Utku Çoban

CS student @virginia_tech

[Utku Çoban @UtkuCoban000]

@gokayfem I love Director, it’s exactly what I’m looking for. But why doesn’t it support H3 Max Turbo?

[Utku Çoban @UtkuCoban000]

@mustafaakin Dataset'i nasıl oluşturdunuz? Bölümlerin sesini başka bir modelle yazıya mı döktünüz yoksa hazır yazı halinde mi bulunuyor?

[Utku Çoban @UtkuCoban000]

@mervenoyann I don't know what they are feeding the 27B models. 3.6 27B punches waay above its weight.

[Utku Çoban @UtkuCoban000]

Try it out here: https://t.co/a8Jy0BOg5U

Technical details:https://t.co/UEy3oBOtmF

[Utku Çoban @UtkuCoban000]

I made an AI Image generation model under 50MB that runs entirely locally in your browser using WebGPU, generates an image in under 100ms, has only 24M parameters, it uses my custom GAN implementation also its open weights

Link in replies

[Utku Çoban @UtkuCoban000]

@aisearchio I love diffusion models because, unlike LLMs, you can actually run frontier models on accessible consumer GPUs.

[Utku Çoban @UtkuCoban000]

Thanks @huggingface community for almost 50 likes!
Version 2 training is underway with much better consistency and fewer model collapses.

Will keep pushing what's possible for generative models using local WASM CPU inference!

https://t.co/ByRuybOOzO

[Utku Çoban @UtkuCoban000]

@sama Tiny LSTM model generating music realtime on any device fully in browser https://t.co/sWRpqSXAgN

https://t.co/ByRuybOOzO

[Utku Çoban @UtkuCoban000]

@sama https://t.co/sWRpqSXAgN

[Utku Çoban @UtkuCoban000]

Almost at 30 likes! #1 among music generation spaces this week

Thank you @huggingface @HuggingModels community!

HF space: https://t.co/ByRuybOOzO

Web app: https://t.co/OBFnl2Uu5y

[Utku Çoban @UtkuCoban000]

You can now use different instruments for playback!
Try it out at: https://t.co/OBFnl2Uu5y

[Utku Çoban @UtkuCoban000]

Sharing your favorite songs just got incredibly easy with the new share button!
https://t.co/OBFnl2Uu5y

The best part? No actual song data is shared, only a simple seed is transmitted, yet the real-time playback will match perfectly every single time.

[Utku Çoban @UtkuCoban000]

@jessu_js Antigravity de kullanılacak tek model 3.5 flashdır şuanda ekstra hızlı inference sunuyorlar tabi sadece hafif/front end işleri için web design da görsellik olarak 5.5 xhigh dan bile daha iyi

[Utku Çoban @UtkuCoban000]

@Naiknelofar788 Looking forward to your feedback! It's also open source and the #1 music generation space on Hugging Face right now.

[Utku Çoban @UtkuCoban000]

Generate endless real-time AI music directly in your browser

No sign-up. Free unlimited generations. Runs locally in your browser with transformers js

Tiny model: ~13M params, ~50MB. Runs in real time on almost any consumer CPU, even mobile.

Try it out

[Utku Çoban @UtkuCoban000]

@earth_ish Thank you for the feedback! I’m aware of this issue, and I believe it comes from noise in the training data. This noise accumulates over time and becomes worse with each prediction, especially because the sequence lengths are quite short.

[Utku Çoban @UtkuCoban000]

Generate endless real-time AI music directly in your browser

No sign-up. Free unlimited generations. Runs locally in your browser with transformers js

Tiny model: ~13M params, ~50MB. Runs in real time on almost any consumer CPU, even mobile.

Try it out

[Utku Çoban @UtkuCoban000]

Infinite original songs. Generated live on your device.

NanoMaestro streams endless piano music right in your browser no sign-ups, no subscriptions, no downloads, and fully offline.

A tiny neural network creates every song in real time.

[Utku Çoban @UtkuCoban000]

A much more polished version of NanoMaestro is now available at https://t.co/OBFnl2Uu5y Try it out!

[Utku Çoban @UtkuCoban000]

@muhammet_bsr Uygulama indirmeye gerek kalmadan tarayıcıda yapay zeka çalıştırmak daha bile zor, bu yüzden evet, kesinlikle uygulama yapılabilir. Farklı müzik tarzlarında da üretim üzerinde çalışıyorum. Sırada daha klasik barok parçalar var.

[Utku Çoban @UtkuCoban000]

Generate endless real-time AI music directly in your browser

No sign-up. Free unlimited generations. Runs locally in your browser with transformers js

Tiny model: ~13M params, ~50MB. Runs in real time on almost any consumer CPU, even mobile.

Try it out

[Utku Çoban @UtkuCoban000]

@FreeFairDOG @zxc68984008 Yes, this model is a 2-layer LSTM model and has no transformer architecture, which is actually very useful for this case. All it has to do is predict the next token (musical note), which has an easy to recognize pattern and does not need a complex neural network architecture.