AllyshipContent Radar

Deep dives 738

Paper và bài viết kỹ thuật

Crawl live-1795068900000 · 13:22 19 thg 11, 2026 · Xuất bản 13:30 19 thg 11, 2026

Kỹ thuật

langchain-anthropic==1.7.11

<p>Changes since langchain-anthropic==1.7.10</p> <p>release(anthropic): 1.7.11 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6461255356" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41929" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41929/hovercard" href="https://github.com/langchain-ai/langchain/pull/41929">#41929</a>)<br> feat(anthropic): send <code>langsmith_gateway</code> headers from the client (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6433424287" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41902" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41902/hovercard" href="https://github.com/langchain-ai/langchain/pull/41902">#41902</a>)<br> feat(anthropic): support Claude Fable and Mythos 5.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6293362193" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41770" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41770/hovercard" href="https://github.com/langchain-ai/langchain/pull/41770">#41770</a>)<br> fix(anthropic): emit a chunk when an empty thinking block starts (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6434857275" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41907" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41907/hovercard" href="https://github.com/langchain-ai/langchain/pull/41907">#41907</a>)<br> feat(partners): Remove EOL'd Python 3.10 from partner packages (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5886024423" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41268" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41268/hovercard" href="https://github.com/langchain-ai/langchain/pull/41268">#41268</a>)<br> chore(model-profiles): refresh model profile data (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6324372002" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41791" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41791/hovercard" href="https://github.com/langchain-ai/langchain/pull/41791">#41791</a>)<br> chore(model-profiles): refresh model profile data (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6302905023" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41781" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41781/hovercard" href="https://github.com/langchain-ai/langchain/pull/41781">#41781</a>)<br> fix(anthropic): raise <code>ModelRefusalError</code> for refusals when tool calls are not forced (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6226735008" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41715" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41715/hovercard" href="https://github.com/langchain-ai/langchain/pull/41715">#41715</a>)<br> fix(anthropic): preserve nested <code>caller</code> on server calls and results (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6205834378" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41678" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41678/hovercard" href="https://github.com/langchain-ai/langchain/pull/41678">#41678</a>)</p>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Time-Series Models in the Streamhouse: A New Frontier and Use Cases

Real-time forecasting, anomaly detection, and optimization, powered by IBM time-series models with more than 50 million downloads. Now available natively on Confluent, in the cloud and on premises, running on CPUs. Every business generates a continuous record of change: a machine reading, a payment, a sale, an order, or a change in inventory. These observations form time series, and they carry signals that can help teams anticipate what happens next. The challenge is turning those signals into decisions while they still matter. A forecast delivered after a production run, a stockout, or a payment authorization may be accurate, but too late to influence the outcome.

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

Tackling GPU Bottlenecks in Plane-Wave DFT with K-Point Batching and Mixed Precision

Training a next-generation machine learning interatomic potential (MLIP) requires ground truth: tens to hundreds of millions of density functional theory (DFT)... Training a next-generation machine learning interatomic potential (MLIP) requires ground truth: tens to hundreds of millions of density functional theory (DFT) calculations of energies, forces, and stresses across diverse structures and chemistries. Plane-wave DFT, with VASP as its dominant implementation, is the gold standard for generating that data. It is accurate… Source

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

The PSF end-of-year fundraiser

<p><strong><a href="https://donate.python.org">The PSF end-of-year fundraiser</a></strong></p> With my Python Software Foundation board director hat on: we've launched our end-of-year fundraiser for the non-profit that supports Python development and the Python community.</p> <p>If your company or your work benefits from Python, please consider donating this year!</p> <p>The PSF looks after the IP behind Python, operates PyPI, sponsors core developer sprints, supports community events (including running PyCon US every year), and employs a small and excellent staff team to keep everything running.</p> <p>We also operate the <a href="https://www.python.org/psf/developersinresidence/">Developers in Residence program</a>, which now includes three security developers (welcome, Brett Smith) focused on helping secure Python, PyPI, and the wider ecosystem. <p>Tags: <a href="https://simonwillison.net/tags/python">python</a>, <a href="https://simonwillison.net/tags/psf">psf</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

NVIDIA RTX Spark Compact Desktops Out Now

New NVIDIA RTX Spark compact desktops from ASUS, Dell and MSI are available today, bringing local AI agents, image and video generation, and model fine-tuning to a dedicated Windows PC. With up to 128GB of unified memory and NVIDIA’s AI software stack, these systems provide room for larger local models alongside RTX-accelerated creative and development tools. A dedicated desktop can keep agents and scheduled tasks running while a laptop travels or handles other work. Alongside the new desktops, an expanding software ecosystem brings more creative and development workflows to RTX Spark laptops and compact desktops. Native Windows on Arm applications from Adobe, the NVIDIA DLSS Ray Reconstruction feature in Blender and local generation workflows in ComfyUI expand what creators and developers can do on RTX Spark — all on one Windows PC. A Windows PC Built for Local AI As workflows expand to include local AI models and agents, more work can continue beyond an active editing session — from generating assets to processing long renders or scheduled tasks. RTX Spark compact desktops give developers, creators and AI enthusiasts a dedicated Windows PC for local models and always-on agents. They bring together image and video generation, model fine-tuning, creative applications and development tools in one compact system. Running supported models locally keeps locally processed code and files on the PC and avoids per-token cloud application programming interface (API) charges for those tasks. RTX Spark compact desktops like the ASUS ProArt GR1X mini PC are built to run personal AI agents 24/7. RTX Spark desktops are equipped with a 20-core NVIDIA Grace CPU, an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores and up to 128GB of unified memory, delivering up to 1 petaflop of AI performance. Ten-gigabit Ethernet connectivity enables fast transfers of large datasets, model files and creative assets between RTX Spark compact desktops and networked storage. This advanced hardware unlocks incredible capabilities for creators, developers and AI enthusiasts, including: Running massive 120-billion-parameter large language models locally Rendering 90GB+ 3D scenes in real time Smoothly editing 4:2:2 video in real time The MSI EdgeMesa N AI+ Series RTX Spark compact desktop. New compact desktops are available for purchase from ASUS, Dell and MSI, joining the RTX Spark family of laptops available today. Mini PCs from Acer, GIGABYTE and HP are coming soon, along with the Microsoft Surface RTX Spark Dev Box . From Local Agents to AI Creation Native CUDA support and Linux development tools through Windows Subsystem for Linux give developers a foundation for building and running local AI workflows. Developers can set up agents with Hermes Agent or NVIDIA NemoClaw blueprints and fine-tune supported models using Unsloth. RTX Spark can run creative agents that connect familiar applications and local AI tools, helping creators move from an idea to finished work with natural-language instructions. NVIDIA Switchyard, now available in Hermes, automatically routes agent tasks between local and cloud models. Everyday tasks can run on device, while advanced reasoning draws on cloud models, helping manage cloud costs across creative workflows. These capabilities come together in creative workflows that connect Photoshop and ComfyUI through Hermes Agent. Using natural-language instructions, creators can turn sketches into rendered images and animations, or animate an image entirely on device by isolating a subject and defining its motion. NVIDIA brought these workflows to life at Adobe MAX, where attendees used natural-language instructions to turn sketches into rendered images and generate video for a shared digital Miami landscape. The RTX Video Frame Generation and RTX Video HDR features, already available in ComfyUI, extend the pipeline: a generated clip can be expanded to 60 frames per second and converted to high dynamic range for color grading, all within a single ComfyUI workflow on RTX Spark. Adobe Accelerates the Full Creative Stack for RTX Spark NVIDIA and Adobe are expanding native Windows on Arm support and RTX acceleration across Adobe’s creative applications, helping video editors, photographers, designers and 3D artists work more efficiently on RTX Spark. Adobe Premiere accelerates a range of AI-powered editing features through NVIDIA TensorRT-RTX , including Enhance Speech, Object Mask, Scene Edit Detection and Auto Reframe — so editors spend less time waiting and more time cutting. GPU acceleration also extends to Film Impact transition and effects plug-ins. The update improves support for professional video formats, including Canon RAW, Red RAW, ARRI RAW, ARRICORE, ARRI HDE, H.264 and HEVC in MXF, ProRes RAW and long-GOP video. This lets editors realize more than 50% quicker turnaround on everyday H.264 video deliverables. Adobe Photoshop adds TensorRT-RTX acceleration for the Remove Tool and ArmRTX Spark-native Camera Raw support, helping photographers remove unwanted objects faster and work with RAW images without relying on emulation. At Adobe MAX, Adobe previewed a next-generation image processing engine with extensive GPU acceleration, including interactive Gaussian blur and Light Leaks filters. Adobe Substance 3D Painter 12.2 is now optimized for RTX Spark with native support on Arm for the first time in public beta. Artists in games, visual effects and visualization can run Substance Painter natively on RTX Spark laptops and compact desktops, with no compatibility layer required. Adobe Illustrator and InDesign now offer native Windows on Arm versions optimized for RTX Spark. Adobe Illustrator delivers up to 30% higher performance than its emulated version, helping accelerate illustration, graphic design and page-layout workflows on RTX Spark. Blender 5.3 Adds DLSS Ray Reconstruction to the Viewport Blender 5.3 integrates DLSS 4.5 Ray Reconstruction into the Cycles viewport, helping artists evaluate light

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

The Silent Consensus

Dean Ball, Head of Strategic Futures at OpenAI, examines how more capable AI could reshape society and how people can retain control of their institutions.

Nguyên văn nguồn
Chi tiết nguồn
Huấn luyện

At MIT, Martha Minow speaks on sustaining democracy, both by using and overcoming AI

<p>Legal scholar Martha Minow delivered this fall’s Compton Lecture at MIT on Thursday, examining the civic foundations of democracy — and describing how artificial intelligence is undermining it, but could be used to help reinforce traditional U.S. governance.&nbsp;</p><p>“It’s not a good time for democracy or constitutionalism,” said Minow, who is the 300th Anniversary University Professor at Harvard University, and a prominent expert in constitutional law and human rights. She noted a litany of challenges to democracy in recent years, including violations of laws and norms, the stifling of free speech, a diminution of congressional authority, disregard of legal rulings, the common vilification of others in society, and more.&nbsp;</p><p>Recent Gallup polling, Minow observed, shows 77 percent of people saying that the founders of the U.S. would be disappointed in the contemporary condition of the country.&nbsp;</p><p>“Many people genuinely wonder whether this nation’s system of government and rule of law, ignited 250 years ago, will survive,” Minow said.&nbsp;&nbsp;</p><p>Her remarks assessed the state of the U.S. in terms of what she called four key “ingredients” of democracy: education, reliable and accessible information, security from violence, and fairness. Minow evaluated how AI has, among other things, influenced these areas of life — but also suggested ways people could use contemporary technology to help.&nbsp;</p><p>“It’s very familiar to hear blame attributed to digital technologies and experts of all stripes, but a turnaround, I believe, is genuinely possible if people at MIT and elsewhere pour time and talent towards real and much-needed solutions,” Minow said.&nbsp;</p><p>A substantial portion of her remarks focused on the pitfalls of AI in education. That issue compounds some other troubling trends: The U.S. already has large local and regional disparities in educational funding, Minow noted, while schools have actually become significantly more segregated in economic and ethnic terms over the last three decades.&nbsp;</p><p>In this context, AI threatens to further weaken independent-thinking skills, Minow suggested, just when greater understanding of civic issues is absolutely vital.&nbsp;</p><p>“As AI surges, I think we have to learn to be smarter about how to use it, how to regulate it, and how to actually educate people about the costs of mindless using,” Minow said. She added: “Dominant AI platforms supercharge the strategies that are deployed by social media companies, inducing passivity in people … Leading AI models and interfaces and sycophantic elements endanger the development particularly of young people’s capabilities to think, to remember, and to engage in critical thinking.”&nbsp;</p><p>That trend away from cognitive effort, she suggested, can be changed — perhaps through AI tools that act more like tutors and are designed to enhance people’s deep thinking.</p><p>“Existing LLMs, model training, and interface design can offer nudges and questions and hints rather than replacing student work, but there’s so much more that could be developed, and more effectively, to be trustworthy, and to deepen human activities and curiosity, rather than replacing people,” Minow said.</p><p>To the audience, she added, “AI tools could do much more of that … And I believe that you, the people here, could develop better tools than the ones that are out there. And make better assessments, more meaningful and cheating-resistant.”&nbsp;</p><p>Minow’s talk, “Constituting Democracy in the Era of AI,” was delivered to a large audience in MIT’s Huntington Hall, Room 10-250 on campus.</p><p>MIT President Sally A. Kornbluth delivered introductory remarks to the audience.&nbsp;The&nbsp;Karl Taylor Compton Lecture Series, introduced in 1957, is the highest-profile event of its kind at MIT. Karl Taylor Compton was MIT’s ninth president, serving from 1930 to 1948, and then becoming chair of the MIT Corporation from 1948 to 1954.</p><p>“He guided MIT through the Great Depression and World War II,” Kornbluth said, “transforming the Institute from an outstanding school for training job-ready engineers to a great global research university.”&nbsp;</p><p>She added: “He was also a deeply committed patriot who helped harness the intellectual power of America’s research universities to serve the nation during World War II.”&nbsp;</p><p>Kornbluth called Minow “a legal scholar of extraordinary intellectual scope,” and a “dedicated educator whose work has shaped how we think about the law and our own humanity, as individuals and as a society.”&nbsp;</p><p>Minow has been a faculty member at Harvard Law School since 1981, and served as dean of the school from 2009 through 2017. Minow earned her BA in history from the University of Michigan, an MA in education from Harvard, and a JD from Yale University.</p><p>Minow’s books include “Making All the Difference” (1990), “Partners Not Rivals” (2002), and “When Should Law Forgive?” (2020).&nbsp;</p><p>In her talk, Minow also discussed the civic effects of misinformation, threats of violence, and multiple dimension of societal fairness. Modern media technology has led to a 75 percent reduction in journalists in recent decades, Minow observed, with an attendant rise in corruption in places without local media. Civic-minded technologists, she proposed, should work towards establishing a widely-used “information commons,” or many forms of them, to help people access more reliable information.&nbsp;</p><p>Meanwhile, when it comes to basic security in society, Minow warned, “We’re on the brink right now. I know a lot of judges, and every one of them has a credible threat of violence against them. In America. This is not acceptable.”&nbsp;</p><p>For all the challenges to democracy and traditional civic norms, however, Minow sounded positive notes throughout her remarks, pointing out that there are many opportunities to create new forms of reliable information, educa

  • Modern media technology has led to a 75 percent reduction in journalists in recent decades, Minow observed, with an attendant rise in corruption in places without local media.
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Inside the Codex sandbox

See how Codex runs commands on your machine while enforcing the filesystem and network limits you control across macOS, Linux, and Windows.

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

Helping AI text detectors keep up with changing conditions

<p>A teacher who believes a student’s term paper may have been written by a large language model (LLM) can use a tool for detecting artificial intelligence-generated text to help confirm their suspicions. But the teacher may want to think twice before putting all their trust in the detector.&nbsp;</p><p>Researchers from MIT and Cornell University showed that such detection tools, which are AI models themselves, can be fooled by AI-generated text disguised to appear as if it were written by a human. Detectors can also fail to identify AI text due to natural shifts in writing habits over time or the emergence of new LLMs.</p><p>In a <a href="https://arxiv.org/pdf/2606.25152" target="_blank">new paper</a>, the researchers demonstrated how a special training technique makes an AI detector more accurate and resilient to deception, even as human and AI writing constantly changes.</p><p>They employed a method called test-time adaptation, which allows a deployed detection model to learn from new, unlabeled data by adjusting its inner workings on the fly. In this way, the model adapts to unseen inputs without being retrained from scratch.</p><p>In experiments, a commercial detector could only identify about 27 percent of AI-generated text that had been disguised, while the new approach achieved 93 percent accuracy.</p><p>This method could help educators, editors, publishers, and hiring managers flag LLM-generated writing that people try to pass off as human-written. It could also be used by cybersecurity experts to prevent email phishing scams or identify search engine optimization fraud.</p><p>“There are many high-stakes contexts where people really care about the identity of the author, from scientific publishing to short story competitions. We’ve shown that test-time adaptation allows your AI detector to be much more robust, even with continual adversarial behavior by AI users,” says study co-author Manish Raghavan, the Drew Houston (2005) Career Development Professor at the MIT Sloan School of Management, who also holds an MIT Schwarzman College of Computing shared position in the Department of Electrical Engineering and Computer Science (EECS), and is a principal investigator in the Laboratory for Information and Decision Systems (LIDS).</p><p>He is joined on the paper by lead author Kevin Ren, graduate student at Cornell Tech, and senior author Nikhil Garg, an assistant professor at Cornell Tech. The research will be presented at the Conference on Neural Information Processing Systems.</p><p><strong>Duping the detectors</strong></p><p>The proliferation of LLMs that generate human-like text has led to a plethora of commercial AI-detection tools to help people identify whether a piece of writing was authored by a human.</p><p>But these detection tools can fail when the text they evaluate significantly differs from the text they were taught to identify. This is known as a distribution shift.</p><p>Detector developers typically train their models by gathering text from the internet that they know to be either AI- or human-generated. They train the model to tell the difference.&nbsp;</p><p>However, once the detector is released, people might change the way they write or the way they use AI to generate text.&nbsp;</p><p>For instance, a human in 2030 may write about different topics using new analogies that weren’t common for someone writing in 2010. A detector trained with 2010 data might be unfamiliar with this new writing style and incorrectly mark human-written text as AI-generated.</p><p>The release of a detection model also creates a cat-and-mouse game between its developers and users who want to bypass it. These individuals can run AI-generated text through an automated “humanizer” software tool that modifies it to appear human-written to the detector.</p><p>“After the company releases a model, a bunch of ‘humanizers’ pop up to try to beat it, so the company tries to incorporate this humanized text into their training so the detector can identify it, and so on. It’s like rinse-and-repeat. The cycle can go on forever,” says Ren.</p><p>The researchers explored whether test-time adaptation (TTA) can make an AI detector resilient to these distribution shifts.</p><p><strong>Adaptation in action</strong></p><p>In TTA, instead of freezing the model so it can’t change during deployment, researchers allow that model to adapt itself to incoming data after it has been released. For an AI text detector, these data are new pieces of text uploaded by users.&nbsp;</p><p>These data are unlabeled, so the detector developers don’t know which data were written by humans versus AI models. But these new data represent contemporary human and AI writing — exactly what the developers want the model to be able to predict.&nbsp;</p><p>With enough examples, the detector learns to identify patterns to determine which data are AI-generated and which are human-generated.</p><p>The researchers studied TTA by creating datasets to contain a certain amount of AI-generated and human-generated text, but they don’t provide those labels to the model. They also used LLMs to generate some examples that resemble text written by AI to specifically evade a detector.&nbsp;</p><p>Then they performed the TTA training process and tested the model on ground-truth data.</p><p>In experiments designed to mimic the natural shift in human language over time, a detector that used TTA was 85 percent accurate, while traditional methods only achieved 73 percent accuracy.</p><p>And when it came to writing designed to thwart detection tools, existing methods could only identify about 27 percent, while TTA enabled the model to catch more than 93 percent of disguised text.</p><p>“TTA has the potential to really help deployed AI-detection models in terms of being able to proactively address this iterative ‘gaming’ behavior by adversarial AI users,” Ren says.</p><p>Adapting to shifts in human writing behavior is a more difficult TTA problem, since the

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

v18.22.2

<h2>@oh-my-pi/pi-ai</h2> <h3>Added</h3> <ul> <li>Added the native Kiro provider: sign in with AWS IAM Identity Center, AWS Builder ID, Google, GitHub, or a Kiro API key (<code>KIRO_API_KEY</code>), stream responses with reasoning, tool calls, and images, and report Kiro credits in <code>/usage</code> (<a href="https://github.com/can1357/oh-my-pi/pull/18955" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18955/hovercard">#18955</a> by <a href="https://github.com/Xytronix">@Xytronix</a>, based on <a href="https://github.com/can1357/oh-my-pi/pull/14215" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/14215/hovercard">#14215</a> by <a href="https://github.com/ajdiyassin">@ajdiyassin</a>)</li> <li>Added a <code>definitive</code> option to <code>OAuthError</code>; <code>isDefinitiveOAuthFailure</code> now takes the error object and disables a credential marked definitive without matching its message (<a href="https://github.com/can1357/oh-my-pi/pull/18955" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18955/hovercard">#18955</a> by <a href="https://github.com/Xytronix">@Xytronix</a>)</li> <li><code>buildAnthropicSearchHeaders</code> takes <code>extraBetas</code>, request-specific betas sent beside the web search beta (<a href="https://github.com/can1357/oh-my-pi/pull/18125" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18125/hovercard">#18125</a> by <a href="https://github.com/will-bogusz">@will-bogusz</a>)</li> <li>Added <code>omitDemotedReasoning</code> to <code>StreamOptions</code>: an Anthropic request leaves out other models' reasoning that it would otherwise replay as text, and a refused request reports how many such blocks it carried in <code>demotedReasoningBlocks</code> (<a href="https://github.com/can1357/oh-my-pi/issues/10407" data-hovercard-type="issue" data-hovercard-url="/can1357/oh-my-pi/issues/10407/hovercard">#10407</a>) (<a href="https://github.com/can1357/oh-my-pi/pull/17610" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/17610/hovercard">#17610</a> by <a href="https://github.com/will-bogusz">@will-bogusz</a>)</li> </ul> <h3>Fixed</h3> <ul> <li>Fixed shutting down or cancelling an omp session making every other session sharing the agent directory see its OAuth accounts as depleted for 5 minutes; still-valid tokens are also served during shutdown (<a href="https://github.com/can1357/oh-my-pi/issues/19607" data-hovercard-type="issue" data-hovercard-url="/can1357/oh-my-pi/issues/19607/hovercard">#19607</a>)</li> <li>OAuth accounts refreshed through an auth broker keep their stored routing metadata (such as an API endpoint) in the key built for the request, instead of losing it after the first refresh (<a href="https://github.com/can1357/oh-my-pi/pull/18955" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18955/hovercard">#18955</a> by <a href="https://github.com/Xytronix">@Xytronix</a>, based on <a href="https://github.com/can1357/oh-my-pi/pull/14215" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/14215/hovercard">#14215</a> by <a href="https://github.com/ajdiyassin">@ajdiyassin</a>)</li> <li>Fenced (```thinking) reasoning with very long unbroken lines no longer slows down as it streams (<a href="https://github.com/can1357/oh-my-pi/pull/19493" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/19493/hovercard">#19493</a> by <a href="https://github.com/H4vC">@H4vC</a>)</li> <li>Fixed Anthropic requests that replay a signed server-compaction block still asking for a message thread, which the API refuses (<code>thread_unsupported_request</code>), so every such request first paid a rejected round trip before resending without the thread (<a href="https://github.com/can1357/oh-my-pi/pull/19473" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/19473/hovercard">#19473</a> by <a href="https://github.com/alphastorm">@alphastorm</a>)</li> <li>Anthropic requests whose connection drops before any response now say how long after the request started it dropped and whether the request body was still uploading or had all been written to the socket, with its size, instead of a bare "Connection error." (<a href="https://github.com/can1357/oh-my-pi/pull/18624" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18624/hovercard">#18624</a> by <a href="https://github.com/will-bogusz">@will-bogusz</a>)</li> <li>Fixed Anthropic requests dropping the reasoning of earlier non-Claude turns signed in under another credential, or kept through a compaction; it is now replayed like other models' reasoning (<a href="https://github.com/can1357/oh-my-pi/issues/7241" data-hovercard-type="issue" data-hovercard-url="/can1357/oh-my-pi/issues/7241/hovercard">#7241</a>) (<a href="https://github.com/can1357/oh-my-pi/pull/17610" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/17610/hovercard">#17610</a> by <a href="https://github.com/will-bogusz">@will-bogusz</a>)</li> </ul> <h2>@oh-my-pi/pi-catalog</h2> <h3>Added</h3> <ul> <li>Added Kiro to the model catalog with automatic discovery of every model available to the signed-in Kiro account (<a href="https://github.com/can1357/oh-my-pi/pull/18955" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/18955/hovercard">#18955</a> by <a href="https://github.com/Xytronix">@Xytronix</a>, based on <a href="https://github.com/can1357/oh-my-pi/pull/14215" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/14215/hovercard">#14215</a> by <a href="https://github.com/ajdiyassin">@ajdiyassin</a>)</li> <li>Added Factory Droid's Claude Fable 5.2, Gemini 4 Carbon (with its launch credit promotion, half the input rate until 2026-12-01) and Kimi K3.1 Pro, each listed once the account's feature flag enables it, and G

Nguyên văn nguồn
Chi tiết nguồn
  • v18.22.2 Can Bölük — Oh My Pi releases · 18 thg 11, 2026
Kỹ thuật

180B Parameters, 32GB of RAM: Pushing MoE Inference to the Limit with Transformers

Streaming MoE experts directly from SSD gets within 6% of Strata on token generation. I've been trying to rely less on online models and run bigger ones locally. At home I have a PC with two RTX 4090s (let's call it Orthrus - the two-headed hound of Greek mythology), and for work I have a MacBook M5 with 32G of unified memory. Orthrus runs on Windows 11, because my kids play games all the time. I had a dual boot with Ubuntu, but their games ended up taking most of the two terabytes of disk.

Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Rumble-2-Flash: an open 0.5B speech model for 40+ languages

Today we're open sourcing Rumble-2-Flash under the Apache 2.0 license. It's a 0.5B-parameter speech recognition model for 40+ languages that competes with models 3-60x its size. audio accepts a local path, a URL, a NumPy array, or a list of these. Rumble-2-Flash uses the Qwen3-ASR architecture for day-0 support. If your stack runs Transformers or vLLM, it runs Rumble-2-Flash, with no forked runtime or trust_remote_code .

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

GPT-6.2 Astra 🧠, Microsoft’s seven models 🚀, Replit Missions 🤖

Get Delinea's 2026 Identity Security Report to learn why standing privilege is creeping back into mature identity programs and how agents become over-permissioned. See the full data and learn how to extend control from the point of access to the moment of action.

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

Keeping Agents Fast: High-Interactivity Decode with FlyDSL MonoKernels in vLLM

How FlyDSL MonoKernels keep agents responsive with high-interactivity decode in vLLM on AMD Instinct GPUs, using model-specific fusion, data-ready handoffs, and specialized-decode integration.

vLLM Blog · Felix Li, Lingpeng Jin, Hattie Wu, Carlus Huang, Fangzhou Ai, Zaidy Ali, Amanzhol Salykov, Hyukjoon Lee, Ziming Huang, Jun Chen, Yifeng Huang, Bin Ding, Xiaobing Zhang, Yi Gan, Vinayak Gokhale, Hongxia Yang, Doug Lehr, Richard Li, Andy Luo, Peng Sun, and Emad Barsoum Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
241–260 / 738
Trang 13 / 37