◕‿◕

both outputs are from 8B Llama but one is "mechanicaly opened" We were doing LLMs wrong this whole time LLMs are much more powerful yes to compare one needs to concentrate but vanilla lama is just slop and repetition (of ideas) and "opened" version actualy thinks! no training etc 'just': sheduling magic, entropix generalization ensemble with smaller 1B model

@lucataco93 @rudzinskimaciej @replicate Thank you 🙏🏼 We’re thinking of open sourcing the code with some more detail behind it after we finish our current LLM scheduling investigation 😌

@neurowelt @rudzinskimaciej @replicate This is awesome! This should be a feature model imo

Along with @rudzinskimaciej we have recently published our Keros diffusion model on @replicate 🔥 Based on SD XL it features several cool parameters to control the image generation process 😌 replicate.com/neurowelt/kero…

Prompt: abstract liquid metal, northern lights, fluorescent stars Richness: 1.1 Contrast: 0.96 Texture: 0.54 Background: 0.0 Focus: 0.16 Variance: -0.08

Prompt: serious man on a job interview for developer job, colorized photograph, circa 2000, liminal space Richness: 1.16 Contrast: 0.96 Texture: 0.84 Background: 0.015 Focus: 0.12 Variance: -0.08

@ai_for_humans I always asked myself this question and it turns out there is an original paper on it from 2003 😄 → ee.columbia.edu/~dpwe/papers/W…

How does Shazam work so well and so fast 🤯

@rudzinskimaciej @GoAbiAryan @musikalkemist @BradenRiggs1 I really enjoyed Coqui.ai TTS, easy to use and has some powerful models in it 😊 --> github.com/coqui-ai/TTS

@GoAbiAryan @musikalkemist @BradenRiggs1 @neurowelt ?

Who has been keeping tabs on all the interesting work coming up with audio-speech synthesis? Any resources/libraries one should know about? CC: @musikalkemist @BradenRiggs1

That's the input image I used 🤖

In Character ~ 2 Trapped in a maze of our own design #stablediffusion

Although #stablediffusion art has taken over, our brain is quite a brilliant artist too! Spent some time working with EEG processing and these are just a snippet of what your brain can draw without anything but its own activity 😍 Now think about using brain to generate art 🙃

After giving @prof_sinister this possibility none of the prompts would fit in a tweet 😅

In Character ~ 1 This week @prof_sinister went back to some character design using #stablediffussion, so I hopped in as well and tried it out – hackers were my first choice as usually

@prof_sinister @tormential I can confirm, soon ☺️

@tormential dear @neurowelt we need to speed up our projects! ;) soon (tm) seriously

How to enter the Weekend #stablediffusion

Cyber Visions ~ 2 Skull formations from wire and glass #stablediffusion

If you know about some EEG recordings that are available for analysis, please do let us know! We're about to dive into some exciting research concerning #brain and the basic principles of human #behavior 🎯

Cyber Visions ~ 1 Creatures moulded by neon lights and fast bandwidth #stablediffusion

Cyberpunk landscapes, all coming from the same simple image. It's amazing to see that the initial colors remain and set the tone of the image 😍 #stablediffusion

From doodles to pretty captivating drawings. I'm trying to figure out how to control creativity of the model while at the same time keeping it close to the structure of the initial image (while not using strength) 😄 #stablediffusion