From 81c3b06b92969df74ae856adf7cd194ba19f1bca Mon Sep 17 00:00:00 2001
From: Kwangsoo Yeo <kyeo@picovoice.ai>
Date: Mon, 3 Jun 2024 10:58:56 -0700
Subject: [PATCH] update

---
 recipes/llm-voice-assistant/web/README.md        |  2 +-
 .../llm-voice-assistant/web/public/controller.js |  7 ++++++-
 recipes/llm-voice-assistant/web/src/main.ts      | 16 ++++++++--------
 3 files changed, 15 insertions(+), 10 deletions(-)

diff --git a/recipes/llm-voice-assistant/web/README.md b/recipes/llm-voice-assistant/web/README.md
index 4a89f87..9850d9b 100644
--- a/recipes/llm-voice-assistant/web/README.md
+++ b/recipes/llm-voice-assistant/web/README.md
@@ -23,7 +23,7 @@ picoLLM Inference Engine supports many open-weight models. The models are on
 Install the required packages using `yarn`:
 
 ```console
-yarn isntall
+yarn install
 ```
 
 or using `npm`:
diff --git a/recipes/llm-voice-assistant/web/public/controller.js b/recipes/llm-voice-assistant/web/public/controller.js
index a754f4f..35728d9 100644
--- a/recipes/llm-voice-assistant/web/public/controller.js
+++ b/recipes/llm-voice-assistant/web/public/controller.js
@@ -97,7 +97,12 @@ window.onload = () => {
     let streamCalls = 0;
 
     try {
-      await Picovoice.init(accessKey.value, {
+      await Picovoice.init(accessKey.value,
+        {
+          modelFile: modelFile.files[0],
+          cacheFilePath: modelFile.files[0].name,
+        },
+        {
         onDetection: () => {
           message.innerText = "Wake word detected, utter your request or question...";
           humanElem = startHumanMessage();
diff --git a/recipes/llm-voice-assistant/web/src/main.ts b/recipes/llm-voice-assistant/web/src/main.ts
index 89463cc..f84bfcc 100644
--- a/recipes/llm-voice-assistant/web/src/main.ts
+++ b/recipes/llm-voice-assistant/web/src/main.ts
@@ -3,7 +3,7 @@ import { Mutex } from 'async-mutex';
 import { BuiltInKeyword, PorcupineDetection, PorcupineWorker } from '@picovoice/porcupine-web';
 import { CheetahTranscript, CheetahWorker } from '@picovoice/cheetah-web';
 import { OrcaStreamWorker, OrcaWorker } from '@picovoice/orca-web';
-import { Dialog, PicoLLMWorker } from '@picovoice/picollm-web';
+import { Dialog, PicoLLMModel, PicoLLMWorker } from '@picovoice/picollm-web';
 import { WebVoiceProcessor } from '@picovoice/web-voice-processor';
 
 type PvObject = {
@@ -26,7 +26,10 @@ type PvCallback = {
 
 let object: PvObject | null = null;
 
-const init = async (accessKey: string, {
+const init = async (
+  accessKey: string,
+  picollmModel: PicoLLMModel,
+  {
   onDetection,
   onEndpoint,
   onTranscript,
@@ -78,9 +81,7 @@ const init = async (accessKey: string, {
 
   const pllm = await PicoLLMWorker.create(
     accessKey,
-    {
-      modelFile: "models/picollm_params.pllm"
-    });
+    picollmModel);
 
   const orca = await OrcaWorker.create(
     accessKey,
@@ -97,9 +98,10 @@ const init = async (accessKey: string, {
   let stopTokens = 0;
 
   const stopPhrases = [
-    '</s>', // Llama-2, Mistral, and Mixtral
+    '</s>', // Llama-2, Mistral
     '<end_of_turn>', // Gemma
     '<|endoftext|>', // Phi-2
+    '<|eot_id|>',  // Llama-3
   ];
 
   const onCheetahFlushed = async (): Promise<void> => {
@@ -109,8 +111,6 @@ const init = async (accessKey: string, {
 
     const { completion, completionTokens } = await pllm.generate(dialog.prompt(), {
       completionTokenLimit: 128,
-      temperature: 0.7,
-      topP: 0.6,
       streamCallback: async token => {
         if (!stopPhrases.includes(token)) {
           onText(token);