Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Continuous-video agents (computer use, robotics, static scenes) burn compute re-ingesting pixels that didn't move. VLMaxxing teaches a frozen video VLM to skip the reruns. 54 fps perception on Gemma 4 26B, training-free, no accuracy drift. w/ JF Bastien (arXiv 2605.03351)

sdamico's profile picture

Sam D'Amico

33,664 subscribers

63,188 Aufrufe • vor 1 Monat •via X (Twitter)

BildungWissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

Part of a training video showing how to use Macedonian shooting on the move
1:00

Part of a training video showing how to use Macedonian shooting on the move

Mon Key

88,314 Aufrufe • vor 2 Jahren

MIT system combines computer vision w/soft robotics to delicately pack groceries. Full video:
1:06

MIT system combines computer vision w/soft robotics to delicately pack groceries. Full video:

MIT CSAIL

48,099 Aufrufe • vor 1 Jahr

Prime Intellect Day Saturday — March 14 (Pi Day) 4-11pm — The Melody of SF Talks on frontier model training, agentic RL, computer use, coding agents, diffusion LLMs, automating AI & science, robotics — and why the singularity will not be singular.
0:24

Prime Intellect Day Saturday — March 14 (Pi Day) 4-11pm — The Melody of SF Talks on frontier model training, agentic RL, computer use, coding agents, diffusion LLMs, automating AI & science, robotics — and why the singularity will not be singular.

Prime Intellect

88,640 Aufrufe • vor 3 Monaten

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!
0:52

Next mlx-vlm release will ship with continuous batching support on the server 🚀 What's coming: → Continuous batching — new requests join the active batch immediately, no waiting. Mixed image + text batches supported → OpenAI-compatible API — field-for-field match with mlx-lm, reasoning/content split for thinking models, tag-aware streaming → Multi-turn tool calling — full tool use support across streaming and non-streaming, works with Gemma4 and other templates → Vision feature caching — cache image embeddings across turns. Gemma4: 228x speedup, Qwen3.5: 23x on cache hit All running locally on Apple Silicon. Check our this demo running 4 concurrent requests (mixed image + text) to gemma-4-26B-A4B-IT by Google Gemma in bf16 using Pi + MLX-VLM server on my M3 Ultra. One of the requests ingests a 8K resolution image!

Prince Canuma

82,085 Aufrufe • vor 2 Monaten

This actually makes Gemma 4 26B-4A usable for a coding agent @ 72tk/s on my MacBook Pro M1 Max. This video is realtime, running completely locally.
0:57

This actually makes Gemma 4 26B-4A usable for a coding agent @ 72tk/s on my MacBook Pro M1 Max. This video is realtime, running completely locally.

Kyle Howells

117,577 Aufrufe • vor 10 Tagen

Excited to share the multiplayer features we built to make running computer use agents easy. Here's a quick video of Claude ordering pizza and setting up Claude Code on the same computer.
3:30

Excited to share the multiplayer features we built to make running computer use agents easy. Here's a quick video of Claude ordering pizza and setting up Claude Code on the same computer.

Spencer Kinney

12,688 Aufrufe • vor 1 Monat

🚨 One orchestrator. 10 parallel agents. 100+ tokens a second. All local. The Google Gemma team just dropped a MASSIVE demo for Gemma 4 26B. They built a concurrent workflow that lets the 26B model coordinate an entire team of sub-agents on your machine. Out of the box, the cookbook lets you run 10 parallel agents to: → Code an entire SVG art gallery in seconds → Translate text simultaneously → Generate ASCII art → Write parallel code Spinning up multi-agent systems locally has never looked this fast or this accessible. 100% free and open-source. repo link in 🧵↓
1:30

🚨 One orchestrator. 10 parallel agents. 100+ tokens a second. All local. The Google Gemma team just dropped a MASSIVE demo for Gemma 4 26B. They built a concurrent workflow that lets the 26B model coordinate an entire team of sub-agents on your machine. Out of the box, the cookbook lets you run 10 parallel agents to: → Code an entire SVG art gallery in seconds → Translate text simultaneously → Generate ASCII art → Write parallel code Spinning up multi-agent systems locally has never looked this fast or this accessible. 100% free and open-source. repo link in 🧵↓

Charly Wargnier

27,691 Aufrufe • vor 3 Tagen

8/ Computer Use Are A Sadness - 🎖️Engagement Award Exploiting computer use agents using LLM security vulnerabilities for fun and profit Hijack agents that try to access your computer
5:50

8/ Computer Use Are A Sadness - 🎖️Engagement Award Exploiting computer use agents using LLM security vulnerabilities for fun and profit Hijack agents that try to access your computer

Alex Reibman 🖇️

25,040 Aufrufe • vor 1 Jahr

A long time coming but new mlx-lm is here with better batching support in the server and Gemma 4. pip install -U mlx-lm Here is a video where a single M3 Ultra serves 5 opencode sessions with Gemma 4 26B that process ~130k tokens in ~1.5 minutes.
2:03

A long time coming but new mlx-lm is here with better batching support in the server and Gemma 4. pip install -U mlx-lm Here is a video where a single M3 Ultra serves 5 opencode sessions with Gemma 4 26B that process ~130k tokens in ~1.5 minutes.

Angelos Katharopoulos

66,095 Aufrufe • vor 2 Monaten

New Google Gemma 4 12B claims near-26B performance - we tested both! We ran both models locally on one RTX 4090 and gave each the same task: write a self-contained HTML5 canvas animation with real physics in one file without libraries. Three scenes - a Galton board, two blocks colliding off a wall, and a chaotic triple pendulum Outputs: Gemma 4 26B-A4B: 15 GB VRAM usage, 6.9k tokens, 138 tok/s Gemma 4 12B: 9 GB VRAM usage, 8.9k tokens, 80 tok/s Same Gemma 4 family, but the 26B-A4B won every scene and ran ~1.7x faster - on just 4B active params. The 12B stayed very close though, on almost half the VRAM - which makes it the ideal model for a 16 GB laptop
0:36

New Google Gemma 4 12B claims near-26B performance - we tested both! We ran both models locally on one RTX 4090 and gave each the same task: write a self-contained HTML5 canvas animation with real physics in one file without libraries. Three scenes - a Galton board, two blocks colliding off a wall, and a chaotic triple pendulum Outputs: Gemma 4 26B-A4B: 15 GB VRAM usage, 6.9k tokens, 138 tok/s Gemma 4 12B: 9 GB VRAM usage, 8.9k tokens, 80 tok/s Same Gemma 4 family, but the 26B-A4B won every scene and ran ~1.7x faster - on just 4B active params. The 12B stayed very close though, on almost half the VRAM - which makes it the ideal model for a 16 GB laptop

atomic.chat

150,374 Aufrufe • vor 18 Tagen

Our preprint on the Perception Test is now on arXiv: We will be benchmarking the cream of the crop multimodal video models in the Perception Test Challenge, happening in ICCV 2023. (1 / 2)
0:47

Our preprint on the Perception Test is now on arXiv: We will be benchmarking the cream of the crop multimodal video models in the Perception Test Challenge, happening in ICCV 2023. (1 / 2)

joao carreira

46,218 Aufrufe • vor 3 Jahren

Instead of asking a VLM to output progress, it reads the model’s internal belief directly from token logits. No in-context learning. No fine-tuning. No reward training. 📈 We introduce: TOPReward, a zero-shot reward modeling approach for robotics using token probabilities from pretrained video VLMs. The simplest way of doing reward modelling for robotics! Project: 🧵👇
1:09

Instead of asking a VLM to output progress, it reads the model’s internal belief directly from token logits. No in-context learning. No fine-tuning. No reward training. 📈 We introduce: TOPReward, a zero-shot reward modeling approach for robotics using token probabilities from pretrained video VLMs. The simplest way of doing reward modelling for robotics! Project: 🧵👇

Jiafei Duan

107,926 Aufrufe • vor 3 Monaten

Check out this amazing combo using Gemma4 + Falcon Perception for video tracking! 1️⃣Give Gemma 4 video frames 2️⃣It describes what it sees 3️⃣Falcon Perception takes those descriptions, segments the objects, and tracks them across the video! The best part? All running locally!
1:04

Check out this amazing combo using Gemma4 + Falcon Perception for video tracking! 1️⃣Give Gemma 4 video frames 2️⃣It describes what it sees 3️⃣Falcon Perception takes those descriptions, segments the objects, and tracks them across the video! The best part? All running locally!

Google Gemma

83,522 Aufrufe • vor 2 Monaten

Robotics fundamentally involves understanding the dynamics of how things change in the world in response to action and force. This is impossible to learn from static images; instead, it’s far more effective and more data-efficient to learn from video. Elvis Nava joins us to talk about Mimic Robotic. One of the key findings from mimic-video is that pretraining on webscale video allows robots to learn physics priors; as a result, policies train faster, generalize better, and are capable of more impressive dexterity, versus training on static images or image-language pairs as per a VLM. Watch Episode #81 of RoboPapers with Michael Cho - Rbt/Acc and Chris Paxton to learn more!
1:08:43

Robotics fundamentally involves understanding the dynamics of how things change in the world in response to action and force. This is impossible to learn from static images; instead, it’s far more effective and more data-efficient to learn from video. Elvis Nava joins us to talk about Mimic Robotic. One of the key findings from mimic-video is that pretraining on webscale video allows robots to learn physics priors; as a result, policies train faster, generalize better, and are capable of more impressive dexterity, versus training on static images or image-language pairs as per a VLM. Watch Episode #81 of RoboPapers with Michael Cho - Rbt/Acc and Chris Paxton to learn more!

RoboPapers

46,190 Aufrufe • vor 1 Monat

NEW FREE Computer Use AI Agents are INSANE 🤯
14:34

NEW FREE Computer Use AI Agents are INSANE 🤯

Julian Goldie SEO

49,308 Aufrufe • vor 1 Jahr

Deepseek Computer Use AI Agents are INSANE (FREE!) 🤯
12:12

Deepseek Computer Use AI Agents are INSANE (FREE!) 🤯

Julian Goldie SEO

91,135 Aufrufe • vor 1 Jahr

Claude Code is now completely free forever 🤯 Google just dropped Gemma 4 on Ollama, which means you can run Claude Code locally with zero API bills and zero subscriptions. Just install Gemma 4, connect it to Claude Code, and suddenly you have free AI agents running directly on your computer. Google may have just killed AI subscription costs. 🚀
0:28

Claude Code is now completely free forever 🤯 Google just dropped Gemma 4 on Ollama, which means you can run Claude Code locally with zero API bills and zero subscriptions. Just install Gemma 4, connect it to Claude Code, and suddenly you have free AI agents running directly on your computer. Google may have just killed AI subscription costs. 🚀

Julian Goldie SEO

605,956 Aufrufe • vor 2 Monaten

Watch Gemma-4-E4B casually identify sea animals by classifying images in a single agentic session using its vision capabilities. (impressive for a 4B model 🚀) I'm convinced: the agents of tomorrow are local, free, fast, and run on every computer!
0:25

Watch Gemma-4-E4B casually identify sea animals by classifying images in a single agentic session using its vision capabilities. (impressive for a 4B model 🚀) I'm convinced: the agents of tomorrow are local, free, fast, and run on every computer!

Victor M

171,280 Aufrufe • vor 2 Monaten

no na's Christy teaches us how to do her iconic jumpalit move in a new video.
1:28

no na's Christy teaches us how to do her iconic jumpalit move in a new video.

About Music

19,226 Aufrufe • vor 4 Monaten