, A as AttachOptions } from './index-3qtk9myB.js';
+export { c as Chat, d as ChatInput, e as ChatInputProps, f as ChatProps, D as DEFAULT_WEBLLM_MODEL, g as Device, h as LLMConfig, i as LLMContextValue, j as LLMLoading, k as LLMLoadingProps, l as LLMProvider, m as LLMProviderProps, n as LLMReady, o as LLMReadyProps, p as LoadProgress, q as LocalLLM, M as MessageRole, Q as Quantization, U as UseChatOptions, r as UseChatReturn, s as UseCompletionOptions, t as UseCompletionReturn, u as UseStreamOptions, v as UseStreamReturn, W as WEBLLM_MODELS, w as createLLM, w as default, x as isWebGPUSupported, y as useChat, z as useCompletion, E as useLLM, F as useStream } from './index-3qtk9myB.js';
 import 'react/jsx-runtime';
 import 'react';
 

package/dist/index.js

@@ -22,7 +22,7 @@ import {
   useCompletion,
   useLLM,
   useStream
-} from "./chunk-QUWLNQIN.js";
+} from "./chunk-OJCFVVNG.js";
 export {
   Chat,
   ChatInput,

package/dist/react/index.d.ts

@@ -1,3 +1,3 @@
-export { c as Chat, H as ChatApp, I as ChatAppProps, d as ChatInput, e as ChatInputProps, f as ChatProps, J as ImageAttachment, i as LLMContextValue, j as LLMLoading, k as LLMLoadingProps, l as LLMProvider, m as LLMProviderProps, n as LLMReady, o as LLMReadyProps, U as UseChatOptions, r as UseChatReturn, s as UseCompletionOptions, t as UseCompletionReturn, u as UseStreamOptions, v as UseStreamReturn, y as useChat, z as useCompletion, E as useLLM, F as useStream } from '../index-DJmiDzY3.js';
+export { c as Chat, H as ChatApp, I as ChatAppProps, d as ChatInput, e as ChatInputProps, f as ChatProps, J as ImageAttachment, i as LLMContextValue, j as LLMLoading, k as LLMLoadingProps, l as LLMProvider, m as LLMProviderProps, n as LLMReady, o as LLMReadyProps, U as UseChatOptions, r as UseChatReturn, s as UseCompletionOptions, t as UseCompletionReturn, u as UseStreamOptions, v as UseStreamReturn, y as useChat, z as useCompletion, E as useLLM, F as useStream } from '../index-3qtk9myB.js';
 import 'react/jsx-runtime';
 import 'react';

package/dist/react/index.js

@@ -9,7 +9,7 @@ import {
   useCompletion,
   useLLM,
   useStream
-} from "../chunk-QUWLNQIN.js";
+} from "../chunk-OJCFVVNG.js";
 export {
   Chat,
   ChatApp,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@blank-utils/llm",
-  "version": "0.5.2",
+  "version": "0.5.4",
   "description": "Run LLMs directly in your browser with WebGPU acceleration. Supports React hooks and eager background loading.",
   "type": "module",
   "main": "./dist/index.js",