AllyshipContent Radar

News 2934

Chủ đề cộng đồng đang thảo luận

Crawl live-1795137300000 · 08:15 20 thg 11, 2026 · Xuất bản 08:31 20 thg 11, 2026

213 / 2934 chủ đề đạt ngưỡng 55 trong khoảng này.

21

langchain-core==1.8.4

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 · langchain-ai/langchain releases
Nguyên văn nguồn
Chi tiết nguồn

<p>Changes since langchain-core==1.8.3</p> <p>release(core): 1.8.4 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6500238240" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41964" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41964/hovercard" href="https://github.com/langchain-ai/langchain/pull/41964">#41964</a>)<br> feat(core): standardize finish reason on message-finish events (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6458937894" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41926" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41926/hovercard" href="https://github.com/langchain-ai/langchain/pull/41926">#41926</a>)<br> feat(langchain): tag synthetic HITL tool messages with source metadata (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6461091843" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41928" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41928/hovercard" href="https://github.com/langchain-ai/langchain/pull/41928">#41928</a>)</p>

22

b12554

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>ggml : add BF16 support to ROPE and SET_ROWS (CPU, CUDA) (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6418314884" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32387" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32387/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32387">#32387</a>)</p> <ul> <li> <p>ggml : add BF16 support to ROPE and SET_ROWS (CPU, CUDA)</p> </li> <li> <p>cuda : SET_ROWS from BF16 into quantized types and F8_E4M3</p> </li> <li> <p>cuda : fuse BF16 rope into a BF16 cache</p> </li> <li> <p>tests : trim the BF16 rope cases to one per code path</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/69881950">https://github.com/ggml-org/llama.cpp/attestations/69881950</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-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/b12554/llama-b12554-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-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/b12554/llama-b12554-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-ubuntu-cuda-12.8-x64.tar.gz">Ubuntu x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/cudart-llama-b12554-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/b12554/llama-b12554-bin-ubuntu-cuda-13.5-x64.tar.gz">Ubuntu x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/cudart-llama-b12554-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/b12554/llama-b12554-bin-ubuntu-cuda-13.5-arm64.tar.gz">Ubuntu arm64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/cudart-llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-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/b12554/llama-b12554-bin-android-arm64.tar.gz">Android arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-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/b12554/llama-b12554-bin-win-cpu-x64.zip">Windows x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-win-cpu-arm64.zip">Windows arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-win-opencl-adreno-arm64.zip">Windows arm64 (OpenCL Adreno)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/llama-b12554-bin-win-cuda-12.4-x64.zip">Windows x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12554/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/b12554/llama-b1255

  • b12554 ggml-org/llama.cpp releases · 20 thg 11, 2026
24

How the BMW Group governs hundreds of agents across thousands of developers

55

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

Dr. Jens Kohl 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>When hundreds of agents serve thousands of developers, enterprise governance can’t be an afterthought.</p> <p>At the BMW Group’s R&amp;D, more than 12,500 developers work on approximately 500 million lines of vehicle software code. This software is augmented by the BMW Connected Vehicle backend, a cloud-based system of more than 1,300 microservices providing digital services for 24 million vehicles in the field. In regulated industry, the bar for (software) quality, stability, and governance compliance is exceptionally high.</p> <p>While research, frameworks, and guidelines now facilitate building agents, industrializing and operating high-quality agents in an efficient, scalable manner remains a significant challenge.</p> <p>Scaling agentic AI across a large organization cannot be achieved by isolated teams. It requires collective effort, locally driven experimentation, and active knowledge sharing across the enterprise. This creates a fundamental governance dichotomy: Development teams need the freedom to innovate rapidly, while the enterprise must enforce standards for safety, security, composability, and cost-efficiency. Without that balance, organizations risk a fragmented landscape of AI islands and costly duplicate work. This presents a real challenge given hundreds of globally distributed teams building agents across the BMW Group.</p> <p>In this post, we detail the practices and patterns that support BMW’s R&amp;D teams in building, governing and operating agents at scale without sacrificing speed, quality or compliance.</p> <h2 id="solution">Solution</h2> <p>Our premise is straightforward: build agents that can cooperate from the start. This requires an architecture that makes diversity composable, governable, and automatable, without requiring a single framework or toolchain.</p> <p>Central to this architecture is the agent harness, a runtime layer wrapped around a large language model (LLM). It provides the execution loop, tool dispatch, memory management, and observability hooks for autonomous operation.</p> <p>Just as the <a href="https://aws.amazon.com/architecture/well-architected/" target="_blank" rel="noopener">AWS Well-Architected Framework</a> standardized cloud infrastructure design, agentic systems need equivalent patterns, reusable blueprints encoding governance, composability, and operational best practices from the start.</p> <p>We organize our solution around five interlocking practices, drawn from hands-on experience building and operating agentic systems at enterprise scale since 2023. Each practice is codified as infrastructure as code (IaC), making adoption repeatable, auditable, and enforceable across teams without blocking DevOps velocity.</p> <h3 id="govern-proactively-not-retroactively">Govern proactively, not retroactively</h3> <p>Governance must be structural from the start and maintained across the full lifecycle. We advocate for three properties of effective agentic governance:</p> <p><strong>Framework-agnostic and externalized:</strong> Governance rules such as authentication, authorization, rate limiting and content filtering are enforced at the integration layer. This includes a Model Context Protocol (MCP) gateway, a central agent registry and orchestration boundary, rather than within individual agent codebases. This means governance survives framework migrations, model swaps, and team reorganizations.</p> <p><strong>Domain-sensitive and graduated:</strong> Not every agent action carries the same risk: drafting documentation, updating customer records, and production database writes demand very different levels of oversight. We implement graduated trust through per-step autonomy levels:</p> <ul> <li><strong>Level 0 (Human-in-the-loop):</strong> Every action requires explicit human approval. This is the starting point for new agents and high-risk operations.</li> <li><strong>Level 1 (Human-on-the-loop):</strong> The agent executes autonomously but surfaces decisions for asynchronous human review. Suitable for medium-risk tasks with defined inputs, outputs, and rollback options.</li> <li><strong>Level 2 (Autonomous within guardrails):</strong> The agent operates independently within strictly defined policy boundaries, enforced deterministically rather than through prompt instructions (see below). Reserved for low-risk, well-understood operations with verified track records.</li> </ul> <p>Assign trust levels per step, not per agent. A single workflow can include steps at all three levels, so teams can start conservatively and incrementally increase autonomy as confidence builds, grounded in measurable evaluation data, not assumptions.</p> <p><strong>Deterministically enforced:</strong> Governance runs as deterministic code outside the agent’s reasoning loop. In practice, this happens mainly in the CI/CD pipeline. Permission checks, policy-as-code rules, and compliance validations gate every deployment there, before an agent reaches production. Because these checks run separately from the agent, neither the agent nor its prompts are in a position to override them. Egress controls (recipient allowlisting, outbound secret scanning) help prevent an agent holding confidential data from disclosing it through an otherwise legitimate call.</p> <h3 id="instrument-verify-and-evaluate-continuously">Instrument, verify, and evaluate continuously</h3> <p>Observability in agentic systems goes beyond traditional application monitoring. The real question is no longer whether a service returned an HTTP 200 status, but whether an agent produced a faithful, policy-compliant output, and how this can be demonstrated through observable execution traces. Answering that requires a ground-truth dataset covering common queries and edge cases. This dataset is used for evaluation from day one, so quality is measured against representative and appropriately governed evaluation datasets for evaluation, not synthetic prompts.</p> <p>This infrastructure cannot be retrofitted an

25

Run your whole selling business in AWS Quick

55

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

Josh Anderson 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Most sellers do not need another AI product to learn. They need the tools and information they already use brought together, so they can see what is happening in their business and act on it.</p> <p>AWS Quick and the Amazon Selling Partner plugin do that together. Quick brings an Amazon seller’s sales data, inventory signals, and marketplace intelligence into one place, where questions become answers and answers become actions. It understands the seller’s full business context, so the seller does not need to re-explain their catalog, their margins, or their competitive landscape in every conversation. It is built on AWS, the same infrastructure Amazon uses to run its own marketplace.</p> <p>At Amazon Accelerate on September 23, 2026, Amazon announced the Selling Partner plugin and launched it in beta in AWS Quick. This post covers what launched, how to claim the free AWS Quick offer, and what the experience looks like day to day. It also covers how the skills and connectors work when a seller puts them to use.</p> <p><strong>Limited-time offer: 12 months of AWS Quick Plus, free. Every primary selling account holder worldwide can claim a free 12-month subscription, plus access for 10 additional users you delegate. Sign up by December 31, 2026.</strong> <a href="https://quick.aws.com/sn?offer_type=sp-quick-promo" target="_blank" rel="noopener">Claim your free AWS Quick Plus offer.</a></p> <h1>1. Why Quick is right for Amazon sellers</h1> <p>Four things make Quick a fit for how Amazon sellers already work: it learns the business, it reaches beyond the Amazon store, it turns data into decisions, and it protects seller data on infrastructure sellers already trust.</p> <h2 id="quick-knows-your-business-not-only-your-prompts">Quick knows your business, not only your prompts</h2> <p>Quick and the Selling Partner plugin connect to a seller’s inventory feeds, sales data, and operational documents. From these, Quick builds a living understanding of the catalog structure, margin thresholds, seasonality, and competitive position. Ask Quick to audit your ASINs for Q4, and it returns a ranked recommendation built on the seller’s real margin data, historical conversion, and inventory availability. Its recommendations get sharper the more the seller works with it.</p> <h2 id="quick-does-not-stop-at-your-amazon-store">Quick does not stop at your Amazon store</h2> <p>A seller’s business runs across many tools: a storefront on another channel, an email inbox, an accounting system, spreadsheets, a supplier portal. Quick connects to where the seller already works and pulls context from across the whole operation. It is a business assistant that happens to know Amazon deeply, so the seller grows overall, not on one channel alone.</p> <h2 id="it-turns-data-into-decisions-you-can-act-on">It turns data into decisions you can act on</h2> <p>Sellers are not short on data. They have Brand Analytics, Search Query Performance, advertising reports, inventory health, and listing quality data. The problem is turning those signals into one clear action. Quick does the synthesis. It watches for change, flags what matters, builds the analysis, and presents the decision with the evidence behind it.</p> <h2 id="your-data-stays-protected-on-the-infrastructure-you-already-sell-on">Your data stays protected on the infrastructure you already sell on</h2> <p>Quick runs on AWS and does not use seller data to train models, not for general improvement and not for anything else. Seller data stays in the seller’s environment and never informs anyone else’s models. For a seller already operating on the Amazon marketplace, there is no new security posture to evaluate. The infrastructure is the same one protecting the seller account today, and the seller’s chosen AWS Region governs where data stays.</p> <h1>The Amazon Selling Partner plugin</h1> <p>The Selling Partner plugin connects AWS Quick directly to live Seller Central data. After one sign-in with a Seller Central account, the plugin gives conversational access to product listings, inventory, and sales data, pulled live rather than exported and reassembled by hand. It installs like an app, with no coding required. It ships with curated Amazon skills, and it pairs with Amazon Ads MCP so campaign performance and spend sit next to the selling data. The plugin is in beta for US stores.</p> <div style="width: 640px;" class="wp-video"> <video class="wp-video-shortcode" id="video-145762-2" width="640" height="360" preload="metadata" controls="controls"> <source type="video/mp4" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22255/1_permissions_setup.mp4?_=2"> </video> </div> <p>Figure 1: The Selling Partner plugin consent screen, where the seller grants None, View, or Edit access per data category</p> <p>The permission screen is worth a pause. The seller grants access per data category, choosing None, View, or Edit for each, and reviews an agreement before authorizing. Nothing is all-or-nothing, and access can be narrowed or revoked later.</p> <h1>2. How to redeem the Quick Plus offer</h1> <ol type="1"> <li>Claim your free AWS Quick Plus subscription (12 months free plus 10 delegated users, by December 31, 2026).</li> <li>Connect the Selling Partner plugin with one sign-in to your Seller Central account. It installs like an app.</li> <li>Ask your first question, something you would normally build a spreadsheet for.</li> </ol> <div style="width: 640px;" class="wp-video"> <video class="wp-video-shortcode" id="video-145762-3" width="640" height="360" preload="metadata" controls="controls"> <source type="video/mp4" src="https://d2908q01vomqb2.cloudfront.net/artifacts/DBSBlogs/ML-22255/2_aws_quick_setup.mp4?_=3"> </video> </div> <p>Figure 2: Connecting the Selling Partner plugin from the AWS Quick sign-up flow</p> <h1>3. What the experience looks like</h1> <p>After the plugin is connected, the work happens in a conversation. You ask in plain language, Quick pulls the releva

26

Gemini 4 Neon is getting rave reviews, so I made a new video comparing it to Opus 5.5, my favorite model for everyday work.

55

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

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

Gemini 4 Neon is getting rave reviews, so I made a new video comparing it to Opus 5.5, my favorite model for everyday work. Some takeaways: 1. Opus is still better at 3D and design from my testing, but Neon is on par or better at writing and personality, which is a huge accomplishment. 2. @Google Gemini 4 has 3 models, but in reality I think I'll use Neon for 99% of tasks. It's much faster than Carbon and its results are just as good. 3. The $100/mo Gemini Ultra subscription is a steal. You get Gemini 4, YouTube Premium Family (my favorite subscription ever) for ad-free YouTube and Music, 20 TB of storage for all your photos and files, and more. 📌 Watch me test Neon vs Opus across 5 use cases here: https://youtu.be/iYIWuI5WfUo

27

AI helps researchers investigate how thousands of medical studies share code

55

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

Bwiza Gihana Deborah | Institute for Medical Engineering and Science 1 bài viết · MIT News: AI
Nguyên văn nguồn
Chi tiết nguồn

<p>The data and code that underpin a research study can tell us what was actually done, and how. In a medical study, for example, how were patients selected for inclusion? How were missing values handled? How were the results evaluated?&nbsp;</p><p>Access to a study's data and code help researchers to answer these questions, to examine the methods, and to build on the work.&nbsp;</p><p>“Studies involve many choices about how to prepare data, develop models, and evaluate performance. These are sometimes called ‘researcher degrees of freedom,’” says Tom Pollard, a research scientist at MIT's Laboratory for Computational Physiology and Computer Science and Artificial Intelligence Laboratory. “The choices we make affect the results, and a methods section in a paper can only capture so much detail.”</p><p>While sharing code can be a first step, in order for code to be understood and reused, it must be sufficiently structured and described. The problem is that, right now, there is little guidance for researchers on what constitutes a well-organized, well-described collection of code.&nbsp;</p><p>This is where an international team led by researchers at MIT and Dana-Farber Cancer Institute sought to help, by examining how code is currently shared in clinical prediction studies. Such studies develop or evaluate models to estimate patient outcomes, such as the risk of developing a disease. Because these models inform decisions about patient care, researchers need to be able to scrutinize how they were built and tested.</p><p>Using large language models (LLMs), the researchers examined nearly 4,000 papers to identify statements about shared code. They then followed the statements to assess hundreds of associated code repositories.</p><p>Their review revealed two key gaps: Code sharing remains uncommon, and making code public does not consistently mean providing the documentation and supporting information needed for reuse. The open-access findings, recently <a href="https://www.nature.com/articles/s41591-026-04691-1">published in <em>Nature Medicine</em></a>, are informing <a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC12888484/">TRIPOD-Code</a>, a reporting guideline being developed to clarify what researchers should share.</p><p><strong>Following the code</strong></p><p>TRIPOD-Code builds on TRIPOD and TRIPOD+AI, checklists that help researchers report how clinical prediction models were developed and evaluated. To understand current code-sharing practices, the team identified studies citing either guideline. The review included almost 4,000 studies that developed, updated, or validated prediction models, and whose full texts were available through PubMed Central’s open-access collection.&nbsp;</p><p>“The challenge was that the information was highly inconsistent. First, we had to identify whether or not a paper shared code” says Thomas Sounack, the study’s first author. He adds that authors “discuss and link to code from a paper in many different ways.”</p><p>The team used an LLM to help screen papers, identify statements about code availability, and extract links. Software then retrieved the linked repositories — online collections of code and related files — for further assessment.</p><p>The workflow went beyond checking whether a link existed. It also assessed repositories against characteristics such as completeness of documentation, software requirements, and terms of reuse.</p><p>“The real benefit of LLMs is their ability to process unstructured information such as text. These models helped us work with information that was scattered across papers and software repositories, rather than neatly organized in a table,” says Pollard, who is a senior author of the study. “That allowed us to examine how code was being shared in much more detail across a large collection of studies.”</p><p>Human reviewers evaluated the work of the LLMs. Two researchers independently checked 500 papers and a separate sample of repositories, resolving disagreements before comparing their assessments with the automated results.&nbsp;</p><p>Agreement was stronger for concrete characteristics, such as whether a license file was present, than for judgments about the adequacy of documentation. Because classification errors remain possible, the authors emphasize the overall patterns rather than treating every count as exact.</p><p><strong>Inconsistent code sharing</strong></p><p>Only about 1 in 8 studies reported sharing code. Where code was shared, the contents of repositories varied widely.</p><p>Some repositories were “extremely well-organized and -documented, while others gave almost no guidance at all,” Sounack says. Good code sharing, he explains, means “giving enough information to understand what the code does and be able to run it to reproduce the results reported in the paper.”</p><p>Practices also varied substantially across journals and countries. Among journals, the proportion of studies sharing code ranged from zero to more than 70 percent. Some journals with low sharing rates nevertheless had written code-sharing policies, suggesting that policies alone may not be enough to motivate code sharing.</p><p>The review also helped to identify examples of effective practice. Some researchers, for example, provided synthetic datasets to assist with code reuse where the original patient records could not be shared publicly.</p><p><strong>Setting a standard for code sharing</strong></p><p>Researchers who share code are already making an effort to make their work transparent. But clinical prediction research brings together people from medicine, statistics, and computer science, whose expectations for documenting and sharing code may differ.</p><p>TRIPOD-Code aims to establish a consensus. Through an expert review process, the team is developing guidance on which elements of shared analytical code should be considered essential.</p><p>“We want to give researchers practical guidance on how to share code, and give j

28

okay @PeterJ_Walker / @brendanwduke and lurkers: https://cottonbureau.com/people/nathan-lambert

55

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

Nathan Lambert 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

okay @PeterJ_Walker / @brendanwduke and lurkers: https://cottonbureau.com/people/nathan-lambert paid newsletter subs get a discount: https://interconnects.ai/perks

30

b12553

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>mtmd : fix YuE2 projector loading (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6490599409" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32606" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32606/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32606">#32606</a>)</p> <ul> <li> <p>Allow cls.output.weight to be absent from the mmproj. It is an optional<br> Qwen3 reranking head that the YuE2 NAR decoder never uses.</p> </li> <li> <p>Without this fix, mtmd_init_from_file() fails with<br> "tensor cls.output.weight is missing from the mmproj", preventing<br> audio generation from starting.</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/69877272">https://github.com/ggml-org/llama.cpp/attestations/69877272</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-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/b12553/llama-b12553-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-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/b12553/llama-b12553-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-ubuntu-cuda-12.8-x64.tar.gz">Ubuntu x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/cudart-llama-b12553-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/b12553/llama-b12553-bin-ubuntu-cuda-13.5-x64.tar.gz">Ubuntu x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/cudart-llama-b12553-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/b12553/llama-b12553-bin-ubuntu-cuda-13.5-arm64.tar.gz">Ubuntu arm64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/cudart-llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-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/b12553/llama-b12553-bin-android-arm64.tar.gz">Android arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-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/b12553/llama-b12553-bin-win-cpu-x64.zip">Windows x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-win-cpu-arm64.zip">Windows arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-win-opencl-adreno-arm64.zip">Windows arm64 (OpenCL Adreno)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/llama-b12553-bin-win-cuda-12.4-x64.zip">Windows x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12553/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/down

  • b12553 ggml-org/llama.cpp releases · 20 thg 11, 2026
32

b12552

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>ggml-openvino: fix OpenVINO CI, minor fixes and optimizations (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6484997714" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32590" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32590/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32590">#32590</a>)</p> <ul> <li>ggml-openvino: round the GPU KV length up to a power of two</li> </ul> <p>Review notes (follow-ups, not blocking):</p> <ul> <li>convert_ggml_input_to_ov pads the KQ mask into a std::vector and then<br> copies it into the ov::Tensor. Writing the padded rows straight into<br> padded.data() avoids the second allocation and copy (up to 16 MB per<br> call at a 16K context).</li> <li>The padding is skipped when the mask is not contiguous, and the KV<br> inputs are then sliced to more rows than the mask has columns. Masks<br> are always contiguous today; a GGML_ASSERT would make that explicit.</li> <li>compute_llm_params reads GGML_OPENVINO_STATEFUL_EXECUTION directly;<br> is_stateful_enabled() is local to ggml-openvino.cpp.</li> </ul> <ul> <li>ggml-openvino: Optimize model recompiles of prefill stage</li> </ul> <p>Optimize the model recompiles of prefill stage when attended size<br> changed between prompts. This can save model recompile time for<br> llama-bench and llama-server.</p> <ul> <li>ggml-openvino: Optimize model recompiles of prefill stage</li> </ul> <p>Optimize the model recompiles of prefill stage when attended size<br> changed between prompts. This can save model recompile time for<br> llama-bench and llama-server.</p> <ul> <li> <p>ggml-openvino: Remove solution C and keep A as default</p> </li> <li> <p>ggml-openvino: match the output count before reusing a compiled NPU prefill</p> </li> </ul> <p>The static prefill model has a fixed inp_out_ids shape [1,1,1,output_len].<br> The new reuse check compared only the attention sizes, so llama-perplexity,<br> which alternates 1 and 512 outputs per ubatch, reused the model compiled<br> for 512 outputs on a 1-output ubatch and failed with "got [1,1,1,1]<br> expecting [1,1,1,512]". Store the output count of the cached prefill<br> model and require it to match.</p> <ul> <li> <p>ggml-openvino: Fix LFM2 context create on GPU and DSpark speculative decoding</p> </li> <li> <p>ggml-openvino: remove leftovers from merging dev into the LFM2/DSpark fix</p> </li> </ul> <ul> <li>compute_model_outputs exported OUTPUT-flagged nodes twice: dev's block<br> ran before the new one and still exported result_norm, so the<br> result_norm exclusion had no effect. Keep only the new block.</li> <li>compute_node_dynamic_dims checked is_activation_input twice in a row.<br> Remove the second copy.</li> <li>process_view_input_new created the dynamic Split on split_dim. With<br> stateful execution the input drops the leading size-1 axis, so use<br> split_axis, which already accounts for the rank difference.</li> </ul> <ul> <li>ggml-openvino: hash the output flags into the model-cache fingerprint</li> </ul> <p>The fingerprint hashed the ops, names and weights but not the outputs, so<br> a blob compiled with another output set was imported. With the result_norm<br> change, an NPU blob written before it gave wrong output, and enabling<br> llama_set_embeddings_layer_inp() reused a blob without the layer-input<br> outputs. Hash each node's output flag and bump the cache version to 8.</p> <ul> <li>ggml-openvino: support f16 SCALE and f16/bf16 TRI</li> </ul> <p>test-backend-ops now runs SCALE on f16 and TRI on f16 and bf16. Both<br> translators built their constants as f32, and Multiply, Add and Select<br> need all inputs in one element type, so building the graph threw for<br> these cases.</p> <p>SCALE now converts a non-f32 input to f32, applies the scale and bias,<br> and converts back once, as the CPU backend computes it. TRI creates its<br> zero constant in the input's element type. f32 graphs are unchanged.</p> <ul> <li>ggml-openvino: [Qwen3.6 A3B]fix static-shape ids view of the last MoE<br> layer</li> </ul> <p>The same-stride single-dim Slice in process_view_input_new was followed by a Reshape<br> whose target baked the static chunk size into the token axis. After the output-ids<br> GET_ROWS the last layer runs on n_outputs tokens (1 in the prefill graph), so the<br> Reshape failed shape validation for qwen35moe with FORCE_STATIC or on the NPU.<br> Skip the Reshape in static graphs when the slice already keeps the rank;<br> the dynamic graphs are unchanged.</p> <ul> <li> <p>enable mamba based models on OV CPU-GPU</p> </li> <li> <p>revert changes in tests</p> </li> <li> <p>docs : add Falcon-H1 to the OpenVINO validated models</p> </li> </ul> <p>Falcon-H1 runs on CPU and GPU since Mamba SSM support was added. Add its<br> row and the known limitations: multi-sequence batches fail, GPU accuracy<br> is slightly lower than CPU, and the NPU fails with an element type<br> mismatch.</p> <ul> <li>ggml-openvino: simplify the SSM_SCAN build check and clean up</li> </ul> <ul> <li>Drop the SelectiveSSM configure check and its guards: OpenVINO 2026.4.1<br> always provides the op. Drop the no-op LANGUAGE CXX source property.</li> <li>Drop the semicolon after OPENVINO_OP("SelectiveSSM"). GCC 15 reports it<br> with -Wextra-semi, which fails builds with fatal warnings.</li> <li>Add ggml_openvino_is_static() for the repeated NPU or<br> GGML_OPENVINO_FORCE_STATIC check.</li> <li>Share the f16 and f32 KQ mask padding code.</li> <li>Drop the redundant inline on in-class static functions.</li> </ul> <hr> <p>Co-authored-by: Suhas Somashekar <a href="mailto:suhas.somashekar@intel.com">suhas.somashekar@intel.com</a><br> Co-authored-by: yzhou51 <a href="mailto:yaowei.zhou@intel.com">yaowei.zhou@intel.com</a><br> Co-authored-by: Ananth, Swathi <a href="mailto:swathi.ananth@intel.com">swathi.ananth@intel.com</a><br> Co-authored-by: Beheshti <a hr

  • b12552 ggml-org/llama.cpp releases · 20 thg 11, 2026
33

@badlogicgames his approach is amazing. it's in fact so amazing that i made an entire framework around that approach a year ago

55

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

devagrawal09 0 bài viết · X
Nguyên văn nguồn
Chi tiết nguồn

@badlogicgames his approach is amazing. it's in fact so amazing that i made an entire framework around that approach a year ago https://devagrawal09.github.io/specter/

34

b12551

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>cuda: Scatter kernel for set_n (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6450527253" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/32489" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/32489/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/32489">#32489</a>)</p> <ul> <li> <p>cuda: add scatter for batched tensor writes</p> </li> <li> <p>update test case to mirror get_n test case</p> </li> <li> <p>save restore timing logs</p> </li> <li> <p>refactor code into batch-copy.cu file</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/69865616">https://github.com/ggml-org/llama.cpp/attestations/69865616</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-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/b12551/llama-b12551-bin-macos-x64.tar.gz">macOS Intel (x64)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-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/b12551/llama-b12551-bin-ubuntu-x64.tar.gz">Ubuntu x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-ubuntu-arm64.tar.gz">Ubuntu arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-ubuntu-s390x.tar.gz">Ubuntu 24.04 s390x (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-bin-ubuntu-vulkan-x64.tar.gz">Ubuntu x64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-ubuntu-vulkan-arm64.tar.gz">Ubuntu arm64 (Vulkan)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-ubuntu-cuda-12.8-x64.tar.gz">Ubuntu x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/cudart-llama-b12551-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/b12551/llama-b12551-bin-ubuntu-cuda-13.5-x64.tar.gz">Ubuntu x64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/cudart-llama-b12551-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/b12551/llama-b12551-bin-ubuntu-cuda-13.5-arm64.tar.gz">Ubuntu arm64 (CUDA 13)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/cudart-llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-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/b12551/llama-b12551-bin-android-arm64.tar.gz">Android arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-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/b12551/llama-b12551-bin-win-cpu-x64.zip">Windows x64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-win-cpu-arm64.zip">Windows arm64 (CPU)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-win-opencl-adreno-arm64.zip">Windows arm64 (OpenCL Adreno)</a></li> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/llama-b12551-bin-win-cuda-12.4-x64.zip">Windows x64 (CUDA 12)</a> - <a href="https://github.com/ggml-org/llama.cpp/releases/download/b12551/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/b12551/llama-b12551-bin-win-cuda-13.5-x64.zip">Windows x64 (CUDA 13)</a> - <a href="https://github.com/g

  • b12551 ggml-org/llama.cpp releases · 20 thg 11, 2026
35

langchain-openai==1.8.4

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 · langchain-ai/langchain releases
Nguyên văn nguồn
Chi tiết nguồn

<p>Changes since langchain-openai==1.8.3</p> <p>release(openai): 1.8.4 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6499336260" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41962" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41962/hovercard" href="https://github.com/langchain-ai/langchain/pull/41962">#41962</a>)<br> chore(langchain): require <code>langgraph</code> 1.2.16 or newer (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6495337612" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41959" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41959/hovercard" href="https://github.com/langchain-ai/langchain/pull/41959">#41959</a>)<br> fix(openai): classify in-stream server errors as retryable (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6480654386" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langchain/issues/41952" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langchain/pull/41952/hovercard" href="https://github.com/langchain-ai/langchain/pull/41952">#41952</a>)</p>

36

Join us next Tuesday, November 24 in San Francisco for an evening with the teams building open weight models

55

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

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

Join us next Tuesday, November 24 in San Francisco for an evening with the teams building open weight models We'll kick things off with the latest Artificial Analysis benchmark data, followed by lightning talks and a panel. Register 👇 https://luma.com/artifi-mkho?utm_source=x

37

Amazon SageMaker AI now detects latency drift, idle GPUs, and missed optimizations on your inference endpoints

55

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

Hazim Qudah 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p>Every inference endpoint is well tuned the day it goes into production. Teams choose an instance type, serving configuration, and scaling settings that suit the model and its traffic at launch. With the endpoint serving smoothly, they understandably move on to the next priority. The workload, however, rarely stays still. Traffic patterns evolve, prompts grow longer, serving engines such as vLLM ship new releases, SageMaker inference develops new optimization techniques, and new fine-tuned checkpoints change the resources your model needs. Over time, a well-tuned endpoint quietly drifts. Latency and time to first token edge upward. A configuration that was never quite ideal keeps missing available performance gains. A GPU provisioned for demand that never materialized sits idle week after week. Because nothing has failed, no alarm sounds, and the drift goes unnoticed until users feel it or the bill reflects it.</p> <p><a href="https://aws.amazon.com/sagemaker/ai/" target="_blank" rel="noopener">Amazon SageMaker AI</a> now monitors your generative AI inference endpoints and tells you when a change could improve performance or lower cost by using <a href="https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-ai-now-supports-optimized-generative-ai-inference-recommendations/" target="_blank" rel="noopener">Generative AI optimized inference recommendations</a>&nbsp;feature with Agentic experience in &nbsp;<a href="https://aws.amazon.com/sagemaker/ai/studio/" target="_blank" rel="noopener">Amazon SageMaker Studio</a>. Monitoring is on by default for SageMaker AI inference endpoints using vLLM serving containers and GPU instances, with no setup required, and you can opt out of endpoints. SageMaker AI looks only at operational signals: latency, GPU utilization, and the endpoint’s serving configuration. It does not read your requests or responses. From those signals, it surfaces three kinds of findings:</p> <ul> <li><strong>Latency drift:</strong> A week-over-week rise in median (p50) end-to-end latency or time to first token.</li> <li><strong>Unused capacity:</strong> GPUs you pay for that sit mostly idle.</li> <li><strong>Configuration gaps:</strong> A serving container that’s missing newer performance improvements or security patches, or an optimization such as speculative decoding or GPU kernel tuning that your model supports but the endpoint doesn’t use.</li> </ul> <p>SageMaker AI sends a finding through AWS Health, with one notification per endpoint that covers that endpoint’s findings. When a notification arrives, you can act on it in one place. It opens the SageMaker AI agentic experience in Amazon SageMaker Studio with the endpoint, and its recent traffic already filled in. From there, you run inference recommendations to benchmark your current configuration against alternatives and compare them side by side on response time, throughput, and cost per million tokens. That turns a judgment call into a decision backed by data. Benchmarks run on separate compute, and nothing changes on your endpoint until you choose to apply a result.</p> <p>In this post, we show how SageMaker AI monitors your endpoints and how to read the notifications it sends through AWS Health. We then show how to go from a notification to a benchmarked decision in SageMaker Studio. We walk through three example scenarios, show how to apply or dismiss a recommendation safely, and explain how to opt an endpoint out of monitoring.</p> <h2 id="solution-overview">Solution overview</h2> <p>SageMaker AI inference recommendations follow a four-step loop: observe, notify, recommend, and decide. Each step is designed so that you stay in control, and your production endpoint isn’t changed without your explicit action.</p> <ol type="1"> <li><strong>Observe –</strong> Monitoring of latency and capacity usage is on by default for eligible endpoints. It doesn’t read your requests or responses, and it carries no additional GPU cost.</li> <li><strong>Notify –</strong>&nbsp;When there is something worth acting on, you receive a plain language notification in AWS Health, plus email, explaining what was observed, why it matters, and a recommended next step.</li> <li><strong>Recommend (opt in) –</strong>&nbsp;One selection opens the SageMaker AI agentic experience in SageMaker Studio, prefilled with the endpoint and its observed traffic, so you can benchmark alternatives on separate compute.</li> <li><strong>Decide –</strong>&nbsp;Compare response time, throughput, and cost side by side. Production is not changed until you apply a result.</li> </ol> <p>The feature produces a few types of findings. The following table summarizes what triggers each one and what gets tested when you choose to benchmark.</p> <table border="1px" width="100%" cellpadding="10px"> <tbody> <tr> <td><strong>Finding</strong></td> <td><strong>What triggers it</strong></td> <td><strong>What a benchmark tests</strong></td> </tr> <tr> <td>Latency drift</td> <td>Response time moves relative to the endpoint’s week-over-week trends</td> <td>Tensor parallelism settings and alternate instance types or sizes</td> </tr> <tr> <td>Underutilized GPUs</td> <td>GPU utilization stays well below capacity over the window</td> <td>Cheaper instance types and smaller sizes at observed peak traffic</td> </tr> <tr> <td>Speculative decoding not enabled</td> <td>Model and container support it, but it is off (configuration based)</td> <td>Current configuration compared to speculative decoding enabled</td> </tr> <tr> <td>Serving container</td> <td>The endpoint’s serving container has a newer version with a performance improvement and security patches for critical and high finding Common Vulnerabilities and Exposures (CVEs).</td> <td>Current compared to updated container</td> </tr> <tr> <td>Untuned GPU kernels</td> <td>Kernel configuration is not tuned for the model and hardware</td> <td>Default compared to tuned kernels</td> </tr> </tbody> </table> <h2 id="prerequisites">Prerequisite

38

check out the results:

55

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

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

check out the results: https://crusbench.com

39

Inside Amazon Connect Talent: The science of structured AI interviews

55

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

Sunitha Srivatsa 1 bài viết · AWS Machine Learning Blog
Nguyên văn nguồn
Chi tiết nguồn

<p><span class="TextRun SCXW252313659 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW252313659 BCX0">For organizations hiring </span><span class="NormalTextRun SCXW252313659 BCX0">hundreds or </span><span class="NormalTextRun SCXW252313659 BCX0">thousands of people, every interview presents the same challenge: How do you give candidates a consistent opportunity to </span><span class="NormalTextRun SCXW252313659 BCX0">demonstrate</span><span class="NormalTextRun SCXW252313659 BCX0"> their skills while gathering the evidence needed to make informed hiring decisions?</span></span><span class="EOP Selected SCXW252313659 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span class="TextRun SCXW193008649 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW193008649 BCX0">Structured behavioral interviews offer a research-backed approach to this challenge. They use consistent, job-related questions and clear scoring criteria to give candidates an opportunity to </span><span class="NormalTextRun SCXW193008649 BCX0">demonstrate</span><span class="NormalTextRun SCXW193008649 BCX0"> their skills. Decades of research in industrial-organizational (I-O) psychology have shown that structured interviews help organizations evaluate candidates and predict job performance (Schmidt &amp; Hunter, 1998; Sackett et al., 2022). Yet delivering </span><span class="NormalTextRun SCXW193008649 BCX0">structured interviews</span><span class="NormalTextRun SCXW193008649 BCX0"> at scale requires </span><span class="NormalTextRun SCXW193008649 BCX0">significant time</span><span class="NormalTextRun SCXW193008649 BCX0"> and </span><span class="NormalTextRun SCXW193008649 BCX0">expertise</span><span class="NormalTextRun SCXW193008649 BCX0"> to develop</span><span class="NormalTextRun SCXW193008649 BCX0"> competencies, </span><span class="NormalTextRun SCXW193008649 BCX0">create q</span><span class="NormalTextRun SCXW193008649 BCX0">uestions, train interviewers, and evaluate responses consistently</span><span class="NormalTextRun SCXW193008649 BCX0">.</span></span><span class="EOP Selected SCXW193008649 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span data-contrast="auto"><a href="https://aws.amazon.com/products/connect/talent/" target="_blank" rel="noopener">Amazon Connect Talent</a> <span class="NormalTextRun SCXW175523192 BCX0">brings established hiring science to AI-led interviews at scale. </span><span class="NormalTextRun SCXW175523192 BCX0">Connect Talent </span><span class="NormalTextRun SCXW175523192 BCX0">uses structured questions, adaptive follow-ups, and evidence-based scoring to help organizations gather consistent, job-related information while giving more candidates the opportunity to demonstrate their strengths.</span></span></p> <p><span class="TextRun SCXW66502508 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW66502508 BCX0">We built Connect Talent around three core design principles and evaluated how well the service </span><span class="NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW66502508 BCX0">applies</span><span class="NormalTextRun SCXW66502508 BCX0"> them through </span><span class="NormalTextRun SCXW66502508 BCX0">a combination of </span><span class="NormalTextRun SCXW66502508 BCX0">internal research and independent expert review. </span><span class="NormalTextRun SCXW66502508 BCX0">The </span><span class="NormalTextRun SCXW66502508 BCX0">findings offer a closer look at the science behind </span><span class="NormalTextRun SCXW66502508 BCX0">Connect Talent’s </span><span class="NormalTextRun SCXW66502508 BCX0">AI-le</span><span class="NormalTextRun SCXW66502508 BCX0">d,</span><span class="NormalTextRun SCXW66502508 BCX0"> structured interviewing.</span></span><span class="EOP Selected SCXW66502508 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <h2><b><span data-contrast="auto">Three core design principles</span></b></h2> <h3><b><span data-contrast="auto">1. Build interviews around job-related evidence</span></b></h3> <p><span class="TextRun SCXW227730746 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW227730746 BCX0">Connect Talent builds each interview around the competencies </span><span class="NormalTextRun SCXW227730746 BCX0">required</span><span class="NormalTextRun SCXW227730746 BCX0"> for the role. The service uses job requirements and </span><span class="NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW227730746 BCX0">established</span><span class="NormalTextRun SCXW227730746 BCX0"> occupational frameworks to create behaviorally anchored scorecards. Each scorecard defines specific strength and concern signals, and every interview question targets the behaviors it aims to uncover.</span></span><span class="EOP Selected SCXW227730746 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <p><span class="TextRun SCXW128739577 BCX0" lang="EN-US" xml:lang="EN-US" data-contrast="auto"><span class="NormalTextRun SCXW128739577 BCX0">The AI interviewer also adapts to each response. After every answer, it checks whether the candidate has provided enough evidence for the competency being assessed. When details are missing, it asks targeted follow-up questions. This approach preserves the structure of the interview while giving candidates more opportunities to explain their experience.</span></span><span class="EOP Selected SCXW128739577 BCX0" data-ccp-props="{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:278}">&nbsp;</span></p> <h3><b><span data-contrast="auto">2. Separate interviewing from scoring</span></b></h3> <p><span c

40

b12550

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>ggml : add backend scheduler tests (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5171818789" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/27258" data-hovercard-type="pull_request" data-hovercard-url="/ggml-org/llama.cpp/pull/27258/hovercard" href="https://github.com/ggml-org/llama.cpp/pull/27258">#27258</a>)</p> <ul> <li> <p>Adds initial scheduler stress tests</p> </li> <li> <p>Adds new sleep op to simplify scheduler testing</p> </li> <li> <p>Adds testcase addressing <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="4474889411" data-permission-text="Title is private" data-url="https://github.com/ggml-org/llama.cpp/issues/23321" data-hovercard-type="issue" data-hovercard-url="/ggml-org/llama.cpp/issues/23321/hovercard" href="https://github.com/ggml-org/llama.cpp/issues/23321">#23321</a></p> </li> <li> <p>Adds experimental vulkan GGML_OP_SLEEP</p> </li> <li> <p>Improves structure and naming of previous tests. Adds new tests,<br> including one which detects the nkvo-bug correctly, and validates the<br> proposed hotfix.</p> </li> <li> <p>Cleanup: Removes CUDA specific comments and nvtx-markers</p> </li> <li> <p>Fix: forgot to add sleep specific test case</p> </li> <li> <p>Fix oversights identified by CI</p> </li> <li> <p>Drops default prints so that it behaves like other ctests. Adds -v flag<br> to toggle this behavior</p> </li> <li> <p>Fix: Early exit in CPU-only settings (cause for graviton CI fails)</p> </li> <li> <p>Improves stress_test_linked_list by removing overlap with<br> test_chain_all_backends and test_pair_user_inputs. The latter ones test<br> either activations or user inputs in isolation and just once, whereas<br> the stress test tests all combinations multiple rounds</p> </li> <li> <p>Exclude ACCEL devices from new tests due to incompatibility</p> </li> <li> <p>Fixes vulkan test failure in llvmpipe</p> </li> <li> <p>Adds asserts, improves comments and some checks</p> </li> <li> <p>Fixes nitpicks in white-space and variable naming</p> </li> <li> <p>Adds comments that certain vk drivers could time OP_SLEEP incorrectly</p> </li> <li> <p>Drops limitiation to CPU/GPU in test-sleep</p> </li> <li> <p>Aligns OP_SLEEP guards in cuda with vk guards (input shape and<br> contiguity)</p> </li> <li> <p>Implements feedback for OP_SLEEP to work in place (no copy)</p> </li> <li> <p>Apply non-code suggestions from JohannesGaesslers code review</p> </li> </ul> <p>Co-authored-by: Johannes Gäßler <a href="mailto:johannesg@5d6.de">johannesg@5d6.de</a></p> <ul> <li> <p>Removes struct backend_supports; enables tests to return skip, instead<br> of only pass/fail</p> </li> <li> <p>Make clang-tidy happy</p> </li> <li> <p>Removes sched_backend_caps, improves create_test_scheduler according to<br> feedback (L149)</p> </li> <li> <p>Removes "name" field from struct sched_check in favor of<br> ggml_set_name/ggml_get_name</p> </li> <li> <p>Drops unnecessary shader_clock.comp test and guards</p> </li> <li> <p>Replaces sched_ prefixes with helper_ prefix</p> </li> <li> <p>Moves test suite to new RAII types for better memory management</p> </li> <li> <p>Adds sleep option to test_chain_all_backends</p> </li> <li> <p>Adds sleep option to first-run arm of the y-shaped test</p> </li> <li> <p>Fix switchcase order for legibility in stress_test_dag</p> </li> <li> <p>Simplfies stress_test_linked_list via double for-loop</p> </li> <li> <p>Adds pseudo-random input to stress_test_linked_list, and runs all<br> transitions both with and without CPU input (next to activation copy as<br> input)</p> </li> <li> <p>Fix RPC Proto patch due to GGML_OPS bump</p> </li> <li> <p>Adds ggml-cpp.h treatment to test-sleep.cpp</p> </li> <li> <p>Adds TODO pointers for subsequent additional tests.</p> </li> <li> <p>Apply suggestion to ggml.h</p> </li> </ul> <p>Co-authored-by: Johannes Gäßler <a href="mailto:johannesg@5d6.de">johannesg@5d6.de</a></p> <ul> <li>Apply suggestion to tests/test-backend-sched.cpp</li> </ul> <p>Co-authored-by: Johannes Gäßler <a href="mailto:johannesg@5d6.de">johannesg@5d6.de</a></p> <ul> <li>Apply suggestion to tests/test-backend-sched.cpp</li> </ul> <p>Co-authored-by: Johannes Gäßler <a href="mailto:johannesg@5d6.de">johannesg@5d6.de</a></p> <ul> <li> <p>Fixes node compatibility check to not take buffer locations into account.<br> This led to skipped tests due to incorrectly rejected ops due to wrong buffer locations in multi-GPU (which is one of the scheduler behaviors under test)</p> </li> <li> <p>Apply batched suggestions from gg code review</p> </li> </ul> <p>Co-authored-by: Georgi Gerganov <a href="mailto:ggerganov@gmail.com">ggerganov@gmail.com</a></p> <ul> <li> <p>Completes callsite change of gg-suggestion</p> </li> <li> <p>Switch test-sleep from std::chrono::steady_clock::now() to ggml_time_us<br> to avoid mac clock drift and spurious test failures.<br> This makes the test a bit nonsensical on CPU (same clock checks itself),<br> but the test still has merit on the other backends where<br> independent device clocks are used.</p> </li> </ul> <hr> <p>Co-authored-by: Johannes Gäßler <a href="mailto:johannesg@5d6.de">johannesg@5d6.de</a><br> 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/69858536">https://github.com/ggml-org/llama.cpp/attestations/69858536</a></li> </ul> <p><strong>macOS/iOS:</strong></p> <ul> <li><a href="https://github.com/ggml-org/llama.cpp/releases/download/b12550/llama-b12550-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_

  • b12550 ggml-org/llama.cpp releases · 20 thg 11, 2026
21–40 / 213
Trang 2 / 11