web-llm-runner 0.1.9 → 0.1.12

package/lib/index.js

@@ -986,7 +986,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/Llama-3.2-1B-Instruct-q4f32_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 1128.82,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -998,7 +998,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/Llama-3.2-1B-Instruct-q4f16_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 879.04,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1024,7 +1024,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/Llama-3.2-1B-Instruct-q0f16-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 2573.13,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1036,7 +1036,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/Llama-3.2-3B-Instruct-q4f32_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 2951.51,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1048,7 +1048,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/Llama-3.2-3B-Instruct-q4f16_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 2263.69,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1483,7 +1483,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-360M-Instruct-q0f16-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 871.99,
-            low_resource_required: true,
+            low_resource_required: false,
             required_features: ["shader-f16"],
             overrides: {
                 context_window_size: 4096,
@@ -1496,7 +1496,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-360M-Instruct-q0f32-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 1743.99,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1508,7 +1508,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-360M-Instruct-q4f16_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 376.06,
-            low_resource_required: true,
+            low_resource_required: false,
             required_features: ["shader-f16"],
             overrides: {
                 context_window_size: 4096,
@@ -1521,7 +1521,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-360M-Instruct-q4f32_1-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 579.61,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1533,7 +1533,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-135M-Instruct-q0f16-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 359.69,
-            low_resource_required: true,
+            low_resource_required: false,
             required_features: ["shader-f16"],
             overrides: {
                 context_window_size: 4096,
@@ -1546,7 +1546,7 @@ const prebuiltAppConfig = {
                 modelVersion +
                 "/SmolLM2-135M-Instruct-q0f32-ctx4k_cs1k-webgpu.wasm",
             vram_required_MB: 719.38,
-            low_resource_required: true,
+            low_resource_required: false,
             overrides: {
                 context_window_size: 4096,
             },
@@ -1554,14 +1554,14 @@ const prebuiltAppConfig = {
         // Transformers.js (ONNX) confirmed models from llm_runner_poc
         {
             model: "https://huggingface.co/Xenova/LaMini-Flan-T5-248M",
-            model_id: "LaMini-Flan-T5-248M",
+            model_id: "LaMini-Flan-T5 (~248MB) — Instruction Tuned",
             onnx_id: "Xenova/LaMini-Flan-T5-248M",
             model_lib: "",
             low_resource_required: true,
         },
         {
             model: "https://huggingface.co/Xenova/flan-t5-small",
-            model_id: "flan-t5-small",
+            model_id: "Flan-T5 Small (~300MB) — Instruction Tuned",
             onnx_id: "Xenova/flan-t5-small",
             model_lib: "",
             low_resource_required: true,
@@ -36727,7 +36727,14 @@ class ONNXEngine {
             this.initProgressCallback({
                 progress: 0,
                 timeElapsed: 0,
-                text: `Initializing ONNX fallback for ${repoId}...`
+                text: `Initializing ONNX engine for ${repoId}...`
+            });
+        }
+        if (this.initProgressCallback) {
+            this.initProgressCallback({
+                progress: 0.05,
+                timeElapsed: 0,
+                text: `Loading model configuration and tokenizer...`
             });
         }
         try {
@@ -37983,18 +37990,55 @@ class WebLLM {
     messages = [];
     downloadProgress = {};
     config;
+    isLowResourceDevice = false;
     constructor(config = {}) {
         this.config = config;
         this.messages = this.loadContext();
+        this.detectDevice();
+    }
+    detectDevice() {
+        if (typeof navigator === "undefined")
+            return;
+        // Check for explicit override in config (the user can "tell" the module)
+        if (this.config.is_low_resource !== undefined) {
+            this.isLowResourceDevice = !!this.config.is_low_resource;
+            console.log(`[WebLLM] Using explicit is_low_resource: ${this.isLowResourceDevice}`);
+            return;
+        }
+        const isMobileUA = /Android|webOS|iPhone|iPad|iPod|BlackBerry|IEMobile|Opera Mini/i.test(navigator.userAgent);
+        // Special check for iPad/modern tablets that report as desktop but have touch points
+        const isTablet = navigator.maxTouchPoints > 1 && !window.MSStream;
+        const isMobile = isMobileUA || isTablet;
+        const hasWebGPU = !!navigator.gpu;
+        // If it's mobile OR if WebGPU is missing, we consider it a low-resource/fallback device
+        this.isLowResourceDevice = isMobile || !hasWebGPU;
+        console.log(`[WebLLM] Device detection: isMobile=${isMobile}, hasWebGPU=${hasWebGPU} -> isLowResourceDevice=${this.isLowResourceDevice}`);
+    }
+    get is_low_resource() {
+        return this.isLowResourceDevice;
     }
     // manage_model endpoints
     get models_available() {
-        return prebuiltAppConfig.model_list.map((m) => m.model_id);
+        let list = prebuiltAppConfig.model_list;
+        if (this.isLowResourceDevice) {
+            // Filter for specific approved mobile models
+            const approvedIds = [
+                "LaMini-Flan-T5 (~248MB) — Instruction Tuned",
+                "Flan-T5 Small (~300MB) — Instruction Tuned"
+            ];
+            list = list.filter(m => approvedIds.includes(m.model_id));
+        }
+        return list.map((m) => m.model_id);
     }
     async local_model_available(model_id) {
         return await hasModelInCache(model_id);
     }
     async download_model(model_id, progressCallback) {
+        // Initial feedback
+        const initText = "Initializing model engine...";
+        this.downloadProgress[model_id] = initText;
+        if (progressCallback)
+            progressCallback(initText);
         this.engine = await CreateMLCEngine(model_id, {
             ...this.config,
             initProgressCallback: (progress) => {