DiffusionGemma: 4X MAIS RÁPIDO com 10% de perda de qualidade
DiffusionGemma is Google's new experimental model that uses text diffusion instead of traditional autoregressive generation, reaching over 1000 tokens per second on dedicated hardware. In this video, I install and run DiffusionGemma locally on a Mac M3 Ultra using llama.cpp, showing in real time how the model rewrites and refines its own response during generation. I tested everything from creating a Flappy Bird clone in HTML to a complete landing page. If you want to understand how diffusion architecture works in practice, why it's up to 4x faster than autoregressive Gemma 4, and what that means for those running AI models locally, this video is for you. 👉 Here's the installation link: https://unsloth.ai/docs/models/diffus... _______ ⚡ Coding Tools: ⏵ GLM - https://z.ai/subscribe?ic=1CUTHGZM5X ⏵ AI Coding Plan - https://www.alibabacloud.com/en/campa... ⏵ Alibaba Cloud - https://www.alibabacloud.com/en/campa... ⚡ TTS ⏵ ElevenLabs - Text-to-Speech: https://try.elevenlabs.io/aiprogbr-tts ⏵ ElevenLabs - Voice Clone: https://try.elevenlabs.io/aiprogbr-vo... ⚡Hardware ⏵ ZimaCube: https://shop.zimaspace.com/products/z... ca_source=ytAi&utm_source=igor-oliveira&utm_medium=affiliate&utm_campaign=zimaspace_affiliate&utm_content=Igor-Oliveira ⏵ ZimaBoard 2: https://shop.zimaspace.com/products/z... ⚠️ Use coupon: AIProgBr15 to receive a $15.00 discount ⏵ Dock to connect the Video Card to the ZimaBoard 2: https://shop.zimaspace.com/products/g... ⏵ ZimaBlade: https://shop.zimaspace.com/products/z... _______ 🔹 Want to learn how to create apps? From scratch to publication with me? Then check out my course: https://programadorbr.com/flutter?src... _______ 0:00 Intro 0:34 Documentation 3:39 Initial Tests 5:53 Flappy Bird 7:05 Pinball 8:22 Landing Page 9:31 Conclusion _______ 👀 If you're curious, take a look at some of my published apps: ⏵ Teleprompter: https://apps.apple.com/app/apple-stor... ⏵ Pomodoro: https://apps.apple.com/app/apple-stor... ⏵ Infinite Tic-Tac-Toe iOS: https://apps.apple.com/app/apple-stor... ⏵ Reverse Play iOS: https://apps.apple.com/ca/app/reverse... _______ ✴️ Also follow my other channels: ⏵ Channel about the Apple universe: / @iprogbr ⏵ Programming channel: / @programadorbr ⏵ Instagram profile: / progbr ⏵ Substack profile: https://progbr.substack.com/ Some of the links in this description may be affiliate links, meaning they may generate commission for the channel at no extra cost to you.

Kimi K2.7 Code: 5x cheaper than Opus 4.8 and does it come close?

Google Just Found a Loophole in AI Hardware Limitations

How to use MTP in llama.cpp to ABSURDLY increase the speed of your local AI

I Built an AI Operating System with Claude Fable 5 (RAW RESULTS)

I Think They Are Lying To You

Google TurboQuant: Local LLM 8x FASTER, in 6x LESS memory, with ZERO quality loss.

I Drank 30 TBSP of Olive Oil in 10 Days… My Body Fat MELTED (Day 3 Wasn’t Normal)

Yann LeCun Says LLMs Have 2 Years Left…

Samsung's 990 Pro SSD warranty policy is a scam; I'm taking them to court.

AMD's Strix Successor Just Caught the M4 Pro

RTX Spark: NVIDIA is trying to HIDE a detail that changes EVERYTHING for local LLMs

I ran Qwen 3.6 35B on 8GB of VRAM at almost 20 t/s (COMPLETE TUTORIAL using llama.cpp)

MiniMax M3 REBORN: multimodal, 1M tokens of CONTEXT and trained on 100T

Why Aliens Would NEVER Invade Africa

I Made Opus 4.8 and Fable 5 Build the Same App (RAW RESULTS)

It's Boring, But It Destroys Your Visceral Fat In 14 Days (Japanese Method)

How Did DeepSeek V4 Make V4 So Cheap?

Kimi K2.7 Code Is HERE – Is THIS the Best Open Coding Model Yet?

The world's BEST AI models NO LONGER EXIST: Fable 5 and Mythos 5 have been BANNED

