Transformers.js: State-of-the-art Machine Learning for the web

[@bs701]

This is going to catapult the chrome memory usage memes to the next level

[@kneelesh48]

Chrome is the OS

[@rude_people_die_young]

If you run the AI powered meme generator in the browser it will create an infinite meme loop 😂

[@yoshmarklund]

Would've been nice if they demoed the models running on a cellphone or Intel/AMD embedded graphics

[@JasonMayes]

Check Moh's talk from Intel also in the playlist

[@naveenkumarpulikonda1998]

Great video. Very useful for people who wants to build AI application in web

[@rhinelandnagel7288]

Liquid LLMs can make in-browser AI even more efficient by dynamically adjusting their complexity based on each task’s needs. For simpler tasks, the model can operate in a “compressed” mode, using fewer resources, and then scale up only when a more complex response is required-all within the same adaptable model. This reduces the memory and processing load on the client side, so the app runs smoothly even with AI features. Combined with transformers.js and WebGPU, liquid LLMs enable responsive, task-specific AI that balances performance and resource use, creating a fast, seamless user experience.

[@FatahChan]

This is amazing, this is what I wanted to do, I am also 24, I am feeling imposter syndrome right now 😢

[@dami8483]

It's normal at 24, even worse at 26. But keep learning, keep failing. Those things build confidence

[@wes555]

We need as many of you guys as we can get. You are not an imposter

[@rhinelandnagel7288]

Does anyone wonder why we save 3 dimensional vectors with vast open spaces between them, why not map to 2d in 3 planes and reduce size, then cache the most used parts of the model as vectors, map the rest, handle model error faults as page faults to update the cache and mappings where mismatches occur just like virtual memory in OS. Index the most used and reconstruct the rest from 2d when fault occurs… no ? Just me ?

[@whatisthis448]

What are the differences from mediapipe library?

[@prakharjain2548]

everything is happening sooooo fassstt...amazed at the limits of what we can achieve here

[@oivoodoo]

wonderful talk. thx

[@pu239]

the word processing is this man's greatest enemy 😁 jk, great video and thanks so much for starting xenova/transformer.js!

[@mubbarikali]

Awesome Work... 👍

[@mariocuric6690]

Wait this is big right?

[@akj3344]

no

[@compton8301]

South Africa to the world. 🎉

[@wes555]

South Africa represent

[@Quaquaquaqua]

I need this!

[@samrathchauhan1271]

This is game changing.

[@CesarAnton]

Inspiring

[@s4ndeep1203]

really impressive !

[@peterhayman]

Webgpu support for candle plz 🕯

[@JeomonGeorge]

game changer

[@kasper369]

web developers moving from web to mobile and now to AI, Less gu!

[@wadexyz]

Wow !

[@vigoworkchannel1681]

Real ML engineers want this working seamlessly in node.js. I’ve ran many models in the browser vs node.js backend. The workflow is better when the model is trained in Node and packed for the web or even let Node handle predictions/inference via rest API or websockets

[@jagadeeshch7870]

wow

[@uchennaofoma4624]

2025 is gonna be fun 🙂

[@MysticZA]

hulle weet nie wat ons weet

[@manu144x]

Can someone write a bitcoin miner for the browser? I need it...for a friend.

[@rns10]

Another hype for AI. We don't need it. Get over your obsession. Fix your manifest v3 and give the user controls over their browser security.

[@vigoworkchannel1681]

I also haven’t been able to think of any long term use cases for active machine learning in the web, but what’s this thing about the manifest v3. Are you talking against their new experimentation to block ad blockers?

[@rns10]

@vigoworkchannel1681 yes. They are deprecating APIs that ad blockers use to detect and remove ads from the DOM. And then they we we are protecting you from insecure extensions. The irony.

[@pu239]

this is not something google has made. this is huggingface and is open source.

[@rns10]

@pu239 but Google is promoting it.

[@rns10]

@@pu239 then why are they promoting it in the first place on their own youtube channel. We dont need AI. Its not even profitable for any company other than open ai. So why are we still debating its good usage. Just fix the web, remove useless content, fix google search, give organic search results.