AllyshipContent Radar

News 2877

Chủ đề cộng đồng đang thảo luận

Crawl live-1795119300000 · 03:15 20 thg 11, 2026 · Xuất bản 03:31 20 thg 11, 2026

203 / 2877 chủ đề đạt ngưỡng 55 trong khoảng này.

21

PolyAI Dialog-RSN-1 is now on Amazon SageMaker JumpStart

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Helen Greul 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Dialog-RSN-1, a reasoning model built for real-time complex dialog, is available today on AWS Marketplace and Amazon SageMaker JumpStart. It powers frontier dialog: agents that hear a live call, reason about it, and act in real time.</p> <p>As an AWS customer, you can <a href="https://aws.amazon.com/marketplace/pp/prodview-l4qyhvfxnkq7e" target="_blank" rel="noopener">subscribe to Dialog-RSN-1 through AWS Marketplace</a> and deploy it as a SageMaker endpoint inside your own AWS environment. You can also deploy directly from Amazon SageMaker Studio following the steps in <a href="https://aws.amazon.com/blogs/machine-learning/deploy-and-evaluate-models-with-amazon-sagemaker-jumpstart/" target="_blank" rel="noopener">Deploy and evaluate models with Amazon SageMaker JumpStart</a>. That means the model designed to handle your customer conversations can be deployed like other Amazon SageMaker models with an AWS Marketplace subscription.</p> <h2 id="deploying-dialog-rsn-1-inside-aws">Deploying Dialog-RSN-1 inside AWS</h2> <p>Dialog-RSN-1 is the first PolyAI model available through <a href="https://aws.amazon.com/marketplace/pp/prodview-l4qyhvfxnkq7e" target="_blank" rel="noopener">AWS Marketplace</a>. It runs inside your own AWS environment on the same AWS cloud that powers PolyAI’s compute and model training and serving.</p> <p>All in a single pass, Dialog-RSN-1 perceives a caller’s audio directly, reasons about what they need, and responds in real time. On SageMaker, it deploys as an endpoint like other models you run there. It is reachable from your own architecture, billed through your AWS account, and covered by the same virtual private cloud (VPC), AWS Identity and Access Management (IAM) roles, and monitoring you already have in place.</p> <p>You keep ownership of orchestration, telephony integration, and business logic. Dialog-RSN-1 takes off your plate the hardest part to build yourself: understanding a live call and responding to it in real time. The endpoint speaks the OpenAI Realtime protocol, so an existing Realtime client connects through PolyAI’s <a href="https://github.com/polyai/dialog-rsn-1-sagemaker" target="_blank" rel="noopener">bridge</a> by changing its base URL. Audio in is 16-bit PCM mono at 16 or 24 kHz (transcode G.711 or Opus first). Connections last up to 30 minutes, and the model is English-only today.</p> <p>To get started, <a href="https://aws.amazon.com/marketplace/pp/prodview-l4qyhvfxnkq7e" target="_blank" rel="noopener">find Dialog-RSN-1 on AWS Marketplace</a>, then deploy it straight into an Amazon SageMaker endpoint in your own account. Or you can deploy directly from <a href="https://aws.amazon.com/sagemaker/ai/jumpstart/" target="_blank" rel="noopener">Amazon SageMaker JumpStart</a>.</p> <p>With Dialog-RSN-1 now onboarded to SageMaker JumpStart, deploying Dialog-RSN-1 takes one click. From the Amazon SageMaker AI console, open the Model Hub, select Dialog-RSN-1 from the JumpStart catalog, and choose Deploy. SageMaker JumpStart then spins up a real-time inference endpoint on a serving stack it has already tuned for you. No containers to wrangle, no instance types to second-guess, and no manual benchmarking. For automated workflows, the same one-click deployment also runs in a few lines of the SageMaker Python SDK.</p> <p>The Dialog-RSN-1 endpoint runs entirely within your own AWS account. It is deployed in your VPC, accessed through your IAM roles, and monitored with the same Amazon CloudWatch logging and metrics you use for every other workload. Because inference happens inside your account, your callers’ audio and transcripts stay within a boundary you control. For a model handling live customer conversation, this helps support your data residency and auditability requirements.</p> <h2 id="built-for-speed-and-accuracy">Built for speed and accuracy</h2> <p>Models built for live calls may make you choose between speed and understanding. And models built for speed may miss the meaning, and ones built for reasoning may lead to wait time for callers. Dialog-RSN-1 avoids the tradeoff.</p> <p>With PolyAI, you can build dialog agents that hear and handle calls the way your human agents would. These agents catch hesitation, hear frustration before it escalates, and know when to speak.</p> <p>Real-time customer conversation is the job PolyAI’s Dialog-RSN-1 is built for, and it is already handling live calls in production today. Because the model reasons directly over the caller’s audio, it picks up tone, hesitation, and timing, the signals a plain transcript loses, and decides for itself the right moment to respond.</p> <h2 id="a-model-that-hears-calls-the-way-humans-do">A model that hears calls the way humans do</h2> <p>Dialog-RSN-1 is a model focused on real-time conversations. There is no separate speech-to-text step and no reasoning model on the appropriate moment to talk, so thereby no multiple handoffs.</p> <p>In practice, that means the callers get responses faster:</p> <ul> <li>Barge-in keeps working even when call volume spikes, instead of getting switched off to protect performance.</li> <li>Responses keep pace, so the model is not lagging behind the conversation while it reasons through what to say.</li> <li>Transcription is produced from the audio itself, because Dialog-RSN-1 is reading the actual audio, not someone else’s best guess at it.</li> </ul> <h2 id="dialog-rsn-1-on-amazon-sagemaker-jumpstart">Dialog-RSN-1 on Amazon SageMaker JumpStart</h2> <p>Amazon SageMaker JumpStart is the machine learning hub inside Amazon SageMaker AI, and it exists to collapse the distance between choosing a model and running it in production. For Dialog-RSN-1, that means the frontier of real-time dialog is a few clicks away. You can discover it, prove it on your own data, and ship it to a secure endpoint, all in one place. Here is what you get.</p> <div style="width: 810px" class="wp-caption alignnone"> <a href="https://d2908q01vomqb2.cloudfront.net/f1f836cb4ea6efb2a

23

Read more in our launch blog, so excited to see where you all take this: https://blog.google/innovation-and-ai/technology/developers-tools/genie-4-in-gemini-api/

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

officiallogank 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Read more in our launch blog, so excited to see where you all take this: https://blog.google/innovation-and-ai/technology/developers-tools/genie-4-in-gemini-api/

24

Introducing Genie 4 in the Gemini API + @GoogleAIStudio, our first world model for developers to build with and create new world model experiences 🔮

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

officiallogank 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Introducing Genie 4 in the Gemini API + @GoogleAIStudio, our first world model for developers to build with and create new world model experiences 🔮 Try it out: https://ai.studio/genie

25

Today we are rolling out Genie 4 on the Gemini API and AI Studio🚀

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

_philschmid 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Today we are rolling out Genie 4 on the Gemini API and AI Studio🚀 Seed a 720p 24fps world from text or image and steer it live with WASD, prompts, and 48 kHz stereo audio. Stream via WebRTC (browser + ephemeral tokens) or WebSocket. Available on: - Gemini API: https://aistudio.google.com/docs/genie - AI Studio: https://aistudio.google.com/live?model=genie-4

26

b12547

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

github-actions[bot] 1 bài viết · ggml-org/llama.cpp releases
Nguyên văn nguồn
Chi tiết nguồn

<details open=""> <p>webgpu: add POOL_2D support (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6387361702" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32250" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32250/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32250">#32250</a>)</p> <ul> <li> <p>Shader, pipeline, and launcher</p> </li> <li> <p>Update docs and add newline to wgsl</p> </li> <li> <p>Formatting fixes</p> </li> </ul> </details> <p><strong>Website:</strong></p> <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></li> </ul> <p><strong>Attestations:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/attestations/69756852">https://github.com/ggml-org/llama.cpp/attestations/69756852</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI enabled) <a href="https://github.com/ggml-org/llama.cpp/pull/23780" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/23780/hovercard">DISABLED</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-xcframework.zip">iOS XCFramework</a></li> </ul> <p><strong>Linux:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-26.04-s390x.tar.gz">Ubuntu 26.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-zdnn-s390x.tar.gz">Ubuntu 26.04 s390x (zDNN)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-cuda-12.8-x64.tar.gz">Ubuntu x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/cudart-llama-b12547-bin-ubuntu-cuda-12.8-x64.tar.gz">CUDA 12.8 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-cuda-13.5-x64.tar.gz">Ubuntu x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/cudart-llama-b12547-bin-ubuntu-cuda-13.5-x64.tar.gz">CUDA 13.5 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-cuda-13.5-arm64.tar.gz">Ubuntu arm64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/cudart-llama-b12547-bin-ubuntu-cuda-13.5-arm64.tar.gz">CUDA 13.5 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-rocm-10.0-x64.tar.gz">Ubuntu x64 (ROCm 10.0)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-openvino-2026.4.1-x64.tar.gz">Ubuntu x64 (OpenVINO)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-sycl-fp32-x64.tar.gz">Ubuntu x64 (SYCL FP32)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-ubuntu-sycl-fp16-x64.tar.gz">Ubuntu x64 (SYCL FP16)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-linux-arm64-snapdragon.tar.gz">Linux arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)</a> - <a href="https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/linux.md">setup guide</a></li> </ul> <p><strong>Android:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-android-arm64.tar.gz">Android arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-android-arm64-snapdragon.tar.gz">Android arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)</a> - <a href="https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/README.md">setup guide</a></li> </ul> <p><strong>Windows:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-win-cpu-x64.zip">Windows x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-win-cpu-arm64.zip">Windows arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-win-opencl-adreno-arm64.zip">Windows arm64 (OpenCL Adreno)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-win-cuda-12.4-x64.zip">Windows x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/cudart-llama-bin-win-cuda-12.4-x64.zip">CUDA 12.4 DLLs</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/llama-b12547-bin-win-cuda-13.5-x64.zip">Windows x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12547/cudart-llama-bin-win-cuda-13.5-x64.zip">CUDA

  • b12547 ggml-org/llama.cpp releases · 20 thg 11, 2026
27

Let's goooo! Super excited to announce our partnership with @Replit & @amasad for our Ramadan Global Buildathon!

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

arblauvelt 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Let's goooo! Super excited to announce our partnership with @Replit & @amasad for our Ramadan Global Buildathon! Limited to 1,000 teams - already past the 100 mark. Learn more & register today: http://Build.LaunchGood.com ps - would it be wrong to participate in my own Buildathon? I have an idea for a physical product 😅 Media

28

For a while I have been obsessed with the idea that there is a fundamental difference between "context" as we have grown to understand it and "knowledge", which is what we actually prize and desire an

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

isidoremiller 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

For a while I have been obsessed with the idea that there is a fundamental difference between "context" as we have grown to understand it and "knowledge", which is what we actually prize and desire and what translates into real-world outcomes. It's become clear to me that as the volume and scope of work delegated to AI grows, legacy systems of context management will not be able to scale cleanly, no matter how much we try to use AI to speed things up. In domains that are not pure code, it is an asymmetric race where the slop generators are always outpacing the context curators. There is no way to break this formation. Unless you shift your perspective and move up a plane. Legacy static context bootstraps synthesized knowledge. Knowledge grows organically and dynamically with rich environmental feedback. It becomes "real" in a sense that legacy context is often not and so can be evaluated and verified and improved and maintained in a truly autonomous and scalable fashion. Data teams and curators do not lose control in this vision of the world; they develop increased leverage, their signal-to-noise ratio improves, and everyone (and their agents) coalesces around the outcomes that matter —knowledge— instead of intermediate busywork. This is a long-term utopian vision for what data work (maybe all knowledge work!) might become, and there is a LOT of work to do, but I'm extremely proud of Hex for taking the first bold steps towards this with our new Knowledge Engine. It's live now for anyone using Hex 4.0. Today, it will immediately improve the accuracy and efficiency of your data work. Tomorrow when you wake up, it will be even better. It will not go stale or fall out of sync with reality. Give it a little breadcrumb, and it will reconcile the entire workspace to fit your desires. "Leverage, says you. I think I feel a change in the wind, says I!" https://hex.tech/blog/knowledge-engine/

29

Into the Omniverse: How Developers Turn Ideas Into Simulations With Frontier AI Agents

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

NVIDIA Writers 1 bài viết · NVIDIA Blog
Nguyên văn nguồn
Chi tiết nguồn

Turning a simulation idea into a working application means assembling assets, connecting physics and rendering, and checking that the scene behaves as intended. Developers are combining frontier AI models with NVIDIA Omniverse libraries to help carry out that work — building applications for exploring scenarios, investigating failures and improving designs. Developers direct AI agents through natural-language instructions, review results and guide changes. Omniverse libraries provide GPU-accelerated physics, rendering and sensor simulation capabilities. Explore the projects below to see frontier AI models such as GPT-6 Astra at work, and check back for new examples from NVIDIA teams and developers across the ecosystem. Physical AI Paves the Path to Autonomous Industrial Operations https://blogs.nvidia.com/wp-content/uploads/2026/11/Nyhamna_footage_take5_30s.mp4 AI agents powered by frontier models pave the way to autonomous operations. They use industrial tools and engineering data to help engineers build digital twins that test plans and processes and evaluate operational decisions. Falkor, a KONGSBERG company, shows how this works in practice. The company’s Falkor Surface platform lets customers turn their computer-aided design and engineering data into facility-scale digital twins using the GPT-6 Astra model. The platform uses NVIDIA Omniverse libraries and NVIDIA RTX PRO 6000 Server Edition GPUs on Microsoft Azure to enable advanced physics simulation. AI agents built on Falkor’s agentic engine understand dense technical documentation, spatial relationships and real-world physics. Engineers can create and test plans in simulation before traveling on site, reducing costly delays and mistakes. They can also inspect equipment and check clearances in a shared view. Read Falkor’s article and watch the livestream f or a closer look at how the company builds these tools and checks the agents’ work. Build a Robot Pick-and-Place Simulation With SimReady Assets Getting a robot model into a scene is only the start. Developers also need to assemble its components, add controls and check whether it can complete a task. NVIDIA developers Dillon Bailey and Rishabh Chadha used 3D assets from the SimReady Central library to build a pick-and-place simulation of a robot moving a can using the NVIDIA Isaac Sim framework. They directed an agent equipped with GPT-6 Sol and Isaac Sim skills to assemble the scene and generate pick-and-place controls. Testing the assembled scene shows how the assets work together, from joint movement and contact behavior to the gripper’s ability to hold, move and release the can. With that baseline in place, developers can adjust object positions and motion paths and repeat the tests to evaluate how each change affects the task. Follow the walkthrough to assemble and test a pick-and-place simulation in Isaac Sim, with sample agent prompts and checks for robot movement and grasp behavior. Build a Voice-Controlled VR Simulation With SimReady Assets Bringing 3D assets into virtual reality (VR) requires more than rendering a scene. Objects need to respond to controllers and behave as expected when grabbed, moved or released. NVIDIA Omniverse community developer Michael Wagner took on this challenge. He directed GPT-6 Astra to use Blender software to model a study environment from two reference images, balancing visual quality with efficient performance in VR. He then used Claude Code to build an application that lets users find SimReady Central assets by voice and interact with them in an immersive VR scene. The Claude Code agent worked for about eight hours, and Wagner built the application over a weekend using an NVIDIA RTX 6000 Ada workstation and a Meta Quest 3 VR headset. The application ran on NVIDIA Omniverse libraries : ovstage shared the OpenUSD scene, ovrtx rendered it, ovxr enabled VR and ovphysx ran physics. Wagner later built a second version with ovnewton , which integrates with ovstage, to compare the physics libraries. The biggest challenge for Wagner was checking what happened inside the headset. Automated tests missed problems he could feel, including objects shaking when grabbed. Wagner reported the symptoms, guided the agent through changes and verified the interaction himself. Updating the grab control once per frame resolved the shaking. His takeaway: Give agents renders and logs to inspect, then check the experience directly. Explore his open source demo f or build instructions, a starting prompt and reusable agent skills. Send Microducks on Rounds in a Simulated Hospital https://blogs.nvidia.com/wp-content/uploads/2026/11/microducks-icu-16x9-1920x1080-new.mp4 Before adapting a robot to a new environment, developers need to test whether its walking policy — the AI that controls how it moves — can keep it stable in that setting. AI developer Maziyar Panahi directed Claude Code, using Claude Opus 5.5, to assemble an intensive care unit from 13 NVIDIA SimReady hospital assets and test Microduck robots inside it. The assets came with Newton physics settings, so Panahi didn’t have to model the furniture or create collision geometry from scratch. The Newton physics engine also helped replicate the ward so hundreds of copies could run together on one NVIDIA RTX PRO 6000 GPU. When the cloud containers lacked graphics support, Panahi and the Claude Code agent separated simulation from rendering, recording robot poses in the cloud and rendering them locally. They also treated the furniture as static collision objects, which enabled larger runs within the solver’s limits. The robots walked using AI they had already been trained on and followed preset points along a route. Panahi checked how many robots remained upright: In a 256-ward run, 766 of 768 Microducks were still standing after 14 simulated seconds. He said the agent’s work took about two hours. Explore the open source demo for a starter prompt, checks and instructions to begin with one ward on a CPU, then sc

30

Enabling Real-Time AI for Scientific Instruments with GPU-Native Data Acquisition

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Cara Laasch 1 bài viết · NVIDIA Developer Blog
Nguyên văn nguồn
Chi tiết nguồn

In science and industry, analyzing sensor measurements in real time can enable AI to adjust processes, guide experiments, and search for better operating... In science and industry, analyzing sensor measurements in real time can enable AI to adjust processes, guide experiments, and search for better operating conditions. Yet limits on how quickly the measured data can be moved and processed often shape how an instrument is designed, from readout electronics to acquisition timing. What could scientific instrument makers simplify—or make possible—if… Source

31

b12546

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

github-actions[bot] 1 bài viết · ggml-org/llama.cpp releases
Nguyên văn nguồn
Chi tiết nguồn

<details open=""> <p>llama : remove redundant fields from llm_graph_params (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6486857483" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32596" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32596/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32596">#32596</a>)</p> <p>llm_graph_params no longer carries hparams, prec_policy and hdmd, since the model reference already provides them. llm_graph_context sources them from the model and keeps prec_policy as a reference.</p> <p>Assisted-by: dsh:llama.cpp/ggml-org/DeepSeek-V4.1-Flash-1017-GGUF (Mac Studio, M5 Ultra, 256GB)</p> </details> <p><strong>Website:</strong></p> <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></li> </ul> <p><strong>Attestations:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/attestations/69749431">https://github.com/ggml-org/llama.cpp/attestations/69749431</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI enabled) <a href="https://github.com/ggml-org/llama.cpp/pull/23780" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/23780/hovercard">DISABLED</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-xcframework.zip">iOS XCFramework</a></li> </ul> <p><strong>Linux:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-26.04-s390x.tar.gz">Ubuntu 26.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-zdnn-s390x.tar.gz">Ubuntu 26.04 s390x (zDNN)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-cuda-12.8-x64.tar.gz">Ubuntu x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/cudart-llama-b12546-bin-ubuntu-cuda-12.8-x64.tar.gz">CUDA 12.8 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-cuda-13.5-x64.tar.gz">Ubuntu x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/cudart-llama-b12546-bin-ubuntu-cuda-13.5-x64.tar.gz">CUDA 13.5 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-cuda-13.5-arm64.tar.gz">Ubuntu arm64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/cudart-llama-b12546-bin-ubuntu-cuda-13.5-arm64.tar.gz">CUDA 13.5 libraries</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-rocm-10.0-x64.tar.gz">Ubuntu x64 (ROCm 10.0)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-openvino-2026.4.1-x64.tar.gz">Ubuntu x64 (OpenVINO)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-sycl-fp32-x64.tar.gz">Ubuntu x64 (SYCL FP32)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-ubuntu-sycl-fp16-x64.tar.gz">Ubuntu x64 (SYCL FP16)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-linux-arm64-snapdragon.tar.gz">Linux arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)</a> - <a href="https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/linux.md">setup guide</a></li> </ul> <p><strong>Android:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-android-arm64.tar.gz">Android arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-android-arm64-snapdragon.tar.gz">Android arm64 (Snapdragon: CPU, Adreno GPU, Hexagon NPU)</a> - <a href="https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/snapdragon/README.md">setup guide</a></li> </ul> <p><strong>Windows:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-win-cpu-x64.zip">Windows x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-win-cpu-arm64.zip">Windows arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-win-opencl-adreno-arm64.zip">Windows arm64 (OpenCL Adreno)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/llama-b12546-bin-win-cuda-12.4-x64.zip">Windows x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12546/cudart-llama-bin-win-cuda-12.4-x64.zip">CUDA 12.4 DLLs</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b1254

  • b12546 ggml-org/llama.cpp releases · 19 thg 11, 2026
32

recommendrd viewing, even if i think the amount of unsubstantiated claims and extrapolations is too damn high. still fun to listen to and think why what is claimed may or may not happen.

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

badlogicgames 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

recommendrd viewing, even if i think the amount of unsubstantiated claims and extrapolations is too damn high. still fun to listen to and think why what is claimed may or may not happen. https://youtu.be/1NHPDzpfR6Y?is=cGj8ducdmD_Z4FO1

33

Show HN: AgentMall – Let your AI agent buy groceries

64

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 19 tương tác đã đo · Nguồn crawl

desilva78_ 1 bài viết · HN: GPT
Nguyên văn nguồn
Chi tiết nguồn

Hey HN, we’re Nel and Kosi. We noticed personal agents like Instinct, Muse were terrible at buying stuff. Most of the time the checkout would fail. So we built reliable automations for agents to checkout at 9 leading grocery merchants. We grapple with anti bot, logins, OTP codes, and changing checkout structures. Loved revisiting this stuff given our backgrounds sneaker-botting and building computer use models. Honestly full circle haha. It currently works on these sites with ~70% checkout success rate: - Walmart - Albertsons - Ocado - Wegmans - FreshDirect - Best Buy - Instacart Our goal is to hill climb to 90%+ across each. Test keys are free and run the real flow without placing an order. New accounts get 10 free completed orders. https://docs.getagentmall.com/quickstart

34

Introducing enhanced infrastructure observability in Amazon SageMaker HyperPod

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Mijanur Palash 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Large-scale model training is one of the most expensive workloads you can run on GPUs, and a single unhealthy accelerator can stall an entire distributed job. When training throughput drops, the hard part is rarely noticing that something is wrong. It is figuring out why and where. Teams often spend hours or days sifting through logs and disconnected dashboards to decide whether a slowdown came from a throttled GPU, a saturated network link, or a stalled data pipeline on the host.</p> <p>Today, we are announcing enhanced observability in Amazon SageMaker HyperPod, a new capability that helps you diagnose GPU training issues faster. It brings together diagnostic and profiling capabilities for training workloads that run on both Amazon Elastic Kubernetes Service (Amazon EKS) and Slurm orchestrators:</p> <p><strong>Diagnostic summary</strong> uses Amazon CloudWatch’s DetectGroupOutliers API to compare each GPU in a training job on utilization, memory, temperature, power, and SM clock, and flags GPUs that deviate from their peers. Static diagnostic rules surface common failure patterns such as GPU hardware faults (XID errors, ECC memory errors), EFA throughput drops, thermal throttling, and overheating, so you catch degradation before it affects training.</p> <h2 id="why-you-need-it">Why you need it</h2> <p>Distributed training is expensive, and small inefficiencies compound into significant wasted GPU hours. Until now, getting visibility into HyperPod infrastructure meant stitching together separate tools, and even then, the metrics told you that performance had degraded without telling you why.</p> <p>Enhanced observability closes three gaps. It isolates a single bad accelerator across hundreds or thousands of GPUs whether through XID errors, ECC memory faults, or thermal throttling, so you spend time on the fix instead of the search.</p> <p>Together, these capabilities reduce troubleshooting time and help you keep expensive accelerators busy.</p> <h2 id="how-it-works">How it works</h2> <p>Each capability targets a different layer of a distributed training job.</p> <h3 id="default-and-advanced-metrics">Default and advanced metrics</h3> <p>Enhanced observability collects telemetry at two levels:</p> <p><strong>Default metrics</strong> (enabled automatically) cover cluster health, CPU/disk/OS-level usage, GPU utilization and temperature, and per-job resource utilization. These metrics are collected every 30 seconds and sent to your chosen metric store with approximately 1–2% CPU overhead and 250–300 MB of memory.</p> <p><strong>Advanced metrics</strong> (opt-in) provide deeper visibility:</p> <ul> <li><strong>Full node exporter suite:</strong> comprehensive host-level metrics.</li> <li><strong>All NVIDIA GPU DCGM and AWS Neuron metrics:</strong> detailed accelerator telemetry including clock speeds, power, PCIe activity, and error indicators.</li> <li><strong>Elastic Fabric Adapter (EFA) metrics:</strong> network interface throughput, drops, retransmissions, and errors.</li> </ul> <p>Advanced metrics are optional because they increase the volume of telemetry sent to your metric store, which has a cost implication. Most customers running small jobs on small clusters may not need them, but they are essential for long-running training jobs on large clusters where patterns emerge over time.</p> <h3 id="metric-stores">Metric stores</h3> <p>You can choose between two metric stores:</p> <ul> <li><strong>Amazon CloudWatch</strong> (new default) provides native monitoring through CloudWatch SageMaker Insights, so you can visualize metrics without leaving the AWS Management Console.</li> <li><strong>Amazon Managed Service for Prometheus</strong> for teams that standardize on Prometheus and Grafana for observability.</li> </ul> <p>Previously, only Prometheus was supported. CloudWatch is now the default, offering lower pricing compared to Amazon Managed Service Prometheus and tighter integration with the SageMaker console.</p> <h3 id="diagnostic-summary">Diagnostic summary</h3> <p>The diagnostic summary uses two complementary approaches to surface issues across your HyperPod cluster.</p> <p><strong>GPU outlier detection</strong> uses CloudWatch’s <em>DetectGroupOutliers</em> API to compare GPUs within a training job across utilization, memory usage, temperature, power draw, and SM clock frequency. When a GPU deviates significantly from its peers, it appears as an outlier card in the diagnostic summary. This works for both EKS and Slurm jobs.</p> <p><strong>Static diagnostic rules</strong> evaluate known failure patterns and organize detected issues by type:</p> <ul> <li><strong>Hardware:</strong> GPU ECC memory errors, XID faults, accelerator failures.</li> <li><strong>Network:</strong> EFA throughput degradation, packet drops, retransmissions.</li> <li><strong>Throttling:</strong> GPU clock throttling because of heat, board, or reliability limits.</li> <li><strong>Overheating:</strong> GPU temperatures exceeding configured limits.</li> </ul> <p>Each detected issue identifies the affected job, cluster, node, accelerator, and rank.</p> <h2 id="how-to-get-started">How to get started</h2> <p>Enhanced observability is delivered as an add-on for EKS-orchestrated clusters and as an extension for Slurm-orchestrated clusters, which means you control who can install or change it through AWS Identity and Access Management (IAM).</p> <h3 id="step-1-create-a-sagemaker-hyperpod-cluster">Step 1: Create a SageMaker HyperPod cluster</h3> <p>If you already have a cluster, you can use it. To create a new SageMaker HyperPod cluster, you can follow the <a href="https://awslabs.github.io/ai-on-sagemaker-hyperpod/" target="_blank" rel="noopener">step-by-step outline</a>. In general, the steps are as follows:</p> <ol type="1"> <li>On the Amazon SageMaker AI console, choose <strong>HyperPod clusters</strong> in the navigation pane.</li> <li>Choose <strong>Create HyperPod</strong> cluster and <strong>Orchestrated on Amazon E

35

b12543

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

github-actions[bot] 1 bài viết · ggml-org/llama.cpp releases
Nguyên văn nguồn
Chi tiết nguồn

<details open=""> <p>metal : CONV_2D tensor API for M5, split-K for all chips (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6468918084" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32542" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32542/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32542">#32542</a>)</p> <ul> <li>metal: add a tensor API conv_2d kernel and split long reductions</li> </ul> <p>The conv_2d matmul kernel runs on simdgroup matrices only, and gives<br> each threadgroup the whole reduction of its output tile. On M5 this<br> leaves the tensor cores unused, and convolutions with few output<br> positions and a long reduction (IC * KW * KH in the thousands) run a<br> handful of threadgroups on an otherwise idle GPU.</p> <p>Add kernel_conv_2d_mm_tensor for devices with the tensor API: like<br> kernel_mul_mm, it gathers the patches of 64 output positions into<br> threadgroup memory as the left operand and reads the kernel rows<br> from device memory as the right operand. Devices without the tensor<br> API keep the simdgroup kernel.</p> <p>Both kernels can now split the reduction until there are enough<br> threadgroups: each split writes its partial sums to a buffer placed<br> after the output, as flash attention does, and a small kernel adds<br> them in a fixed order, so the result stays deterministic. CONV_2D is<br> marked as an op whose allocation may exceed its output size.</p> <p>test-backend-ops -o CONV_2D passes, with and without<br> GGML_METAL_TENSOR_DISABLE.</p> <p>CONV_2D, F16 kernel, us/run on an M5 Max:<br> [19,19,256,16] k [4,4,256,4096] 19008 -&gt; 5235<br> [16,16,128,8] k [3,3,128,512] 333 -&gt; 124<br> [1536,1536,32,1] k [3,3,32,64] 12086 -&gt; 6512<br> [33,1,1280,1] k [3,1,1280,1280] s2 659 -&gt; 33<br> [2054,1,768,1] k [7,1,768,768] 3161 -&gt; 699</p> <ul> <li>metal: walk the conv_2d patches incrementally, size splits by output</li> </ul> <p>The patch gather of both conv_2d matmul kernels spent most of its<br> time on address math: 64-bit multiplies, divisions and four bound<br> checks per value. Walk the reduction index with 32-bit byte offsets<br> inside the image, updated by additions, with one unsigned compare per<br> bound; the host checks that an image fits. The tensor kernel no<br> longer gathers the reduction indices past the end of its split.</p> <p>The number of splits now targets about 1M partial sums, with at least<br> 3 reduction steps per split, instead of counting the tiles of one<br> kernel.</p> <p>CONV_2D, F16 kernel, us/run on an M5 Max:<br> [19,19,256,16] k [4,4,256,4096] 4998 -&gt; 4512<br> [1536,1536,32,1] k [3,3,32,64] 6721 -&gt; 5354<br> [4096,1,1024,1] k [7,1,1024,1024] 2701 -&gt; 2236<br> [2054,1,768,1] k [7,1,768,768] 1050 -&gt; 871<br> with GGML_METAL_TENSOR_DISABLE:<br> [4096,1,1024,1] k [7,1,1024,1024] 9047 -&gt; 8283</p> <ul> <li>nit: share the CONV_2D simdgroup count between host and kernel</li> </ul> <p>kernel_conv_2d_mm and its pipeline both hardcoded 4 simdgroups per<br> threadgroup. N_SG_CONV_2D in ggml-metal-impl.h now feeds the kernel<br> constant and res.nsg, so the dispatch and the kernel cannot drift apart.</p> <ul> <li>metal: fall back to the generic conv_2d kernel for images past 4 GiB</li> </ul> <p>The matmul conv_2d kernels walk the patches with 32-bit byte offsets<br> inside an image. Instead of asserting on nb13, compute the largest<br> offset actually read from ne/nb, which also covers strided views of<br> the input, and route images that do not fit to the generic kernel.</p> <ul> <li>cont : comments</li> </ul> <p>Co-authored-by: Georgi Gerganov <a href="mailto:ggerganov@gmail.com">ggerganov@gmail.com</a></p> <hr> <p>Co-authored-by: Georgi Gerganov <a href="mailto:ggerganov@gmail.com">ggerganov@gmail.com</a></p> </details> <p><strong>Website:</strong></p> <ul> <li><a href="https://llama.app" rel="nofollow">https://llama.app</a></li> </ul> <p><strong>Attestations:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/attestations/69742362">https://github.com/ggml-org/llama.cpp/attestations/69742362</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-macos-arm64.tar.gz">macOS Apple Silicon (arm64)</a></li> <li>macOS Apple Silicon (arm64, KleidiAI enabled) <a href="https://github.com/ggml-org/llama.cpp/pull/23780" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/23780/hovercard">DISABLED</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-xcframework.zip">iOS XCFramework</a></li> </ul> <p><strong>Linux:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-26.04-s390x.tar.gz">Ubuntu 26.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-zdnn-s390x.tar.gz">Ubuntu 26.04 s390x (zDNN)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12543/llama-b12543-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com

  • b12543 ggml-org/llama.cpp releases · 19 thg 11, 2026
36

Launch HN: Onlook (YC W25) – An open-source developer tool for designers

68

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 90 tương tác đã đo · Nguồn crawl

D_R_Farrell 1 bài viết · HN: LLM
Nguyên văn nguồn
Chi tiết nguồn

Hey HN, I'm Daniel, founder of Onlook ( https://github.com/onlook-dev/onlook and onlook.com). Onlook is a desktop app that puts a canvas on top of your real codebase, letting you design directly in code with the agent you already use. Over 25,000 designers and developers have supported the project, 100+ have contributed, and it's been the #1 trending repo in the world twice. We just rebuilt the entire product as an agentic design tool, and open-sourced it to the community. Here’s a demo video: ( https://youtu.be/BogcYWf4siM ) If you're an engineer, traditional engineering tools are largely built for convergence, whereas designers tend to explore and diverge on ideas. Onlook gives designers a familiar canvas while managing all of the dev processes that get between them and making mind-blowing designs in code. Designers can work on a multiple simultaneous worktrees that run on the real app locally, and every change lands as a PR you can review or throw away, so there aren’t mockups to rebuild. We built Onlook because we believe there needs to be a developer tool that works the way designers do. More and more, designers are expected to work in code. Designers think by seeing, drafting, arranging variations, using references, and remixing ideas. That’s why the canvas has been the signature feature of design tools because it genuinely is useful for how messy the creative process is. Most of the challenge of using engineering tools for design work is that the first half of starting a project often takes mental bandwidth in of itself (What branch am I on? What page was that last exploration on?) and the second half is constrained by configuring your coded design to present itself (prompting design tools on top of what you’ve built or making temporary navigation for that specific state). Some of you may remember our earlier Show HNs (July 2024: https://news.ycombinator.com/item?id=40904862 , August 2024: https://news.ycombinator.com/item?id=41390449 , May 2025: https://news.ycombinator.com/item?id=44127653 ). Since then, we built a web version (too early), then a Storybook-native build (a mistake). We noticed that our desktop product from July 2024 was still getting hundreds of downloads a month despite zero promotion, so we decided to rebuild it for the new era of designers working in code. How it works… Onlook shines at exploring tons of ideas at once. Ask it to lay out every flow in your product, show every state of a component, or riff on ten versions of one screen. Your designs all show up side by side on an infinite canvas, with each frame as a real webview running your actual app with real data. Not as a mockup, but the real deal. When we built this editor, we had an epiphany that it could be a new kind of IDE for design. Design tools used to be limited to a proprietary setup or closed schema, but now you can push and pull the medium of code to get whatever result you want, so why not give agents an infinite canvas of browsers and see what they do? It turns out, the agents can figure out how to better present coded work by adjusting the urls of browser frames and arranging and resizing them on the canvas. The concept is simple, but it requires a lot of product thinking. For example, because all of the frames on the canvas share a single browser session, and Chromium only allows six connections per host per session, ten frames of the same app end up fighting over limited connections. So, we had to plan around that, and there are still tons of things we need to do to make the canvas ultra performant. Onlook is an Electron app where you can import a repo from GitHub or upload a local folder. Then, Onlook creates a git worktree, starts your dev server, and has the agent sort out missing dependencies or other requirements. You never open a terminal and everything the agent does is a normal git diff that can go to GitHub as a PR. We inject a small script into each frame that draws selection outlines and reads the page, so that way you can click an element, say "make this a card layout," and the agent knows exactly what "this" is. This time we didn't build our own agent. Onlook runs Claude Code or Codex on your own subscription, and we turn whatever the agent emits into one typed event log and draw the UI from that in our own chat interface. On top of that we give the agent a bespoke set of design-specific tools, from arranging frames on the canvas, to making suggestions, to generating multiple iterations of a selection – which is especially useful because the agent decides how to get the app into a state that you need based on your prompt. We also have our own Artifacts which are quick HTML sketches that live in the editor, but outside your repo. They’re almost like components in other design tools in that they can be carried and edited across every project but are detached from the project with no network access. As an example, you can ask the agent to turn your interface into an Artifact, and use it as a reference for another exploration, or download it as HTML and send it to a coworker as an interactive export of something you have in progress. I frequently go back to the artifacts I’ve made as visual snippets when I need new ideas or need to borrow something that’s already been built. The desktop editor is free (you can download it from our website at onlook.com) and open source under Apache-2.0, and we'll eventually offer a paid cloud product for teams. We'd love to hear which repos work and which don't, what you try to do on the canvas, and what's missing. We’re so excited for this to be a new way for designers to work in code! The Onlook team (Joel - thebrubaker and Kris - krisodd), and I will be here all day!

37

Build Tool-Calling Parsers in Rust with AI Agents

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

Keiven Chang 1 bài viết · NVIDIA Developer Blog
Nguyên văn nguồn
Chi tiết nguồn

A tool call can arrive in pieces, an argument can be cut short, or model-specific markup can leak into the answer. A frontend parser turns that output, received... A tool call can arrive in pieces, an argument can be cut short, or model-specific markup can leak into the answer. A frontend parser turns that output, received as token IDs or decoded text, into visible text, reasoning, and tool calls an application can use. In this post, you’ll learn how tool-call parsing works and use an AI agent to build a parser for a new model. You’ll start with a… Source

38

Open weights are the bare minimum of open.

55

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · chưa có số tương tác · Nguồn crawl

ifm_ai 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

Open weights are the bare minimum of open. PAN 2 ships with the weights, the inference code, the training code and an Apache 2.0 license. Few open world models come with all four. Inference code lets you run it. Training code lets you make it your own. Apache 2.0 lets you keep it. https://huggingface.co/IFM/PAN2

39

From Static to Current

61

1 tiếng nói · chưa gắn nhãn trưởng lab · 1 nền tảng · 7 tương tác đã đo · Nguồn crawl

whakim 1 bài viết · HN: GPT
Nguyên văn nguồn
Chi tiết nguồn
21–40 / 203
Trang 2 / 11