From e82460b93edae6679d626259e4b4e8ab638daeb6 Mon Sep 17 00:00:00 2001
From: zoollcar <zoollcar@qq.com>
Date: Thu, 14 Nov 2024 16:34:09 +0800
Subject: [PATCH] add mobile model

---
 src/App.tsx                     |  6 +++---
 src/models/llm/LLMChatWebLLM.ts | 19 +++++++++++--------
 src/models/tts/textToSpeech.ts  |  1 -
 3 files changed, 14 insertions(+), 12 deletions(-)

diff --git a/src/App.tsx b/src/App.tsx
index e6b6d97..74d733f 100644
--- a/src/App.tsx
+++ b/src/App.tsx
@@ -246,16 +246,16 @@ function App() {
     if (chat instanceof LLMChatWebLLM) {
       if (chat.getInitStatus() === "not start") {
         const answer = confirm(
-          "webLLM need to load at every time, first time may need some time to download model(~1.5G). load now?"
+          "webLLM need to load every time, first time need some time to download model(~1.5G(PC)/~800MB(phone)). load now?"
         );
         if (answer) {
           const timer = setInterval(() => {
             setSubtitle(chat.initProgress || "webLLM loading");
           }, 1000);
           chat.init().then(() => {
-            alert("webLLM loaded");
             clearInterval(timer);
-            setSubtitle("");
+            if (chat.getInitStatus() === "done") setSubtitle("webLLM loaded");
+            else setSubtitle("webLLM error");
           });
         }
         return;
diff --git a/src/models/llm/LLMChatWebLLM.ts b/src/models/llm/LLMChatWebLLM.ts
index ab041ba..c63c91a 100644
--- a/src/models/llm/LLMChatWebLLM.ts
+++ b/src/models/llm/LLMChatWebLLM.ts
@@ -24,21 +24,24 @@ export default class LLMChatWebLLM {
 
   async init() {
     // This is an asynchronous call and can take a long time to finish
+    // models are https://github.com/mlc-ai/web-llm/blob/main/src/config.ts
     this.initStatus = "working";
-    const maxStorageBufferBindingSize =
-      await this.client.getMaxStorageBufferBindingSize();
-    console.log(maxStorageBufferBindingSize);
-    let selectedModel = "RedPajama-INCITE-Chat-3B-v1-q4f16_1-1k";
-    if (maxStorageBufferBindingSize >= 2147480000) {
-      // ~2G
-      selectedModel = "Llama-3.2-3B-Instruct-q4f16_1-MLC";
-    }
     try {
+      const maxStorageBufferBindingSize =
+        await this.client.getMaxStorageBufferBindingSize();
+      // alert(maxStorageBufferBindingSize);
+      let selectedModel = "SmolLM2-135M-Instruct-q0f32-MLC";
+      if (maxStorageBufferBindingSize >= 1073741800) {
+        // ~1G
+        selectedModel = "Llama-3.2-3B-Instruct-q4f16_1-MLC";
+      }
+      console.log(`webLLM: select ${selectedModel}`);
       await this.client.reload(selectedModel);
       this.initStatus = "done";
     } catch (e) {
       this.initStatus = "error";
       this.initProgress = (e as Error).message;
+      alert("Error: " + (e as Error).message);
     }
   }
 
diff --git a/src/models/tts/textToSpeech.ts b/src/models/tts/textToSpeech.ts
index a45b95a..5ea135f 100644
--- a/src/models/tts/textToSpeech.ts
+++ b/src/models/tts/textToSpeech.ts
@@ -1,6 +1,5 @@
 import { getBackendEndpoint } from "../appstore";
 
-// const vitsWebWorker = new Worker("./vitsWeb");
 const vitsWebWorker = new Worker(new URL("./vitsWeb.ts", import.meta.url), {
   type: "module", // Specify that the worker is a module
 });