Video yükleniyor...

Video Yüklenemedi

Bu video yüklenirken bir sorun oluştu. Bu geçici bir ağ sorunundan kaynaklanıyor olabilir veya video kullanılamıyor olabilir.

Ana Sayfaya Dön

Training AI agents takes more than fast token generation. Tools run. Tests stall. Training batches wait. We cut fixed-weight batch collection time by 66.9% on a Terminal-Bench-based workload by improving routing, sandbox execution and scheduling. Here’s how:

Nebius Token Factory

13,217 subscribers

13,368 görüntüleme • 1 ay önce •via X (Twitter)

Bilim & Teknoloji

Anya Rossi• Live Now

Private livecam show

3 Yorum

Curious Nori | AI · Food · Travel profil fotoğrafı

Curious Nori | AI · Food · Travel1 ay önce

Big fan of this.

bilal p profil fotoğrafı

bilal p1 ay önce

Looks like you just turned the AI’s coffee break into a sprint, and the training queue finally stopped napping.

George Cooper profil fotoğrafı

George Cooper1 ay önce

Yup, makes sense! Cutting down processing time can have a huge impact on the overall workflow, glad you guys were able to optimize things!.

Benzer Videolar

how I run Vercel AI SDK agents in a cloud sandbox agents have full filesystem, bash and git access takes 2 minutes

how I run Vercel AI SDK agents in a cloud sandbox agents have full filesystem, bash and git access takes 2 minutes

Josh tried coding

24,803 görüntüleme • 7 ay önce

OpenAI intern attempting to return rogue AI agents to the sandbox during weekend training run (2026, colorized)

OpenAI intern attempting to return rogue AI agents to the sandbox during weekend training run (2026, colorized)

Ed Leon Klinger

40,173 görüntüleme • 1 ay önce

Agents on Solana are running continuous, real-time decision loops. To power this new workload, Sonic launches North Star: Instant private execution sessions for real-time agents. Every Solana agent and dApp gets its own isolated execution session. Run fast, settle back to Solana.

Agents on Solana are running continuous, real-time decision loops. To power this new workload, Sonic launches North Star: Instant private execution sessions for real-time agents. Every Solana agent and dApp gets its own isolated execution session. Run fast, settle back to Solana.

Sonic SVM

36,337 görüntüleme • 4 ay önce

Goodbye manus The next generation of ai agents wont be judged by how smart they are they will be judged by ROTI, return on token investment AllyHub just turned that into a real metric, the same task costing less every single time you run it A thread on the agent that gets cheaper the more you use it

Goodbye manus The next generation of ai agents wont be judged by how smart they are they will be judged by ROTI, return on token investment AllyHub just turned that into a real metric, the same task costing less every single time you run it A thread on the agent that gets cheaper the more you use it

Nawi

56,327 görüntüleme • 5 ay önce

An orthopedic surgeon and a performance coach developed a resistance training routine that protects and strengthens knees by improving the body’s ability to move. Here’s how to do these exercises:

An orthopedic surgeon and a performance coach developed a resistance training routine that protects and strengthens knees by improving the body’s ability to move. Here’s how to do these exercises:

The Washington Post

207,161 görüntüleme • 19 gün önce

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding them.:

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding them.:

Microsoft Research

18,815 görüntüleme • 1 ay önce

Long video generation is a systems problem. Introducing LongLive-2.0 from NVIDIA Research: an end-to-end NVFP4 training and inference system for long video generation. Low-precision deployment often relies on post-training quantization, creating a gap between how models are trained and how they run. LongLive-2.0 aligns NVFP4-aware training, distillation, and W4A4 inference, maintaining strong benchmark quality while improving speed and memory efficiency.

Long video generation is a systems problem. Introducing LongLive-2.0 from NVIDIA Research: an end-to-end NVFP4 training and inference system for long video generation. Low-precision deployment often relies on post-training quantization, creating a gap between how models are trained and how they run. LongLive-2.0 aligns NVFP4-aware training, distillation, and W4A4 inference, maintaining strong benchmark quality while improving speed and memory efficiency.

NVIDIA AI

60,715 görüntüleme • 5 ay önce

It’s velocity based training… Not fast-based training Vitruve

It’s velocity based training… Not fast-based training Vitruve

Cody Hughes

19,656 görüntüleme • 1 yıl önce

Open-weight AI models are getting seriously capable. IQuest-Q1 just dropped, and it’s built for more than generating code. → ~320B total parameters, ~15B active per token → Strong results on NL2Repo and CyberGym → Competitive on Terminal-Bench 2.1 and Agents’ Last Exam → Can build interactive 3D apps and games from natural language → Can use tools, inspect logs, modify code, and verify fixes → Handles long-horizon, multi-step tasks across real work environments The wildest part? IQuest-Q1 helps diagnose capabilities, design training steps, and execute parts of R&D. Open weights. Open research. This feels less like “another LLM” and more like a glimpse at AI that can actually do the work. Technical Blog： Hugging Face： GitHub：

Open-weight AI models are getting seriously capable. IQuest-Q1 just dropped, and it’s built for more than generating code. → ~320B total parameters, ~15B active per token → Strong results on NL2Repo and CyberGym → Competitive on Terminal-Bench 2.1 and Agents’ Last Exam → Can build interactive 3D apps and games from natural language → Can use tools, inspect logs, modify code, and verify fixes → Handles long-horizon, multi-step tasks across real work environments The wildest part? IQuest-Q1 helps diagnose capabilities, design training steps, and execute parts of R&D. Open weights. Open research. This feels less like “another LLM” and more like a glimpse at AI that can actually do the work. Technical Blog： Hugging Face： GitHub：

Atul Kumar

79,465 görüntüleme • 1 ay önce

What if Otto (and other agents) could run predictive analysis on any coin/token? Here’s a quick screen video of our devs experimenting during the holidays with a predictive analysis tool (NFA) — powered by AI Agents like Otto, created via Hybrid's agent launcher.

What if Otto (and other agents) could run predictive analysis on any coin/token? Here’s a quick screen video of our devs experimenting during the holidays with a predictive analysis tool (NFA) — powered by AI Agents like Otto, created via Hybrid's agent launcher.

Hybrid

38,289 görüntüleme • 1 yıl önce

Run system commands and AI coding agents directly inside MATLAB desktop and Simulink Canvas ➡️ - Run CLI tools without leaving MATLAB and Simulink - Launch your AI coding agents from a terminal docked in MATLAB or Simulink. - Free and open source

Run system commands and AI coding agents directly inside MATLAB desktop and Simulink Canvas ➡️ - Run CLI tools without leaving MATLAB and Simulink - Launch your AI coding agents from a terminal docked in MATLAB or Simulink. - Free and open source

MATLAB

13,639 görüntüleme • 4 ay önce

"Fast 4D Mesh Generation by Spatio-Temporal Attention Chains" TL;DR: training-free spatio-temporal attention chains generate topology-consistent 4D meshes from video 13× faster while improving temporal correspondence quality

"Fast 4D Mesh Generation by Spatio-Temporal Attention Chains" TL;DR: training-free spatio-temporal attention chains generate topology-consistent 4D meshes from video 13× faster while improving temporal correspondence quality

Alexandre Morgand

13,807 görüntüleme • 5 ay önce

This is how we get more controllers in the towers ✈️⬇️ The FAA ✈️ is sending new technology to Sioux Falls Regional Airport to provide realistic training scenarios which will cut training time by almost 30%❗️ Dakota News Now

This is how we get more controllers in the towers ✈️⬇️ The FAA ✈️ is sending new technology to Sioux Falls Regional Airport to provide realistic training scenarios which will cut training time by almost 30%❗️ Dakota News Now

U.S. Department of Transportation

17,198 görüntüleme • 1 ay önce

Training the next the generation to run faster than their problems

Training the next the generation to run faster than their problems

Morara Kebaso

21,063 görüntüleme • 8 ay önce

🚀 Where does intelligence reside? The paradigm in modern AI suggests: scale up a single model, train it on more data, and general intelligence may emerge. Yet Society of Mind proposed another possibility: intelligence may emerge from interactions of less intelligent agents. As we keep scaling up a few generalist agents shared by everyone, can we scale up societies of decentralized agents that work together and collectively achieve more than any one of them could alone? We introduce MPT: Market-based Post-Training, a framework to train such societies of agents by jointly learning agent policies and self-orchestration in a decentralized manner.

🚀 Where does intelligence reside? The paradigm in modern AI suggests: scale up a single model, train it on more data, and general intelligence may emerge. Yet Society of Mind proposed another possibility: intelligence may emerge from interactions of less intelligent agents. As we keep scaling up a few generalist agents shared by everyone, can we scale up societies of decentralized agents that work together and collectively achieve more than any one of them could alone? We introduce MPT: Market-based Post-Training, a framework to train such societies of agents by jointly learning agent policies and self-orchestration in a decentralized manner.

Guowei Xu

37,387 görüntüleme • 17 gün önce

"It was about execution. It was about mental toughness." ⠀ Coach Christie takes us inside the huddle between drills on the first day of Kings Training Camp. ⠀ Training Camp presented by Sutter Health

"It was about execution. It was about mental toughness." ⠀ Coach Christie takes us inside the huddle between drills on the first day of Kings Training Camp. ⠀ Training Camp presented by Sutter Health

Sacramento Kings

13,296 görüntüleme • 1 ay önce

Autonomous agents + tools = new risks OpenClaw agents are powerful, but they could also leak data and burn through API spend fast. Here’s how to run OpenClaw🦞 with Venice for more private inference and lower, predictable costs: (+ plus extra resources 🧵)

Autonomous agents + tools = new risks OpenClaw agents are powerful, but they could also leak data and burn through API spend fast. Here’s how to run OpenClaw🦞 with Venice for more private inference and lower, predictable costs: (+ plus extra resources 🧵)

Venice

11,735 görüntüleme • 9 ay önce

Today we're announcing cua-bench: a framework for benchmarking, training data, and RL environments for computer-use AI agents. Why? Current agents show 10x variance across minor UI changes. Here's how we're fixing it.

Today we're announcing cua-bench: a framework for benchmarking, training data, and RL environments for computer-use AI agents. Why? Current agents show 10x variance across minor UI changes. Here's how we're fixing it.

Cua

189,823 görüntüleme • 10 ay önce

Ralf #Rangnick - Training Methods Explaining his training methods and use of exaggeration & provocational rules ” We want to train in a way that makes the training even harder than the actual game by limiting time, space and learn under pressure ”

Ralf #Rangnick - Training Methods Explaining his training methods and use of exaggeration & provocational rules ” We want to train in a way that makes the training even harder than the actual game by limiting time, space and learn under pressure ”

Albin Sheqiri

68,394 görüntüleme • 3 yıl önce

New #NVIDIA Paper We introduce Motive, a motion-centric, gradient-based data attribution method that traces which training videos help or hurt video generation. By isolating temporal dynamics from static appearance, Motive identifies which training videos shape motion in video generation. 🔗 1/10

New #NVIDIA Paper We introduce Motive, a motion-centric, gradient-based data attribution method that traces which training videos help or hurt video generation. By isolating temporal dynamics from static appearance, Motive identifies which training videos shape motion in video generation. 🔗 1/10

Xindi Wu

145,942 görüntüleme • 9 ay önce