Windows 11 Pro 22631.3880 - Windows Feature Experience Pack 1000.22700.1020.0
Chrome Canary v128.0.6589.0

Enabled WebGPU, prompt-api-for-gemini-nano and optimization-guide-on-device-model in all possible combinations and the Optimization Guide On Device Model isn't showing up even after restarting the browser or the PC.

Gemini Nano weights from Google Chrome are on HuggingFace. You can run the inference using this model with MediaPipe LLM inference based on WebGPU.

Please add it as a fallback for other browsers, so that they could chat offline with the on-device LLM. The model could be loaded from
local weights.bin file, like in MediaPipe Studio.

@lightning-joyce

In chrome canary 128.0.6589.0 (Official Build) canary (64-bit), when send prompt, there is a "Unhandled Runtime Error TypeError: model.execute is not a function" error

I fixed it in my local env, update code :

to "let aiReplay = (await model.prompt(prompt)) as unknown as string;", it will works

Hi, I found an internal page that might be able to manually load the model (chrome://on-device-internals/).
As I can't find a way to make the 'Optimization Guide On Device Model' option appear in chrome://components, I can't test that out myself.
I'm wondering if there's any way you might be able to perhaps provide a copy for the Gemini nano model so that users may load manually.

I've followed all steps in the guide however on chrome://components, the required component is not shown. I've already set my primary language to english like mentioned in some guides but it doesn't show up nontheless.

I followed all steps and still facing error, i have also enabled WebGPU.

RT
flags里开开关关试过了，dev版比canary多出一个screenai library，还要就是这个需要登录账号吗？我尝试了登录账号，退出账号，component里都还是没有该选项。

Only for Canary Version: 130.0.6679.0 or above (check chrome://version)

As the title says, createTextSession is no longer valid in the newest canary version.

The new way to create a session is now:

const model = await window.ai.assistant.create()
await model.prompt("What is 2+2?"); // -> ' 2 + 2 = 4'

To get the ready state of the ai you would do:

const status = await window.ai.assistant.capabilities()
const ready = status.available

if (ready === 'readily') {
    // create ai session
} else {
    throw new Error('The AI is not ready!')
}

and lastly, before doing anything with the ai check if ai is a property to avoid errors:

const buildSupported = window.hasOwnProperty("ai") === true
if (buildSupported) {
   // Ai functions
}

A full example of how I would recommend the AI features here:

async function createAISession() {
    const buildSupported = window.hasOwnProperty("ai") === true
    if (buildSupported) {
        const status = await window.ai.assistant.capabilities()
        const ready = status.available
        if (ready === 'readily') {
            const model = await window.ai.assistant.create()
            return model;
        } else {
            throw new Error('The AI is not ready! Please try again later!')
        }
    } else {
        throw new Error('Your current chrome build doesn\'t support window.ai features. Please check the documentation and repeat the steps if you\'ve missed any')
    }
}

with this, the AI should be fully initialized and usable

• thanks for reading.

After I have fully configured it, enter 'await window.ai.canCreateTextSession(); ',returning 'readily', but encountering this issue when calling again, what should I do to use it?