Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

DiffusionGemma is out 🔥 it's compute-bound so 4x faster compared to other Gemma-4 models (1k tok/s on H100) 💨 also great on coding, generate and iterate on any code from 3D generation to front-end ⤵️

merve

87,806 subscribers

40,717 views • 2 months ago •via X (Twitter)

Education Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

what if you could instantly clone any object? Gemma 4 is now on Cerebras Inference, running up to 10x faster than GPUs (1,500 tokens/sec). Multimodal generations you can iterate on in real time :)

what if you could instantly clone any object? Gemma 4 is now on Cerebras Inference, running up to 10x faster than GPUs (1,500 tokens/sec). Multimodal generations you can iterate on in real time :)

Sarah Chieng

32,811 views • 1 month ago

Want to quickly craft and iterate on prompts? Check out Prompt builder in the AI Toolkit for VS Code ( Experiment with local or remote models, generate prompts using AI, and test multi-turn conversations. Plus, view the code to start building right away! 💡

Want to quickly craft and iterate on prompts? Check out Prompt builder in the AI Toolkit for VS Code ( Experiment with local or remote models, generate prompts using AI, and test multi-turn conversations. Plus, view the code to start building right away! 💡

Visual Studio Code

52,751 views • 1 year ago

Finally got my hands on the big one. Qwen3.5-122B-A10B — 122 billion parameters. Too big for any single consumer GPU. So I rented 4 of each... and then one professional card to see if brute force even matters. - 1x RTX PRO 6000 (96GB): 101.4 tok/s - 4x 5090 (128GB): 87.0 tok/s - 4x 4090 (96GB): 25.1 tok/s - 4x 3090 (96GB): 20.8 tok/s One single $8,500 card beat four RTX 5090s

Finally got my hands on the big one. Qwen3.5-122B-A10B — 122 billion parameters. Too big for any single consumer GPU. So I rented 4 of each... and then one professional card to see if brute force even matters. - 1x RTX PRO 6000 (96GB): 101.4 tok/s - 4x 5090 (128GB): 87.0 tok/s - 4x 4090 (96GB): 25.1 tok/s - 4x 3090 (96GB): 20.8 tok/s One single $8,500 card beat four RTX 5090s

stevibe

195,832 views • 5 months ago

to run gemma 4 on your phone without any login or internet: - download Google AI Edge Gallery - select Gemma 4 E2B/E4B - thats it I wish they give support to add other edge models in future like locallyAI here is a video of gemma 4 E4B in action at 13 tps on iphone 15 pro max

to run gemma 4 on your phone without any login or internet: - download Google AI Edge Gallery - select Gemma 4 E2B/E4B - thats it I wish they give support to add other edge models in future like locallyAI here is a video of gemma 4 E4B in action at 13 tps on iphone 15 pro max

Harveen Singh Chadha

13,552 views • 4 months ago

Adobe is entering the image-to-3D game! Their new method, LRM, can create high-fidelity 3D meshes from a single image in just 5 seconds 🔥 It's trained on 1 million objects and is able to generate objects from real-world images and generative AI models.

Adobe is entering the image-to-3D game! Their new method, LRM, can create high-fidelity 3D meshes from a single image in just 5 seconds 🔥 It's trained on 1 million objects and is able to generate objects from real-world images and generative AI models.

Dreaming Tulpa 🥓👑

270,942 views • 2 years ago

i keep hearing about how GLM-5.2 not only is great at coding but also that is on Opus 4.8 level when it comes to front end design I wanted to find out so... i put it to the test.. you can be the judge

i keep hearing about how GLM-5.2 not only is great at coding but also that is on Opus 4.8 level when it comes to front end design I wanted to find out so... i put it to the test.. you can be the judge

Hermes Agent Tips

71,024 views • 1 month ago

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined @TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined @TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

Inception

27,108 views • 5 months ago

Google's new Gemma 4 is excellent, and the 26B MoE version is likely the best model to run on a 32GB Framework Desktop. It's fast, smart, and also great for tool calling if you use it with OpenClaw🦞 or other local agent platforms.

Google's new Gemma 4 is excellent, and the 26B MoE version is likely the best model to run on a 32GB Framework Desktop. It's fast, smart, and also great for tool calling if you use it with OpenClaw🦞 or other local agent platforms.

Framework

58,593 views • 4 months ago

We compared 1-bit Kimi K3 to Claude Opus 5 and GPT 5.6. We gave 4 models the same prompt: Create a glass aquarium whose side panel develops a visible crack and then bursts. 1-bit Kimi K3 GGUF ran locally on 4x B200s at 36 tok/s. GitHub repo:

We compared 1-bit Kimi K3 to Claude Opus 5 and GPT 5.6. We gave 4 models the same prompt: Create a glass aquarium whose side panel develops a visible crack and then bursts. 1-bit Kimi K3 GGUF ran locally on 4x B200s at 36 tok/s. GitHub repo:

Unsloth AI

413,594 views • 17 days ago

We are thrilled to release Chat-to-3D on CSM AI, to enable unprecedented mesh quality and controllability. This entire character was prompted end-to-end using this release. 🛠️ Automatically generate parts for complex assets 🌐 Generate T-pose outputs from any input 🎨 Generate style variations in one-shot Image generation is powered by gpt-image. Available to all users (creative pro+ members get highest resolution outputs).

We are thrilled to release Chat-to-3D on CSM AI, to enable unprecedented mesh quality and controllability. This entire character was prompted end-to-end using this release. 🛠️ Automatically generate parts for complex assets 🌐 Generate T-pose outputs from any input 🎨 Generate style variations in one-shot Image generation is powered by gpt-image. Available to all users (creative pro+ members get highest resolution outputs).

Common Sense Machines

30,404 views • 1 year ago

⚙️Roblox Integration & Plugin We're working on a two-step Roblox integration. First, users can connect their Roblox account on our web app, generate 3D models, and upload them directly to Roblox. Next, we’ll add a Roblox plugin to generate 3D models within Roblox, already in your account. We'll also integrate our landscape generation into this process. 🖥 Discover the full details in the video!

⚙️Roblox Integration & Plugin We're working on a two-step Roblox integration. First, users can connect their Roblox account on our web app, generate 3D models, and upload them directly to Roblox. Next, we’ll add a Roblox plugin to generate 3D models within Roblox, already in your account. We'll also integrate our landscape generation into this process. 🖥 Discover the full details in the video!

NeuralAI

20,857 views • 1 year ago

Cursor with Kimi K2.5. Don't sleep on this combo. From a prompt to a personal HN feed in about ~60 seconds. The future of building is going to be so wild. With faster models, you can quickly iterate on more ideas, while improving quality.

Cursor with Kimi K2.5. Don't sleep on this combo. From a prompt to a personal HN feed in about ~60 seconds. The future of building is going to be so wild. With faster models, you can quickly iterate on more ideas, while improving quality.

elvis

11,283 views • 5 months ago

got gemma 4 31B with MTP running on my DGX Spark. Hermes Agent did most of the legwork. baseline vs MTP on GB10: • c=1: 3.65 → 6.37 tok/s (1.74x) • c=4: 14.34 → 23.59 tok/s (1.65x) • c=8: 14.37 → 24.18 tok/s (1.68x) google says "up to 2x" — we're not quite there but it's real, not vapor. stack: DGX Spark / GB10 + gemma-4-31b-it + gemma-4-31b-it-assistant (MTP drafter) + vLLM built from PR 41745 MTP is basically a lightweight draft model that predicts multiple tokens while the big model verifies them all at once. smaller model does the busywork, bigger model just says yes/no. simple idea, weird to implement. next: tune the draft block size and see if we can push past 2x. also want to try it with Hermes Agent feeding prompts end to end. p.s: this was all done from telegram. Google DeepMind NVIDIA AI Developer

got gemma 4 31B with MTP running on my DGX Spark. Hermes Agent did most of the legwork. baseline vs MTP on GB10: • c=1: 3.65 → 6.37 tok/s (1.74x) • c=4: 14.34 → 23.59 tok/s (1.65x) • c=8: 14.37 → 24.18 tok/s (1.68x) google says "up to 2x" — we're not quite there but it's real, not vapor. stack: DGX Spark / GB10 + gemma-4-31b-it + gemma-4-31b-it-assistant (MTP drafter) + vLLM built from PR 41745 MTP is basically a lightweight draft model that predicts multiple tokens while the big model verifies them all at once. smaller model does the busywork, bigger model just says yes/no. simple idea, weird to implement. next: tune the draft block size and see if we can push past 2x. also want to try it with Hermes Agent feeding prompts end to end. p.s: this was all done from telegram. Google DeepMind NVIDIA AI Developer

Joey

23,088 views • 3 months ago

🌘 Meet Kimi K2.7 Code HighSpeed! A high-speed mode of our latest open-source multimodal coding model, Kimi K2.7 Code. ⚡️ Up to 6× faster: Around 180 tok/s on coding tasks with median-length inputs, and up to 260 tok/s on shorter-context tasks. 🔷 Rolling out to Kimi Code Beta Program members, Kimi API developers, and Kimi Business users. (Access will remain limited for now due to capacity constraints.) 🔷 No invite needed. Anyone who joins the Beta Program has a chance to get access 👉 Open intelligence should be instant, affordable, and borderless. We'll continue improving the model and expanding access as more capacity becomes available! 🔗 Kimi Code: 🔗 API:

🌘 Meet Kimi K2.7 Code HighSpeed! A high-speed mode of our latest open-source multimodal coding model, Kimi K2.7 Code. ⚡️ Up to 6× faster: Around 180 tok/s on coding tasks with median-length inputs, and up to 260 tok/s on shorter-context tasks. 🔷 Rolling out to Kimi Code Beta Program members, Kimi API developers, and Kimi Business users. (Access will remain limited for now due to capacity constraints.) 🔷 No invite needed. Anyone who joins the Beta Program has a chance to get access 👉 Open intelligence should be instant, affordable, and borderless. We'll continue improving the model and expanding access as more capacity becomes available! 🔗 Kimi Code: 🔗 API:

Kimi.ai

569,736 views • 2 months ago

Move from question to insight faster with the AI first Colab, powered by Gemini. Everything you need to accelerate your data science workflow is here. ✨ Generate code and explore data through simple prompts 🛠️ Rewrite code to fix errors on the fly 📋 Propose plans and execute end-to-end workflows Meet the new Colab →

Move from question to insight faster with the AI first Colab, powered by Gemini. Everything you need to accelerate your data science workflow is here. ✨ Generate code and explore data through simple prompts 🛠️ Rewrite code to fix errors on the fly 📋 Propose plans and execute end-to-end workflows Meet the new Colab →

Google for Developers

27,740 views • 2 months ago

Sam Altman says OpenAI has reached another GPT-3-style moment Something important is working so well that other strong projects have to give way That is why OpenAI is focusing its compute on next-generation models and the automated researchers and companies they can power

Sam Altman says OpenAI has reached another GPT-3-style moment Something important is working so well that other strong projects have to give way That is why OpenAI is focusing its compute on next-generation models and the automated researchers and companies they can power

Haider.

59,715 views • 4 months ago

I recorded a video tutorial on building an AI Code Assistant. Going through the full stack - from data models to the front end. Using Postgres, pg_vector, LangChain, OpenAI, and NextJS.

I recorded a video tutorial on building an AI Code Assistant. Going through the full stack - from data models to the front end. Using Postgres, pg_vector, LangChain, OpenAI, and NextJS.

Gwen (Chen) Shapira

30,463 views • 2 years ago

⚡️ NEW: Nosana launches AI Grants. Up to $50K for startups, researchers, and solo builders working on what’s next in AI. Selected teams also get free GPU compute credits, cutting infrastructure costs so you can move faster from idea to deployment. Nosana operates a global GPU marketplace focused on accessible and affordable compute. [Brought to you by Nosana]

⚡️ NEW: Nosana launches AI Grants. Up to $50K for startups, researchers, and solo builders working on what’s next in AI. Selected teams also get free GPU compute credits, cutting infrastructure costs so you can move faster from idea to deployment. Nosana operates a global GPU marketplace focused on accessible and affordable compute. [Brought to you by Nosana]

Cointelegraph

58,396 views • 8 months ago

Diffusion Gemma is 4x faster, but makes 6x more mistakes! We benchmarked the new diffusion LLM against its autoregressive twin on a single H100 (FP8). We gave each the same three tasks: write a Steve Jobs biography, the history of Tetris, and the story of BeOS - every next topic less popular than the previous one. Then we fact-checked every claim in every answer. Gemma4 got 45 facts right, 5 wrong. DiffusionGemma got 33 right, 28 wrong. The less popular the topic, the worse it got: 4 mistakes on Jobs, 12 on Tetris, 12 on BeOS. It named Clara Clley as Steve Jobs' mother, invented a colleague for Pajitnov named Geri Gulovik and priced the BeBox at $9,999. The real one cost $1,600. Outputs: Gemma4 26B A4B: 218 tok/s · 15.1s total · 45 facts · 5 mistakes DiffusionGemma 26B A4B: 763 tok/s · 3.7s total · 33 facts · 28 mistakes The reason is simple. DiffusionGemma throws 256 tokens on the screen at once and polishes them pass after pass until the text sounds smooth. Smooth is all it cares about: a fake name, date or number sounds just as smooth as a real one, so it stays. Regular Gemma4 meanwhile writes one word at a time and checks every new word against everything before it. Google says it themselves in the launch post: quality is lower, use regular Gemma 4 when facts matter.

Diffusion Gemma is 4x faster, but makes 6x more mistakes! We benchmarked the new diffusion LLM against its autoregressive twin on a single H100 (FP8). We gave each the same three tasks: write a Steve Jobs biography, the history of Tetris, and the story of BeOS - every next topic less popular than the previous one. Then we fact-checked every claim in every answer. Gemma4 got 45 facts right, 5 wrong. DiffusionGemma got 33 right, 28 wrong. The less popular the topic, the worse it got: 4 mistakes on Jobs, 12 on Tetris, 12 on BeOS. It named Clara Clley as Steve Jobs' mother, invented a colleague for Pajitnov named Geri Gulovik and priced the BeBox at $9,999. The real one cost $1,600. Outputs: Gemma4 26B A4B: 218 tok/s · 15.1s total · 45 facts · 5 mistakes DiffusionGemma 26B A4B: 763 tok/s · 3.7s total · 33 facts · 28 mistakes The reason is simple. DiffusionGemma throws 256 tokens on the screen at once and polishes them pass after pass until the text sounds smooth. Smooth is all it cares about: a fake name, date or number sounds just as smooth as a real one, so it stays. Regular Gemma4 meanwhile writes one word at a time and checks every new word against everything before it. Google says it themselves in the launch post: quality is lower, use regular Gemma 4 when facts matter.

atomic.chat

76,789 views • 2 months ago

On large scale frontier MoE models, we are seeing GB200 NVL72 FP4 is 10x more power efficient compared to H200 FP8 single node at certain interactivity (tok/s/user). We are excited to implement disagg prefill/multinode wide expert parallelism on h100/h200 to see what is the power efficient gains between GB200 & H200 once we have that implemented. The power efficiency will probably be reduced to 4-6x once we have that implemented on hopper, which is still pretty good! visit inferencemax dot ai for the latest nuanced results.

On large scale frontier MoE models, we are seeing GB200 NVL72 FP4 is 10x more power efficient compared to H200 FP8 single node at certain interactivity (tok/s/user). We are excited to implement disagg prefill/multinode wide expert parallelism on h100/h200 to see what is the power efficient gains between GB200 & H200 once we have that implemented. The power efficiency will probably be reduced to 4-6x once we have that implemented on hopper, which is still pretty good! visit inferencemax dot ai for the latest nuanced results.

SemiAnalysis

56,872 views • 10 months ago