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

Tác tử

Last year we built megakernels. This year we deleted them: The road to Together Compiler

Together Compiler uses AI agents to rewrite and fuse the GPU kernels in a model’s forward pass. It beats hand-written megakernels by 28% on B200 (2,118 vs 1,649 tokens/s on Llama 3.2 1B). This is how we got from one to the other.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

datasette-maplibre 0.1a0

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-maplibre/releases/tag/0.1a0">datasette-maplibre 0.1a0</a></p> <p>The first new Datasette geospatial plugin in quite a while! This one combines the excellent <a href="https://maplibre.org/maplibre-gl-js/docs/">MapLibre GL</a> library with <a href="https://openfreemap.org">OpenFreeMap</a>, the free vector map server operated by Zsolt Ero (you should <a href="https://github.com/sponsors/hyperknot">sponsor Zsolt on GitHub</a>.)</p> <p>I built most of this with a single prompt to Claude Fable 5.2:</p> <blockquote> <p><code>Look at datasette-cluster-map and datasette-leaflet-geojson plugins, then look at ~/dev/tools/openfreemap-demo.html - then build a brand new Datasette plugin that uses maplibreGL and openfreemap and combines the features of those other two plugins - but be creative and think outside the box for other useful features the plugin might have</code></p> </blockquote> <p>"Be creative and think outside the box" resulted in a few neat ideas - my favorite is that it displays SVG previews of GeoJSON shapes, then loads a WebGL preview only when they scroll into view.</p> <p>I also had Claude build a comprehensive demo which works in the latest Datasette Lite, so you can <a href="https://lite.datasette.io/?install=datasette-maplibre&amp;url=https://gist.githubusercontent.com/simonw/39fa5d399465766039549912e2b3d291/raw/f6dfba99a75cbb8b5d9386223220eb52272767ee/demo.db&amp;metadata=https://gist.githubusercontent.com/simonw/39fa5d399465766039549912e2b3d291/raw/f6dfba99a75cbb8b5d9386223220eb52272767ee/metadata.yml">visit this link</a> to load the plugin and a demo database in your browser via Pyodide and WebAssembly.</p> <p>I like <a href="https://lite.datasette.io/?install=datasette-maplibre&amp;url=https://gist.githubusercontent.com/simonw/39fa5d399465766039549912e2b3d291/raw/f6dfba99a75cbb8b5d9386223220eb52272767ee/demo.db&amp;metadata=https://gist.githubusercontent.com/simonw/39fa5d399465766039549912e2b3d291/raw/f6dfba99a75cbb8b5d9386223220eb52272767ee/metadata.yml#/demo/capitals_and_countries%23map=0/4.2/180&amp;color=continent">this demo page</a> showing both GeoJSON shapes and lat/lon points on the same map:</p> <p><img alt="Screenshot of a web map interface. Toolbar: &quot;View&quot; set to &quot;Clusters&quot;, &quot;Colour by&quot; set to &quot;continent&quot;, plus &quot;Filter to this area&quot; and &quot;Download GeoJSON&quot; buttons. A world map shows countries outlined and shaded by continent, with coloured dots and numbered cluster circles. A popup for a point in South America reads: capital Caracas, country Venezuela, continent South America, capital_population 2985000, latitude 10.502944, longitude -66.918983. Legend: Africa (52), Asia (45), Europe (39), North America (16), South America (13), Oceania (6). Text: &quot;All 171 rows are on the map&quot;. Below, a table with columns country, continent, capital_population, latitude, longitude, boundary. First row: Afghanistan, Asia, 3277000, 34.518636, 69.181314, with a small map in the boundary column showing Afghanistan outlined and shaded in blue." src="https://static.simonwillison.net/static/cors-allow/2026/datasette-maplibre.webp" /></p> <p>Tags: <a href="https://simonwillison.net/tags/visualization">visualization</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/datasette-lite">datasette-lite</a>, <a href="https://simonwillison.net/tags/claude-mythos-fable">claude-mythos-fable</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/geospatial">geospatial</a>, <a href="https://simonwillison.net/tags/openstreetmap">openstreetmap</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

datasette 0.65.6

<p><strong>Release:</strong> <a href="https://github.com/simonw/datasette/releases/tag/0.65.6">datasette 0.65.6</a></p> <p>I backported a minor security fix from 1.0a43:</p> <blockquote> <p>Fixed a security issue where a database registered using <code>add_database(db, name=..., route=...)</code> with a route that differs from its name could be downloaded from <code>/&lt;route&gt;.db</code> by actors who were denied access to it. [...]</p> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/security">security</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Our new blog for science

The OpenAI science blog shares new findings, practical methods, and firsthand accounts from researchers using AI to advance scientific discovery.

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

Announcing Humming: Fast, Flexible Kernels for Quantized Inference

Introducing Humming, a quantized inference kernel library from Ant Group and the vLLM community, with broad GPU and quantization support, dense and MoE performance results, and vLLM integration.

  • Introducing Humming, a quantized inference kernel library from Ant Group and the vLLM community, with broad GPU and quantization support, dense and MoE performance results, and vLLM integration.
Nguyên văn nguồn
Chi tiết nguồn
Kiến trúc

Understanding Positional Embeddings in Modern LLMs

Câu hỏi But why spending so much time writing/reading about this rather minor building block?

This article explains how LLMs encode and process token position information by covering classic and contemporary methods like RoPE. It’s essentially a comprehensive journey where we will work through the mechanisms with visual examples, a little math, and a small PyTorch implementation. And, of course, we will then also look at how modern architectures adapt or omit explicit positional encodings. Below, you can find a preview of which positional embedding choices are used in 2026 open-weight models. We will discuss these choices later after covering the underlying mechanisms. Figure 1: A preview of the gallery’s 2026 architectures, grouped by positional encoding. Architecture figures taken from my LLM gallery . But why spending so much time writing/reading about this rather minor building block? LLMs keep evolving, and as I was preparing a new article on the latest LLM architecture trends this year, I realized that I never covered the positional information methods used in LLMs in any article so far. So, instead of letting the other article get out of hand, I thought it might make sense to publish this as a standalone building block for the upcoming LLM architecture and trends analysis. Also, over the years, many readers have asked me for a tutorial on relative positional embeddings (aka RoPE) to follow up on the learned absolute embeddings used in GPT. It’s been on my TODO list for a long time, so I finally revisited my old draft and put this walkthrough together, which will cover different positional methods in LLMs; (finally) how RoPE works; what methods are used by recent open LLMs. Note that this article assumes some familiarity with basic LLM architectures and works well as a follow-up to Build a Large Language Model (From Scratch) . If the concepts and computation of the attention internals, particularly query, key, and value vectors, are a little fuzzy, I recommend checking my self-attention walkthrough as a refresher: The article is also very technical and likely too detailed for some. However, I really wanted to have this foundation out there for future articles to reference. And no worries, the next one will return to a more comfortable style. PS: I started experimenting with Substack’s new “perks” system. As a thank-you to supporters, I’ve added image exports and metadata from the architecture gallery for you to use in your own analyses. PPS: If you are already familiar with how position embeddings work, feel free to fast-forward to Section 6, Positional choices in modern architectures . Note that this article may be best read in a web browser than an email client. I recommend checking out the web view here , which also has a navigation bar on the left side. 1. Where position information enters an LLM Let’s start with the overall architecture. A tokenizer encodes the input text into token IDs. Next, the input token embedding layer of an LLM maps each of these IDs to a vector of numbers. We usually refer to these entries in that vector as “features” (similar to classic ML). The transformer blocks inside the LLM then modify these vectors using attention and feedforward layers. Lastly, the model’s output layer produces logit scores (converted into probabilities) for the next token. The following shows this path in GPT-2 (a relatively classic LLM) and MiniMax-M3 (a more recent LLM released this year; a placeholder for most other recent LLMs). As we can see, the architectures are not too different from each other, but before we get to the positional information parts, let’s focus on the orange blocks for now (the token embedding layers) and the transformer blocks above them; the other architecture differences aren’t needed for this explanation. Figure 2: Token embedding layers in GPT-2 and MiniMax-M3, highlighted in orange. (Side note: M3’s first three out of 60 feedforward layers are dense not MoEs.) A token embedding layer, shown in the figure above, maps each token ID to a vector. Position information is then added afterward. Concretely, and as we will discuss in more detail later, GPT-2 adds a position vector to each token embedding before the modified embedding enters the first transformer block. On the other hand, MiniMax-M3 adds the positional information in each transformer block using a method called RoPE inside attention. We will talk about these methods in more detail later, but first, let’s start with the big picture information explaining why explicit positional information is added or needed. 1.1 Why add positional information? Consider the following two input sentences as a simple example: “The red fox jumps over the gray fox.” “The gray fox jumps over the red fox.” Changing the word order changes which fox does the jumping, as illustrated in the following figure. Figure 3: Changing the word order changes which fox jumps. For the examples in this article, we will treat each word as one token for simplicity. (As you know, real tokenizers can split words into smaller subword tokens.) The first sentence then has eight token positions, from 0 through 7. Figure 4: Token IDs and positions in the toy sentence. # Assume the following small vocabulary. # Real vocabularies typically have 50k to 200k entries vocab = { "The": 0, "red": 1, "fox": 2, "jumps": 3, "gray": 4, "over": 5, "the": 6, } (If you are interested in how vocabularies are built, check out my tokenizer materials on GitHub .) Note that “fox”, which appears twice in the input sentence, gets the same token ID 2 in both cases. And since they both have the same token ID, irrespective of the position in the input sentence, they would get the same embedding vector from the embedding layer. (For those who are interested in more details on how this selection from the embedding table looks like, check Appendix A .) Now, let’s consider a regular self-attention mechanism with no positional encoding (and no causal mask), where every input token can attend to every other token in the input. In this scenario, swapping any tw

Nguyên văn nguồn
Chi tiết nguồn
Đánh giá

30 Months on the Exponential (in Charts)

My second book on AI, Co-Existence, is out today (and you can get it here ), 30 months after the release of Co-Intelligence in early 2024. Looking back across the gap between these books is a good way to get a sense of AI progress. And, thanks to AI, I can describe this progress with some quite lovely charts. Making the charts is itself a good indicator of how far things have come. In the era of chatbots, the era of Co-Intelligence , I would have had to work back-and-forth with an AI in a chat window, guiding it through each new step. In the current era of agentic AI that I cover in Co-Existence , I delegated the work and let the AI handle it. In this case, I used Opus 5.5 in Claude Projects, which allows you to talk to a single “orchestrator” agent that then spins off additional agents as needed, giving them instructions and relaying information back to you. Getting the work done was straightforward. I gave the AI a couple sentences of suggestions and it did what would have taken me hours of work (all the research, all the graphing, all the layouts, all the images, all the citations…) in minutes. And since I didn’t have to do the work, I had the AI make many options for me. I picked some and gave handwavy reactions to others that would drive a human team mad (“make it better”). It even caught that I’d gotten the math wrong. I was going to call this post “28 months on the exponential.” Some examples of interacting with Claude to make the graphics for this post. Notice that it corrected me. “This is the worst AI you will ever use” I wrote this in my last book on AI, and it has continued to be literally true at every moment since. From early 2023 through early 2024, there was a single AI model that was undeniably the best in the world: GPT-4 from OpenAI. This was the AI model that most people remember from the first couple years of AI use, it was the one I thought about while writing Co-Intelligence , and it was the one that gave ChatGPT its status as the default AI for many people. Since then, twenty different AI models have held the lead at one time or another, all from Anthropic, Google, or OpenAI. Which company had the best AI model in the world. But the changing lead alone hides an even bigger change. A total of 180 AI models are now better than GPT-4 was at its release, many of which are open weights models, which anyone is free to use or modify. Nineteen of them can run on a modern smartphone. We are living in an exponential The data below draws on the METR Long Horizon Benchmark, which used to be the most famous chart in AI. The benchmark measured the amount of human work that an AI can do with at least a 50% chance of success from a single prompt. Gains on this benchmark have been startlingly regular. From the release of GPT-4 to the end of the METR measure, the amount of work AI could do doubled eight times. In 2024, an AI could do a four minute task. At the moment I turned in my draft of Co-Existence in April of 2026, an AI could do 17 hours of human work. And then the benchmark ran out. METR said they could no longer accurately measure the amount of work an AI could do using their current methods. Other studies suggest that number might be 17 weeks of work, or more. And, even as ability has increased, prices have dropped. My old favorite AI benchmark, the Google Proof Question and Answer Exam, which gives the AI problems that regularly stump graduate students, used to cost 30 cents in AI usage to get a single correct answer. Now, it costs four hundredths of a cent to get the same result. How humans work with AI is changing Despite all the progress, the frontier of AI ability is still jagged, just as it was 30 months ago. Some areas have stayed stubbornly weak (AI is still quite a bad long-form writer), while others, like solving math problems, went from terrible to superhuman in the same period. And, as I discuss in the book, even where AI is excellent, small gaps create large needs for humans, at least for now. But AI’s gains have been so large that how we work with it has to change. In Co-Intelligence , I wrote about centaur work (divided between AI and humans) and cyborg work (blended between AI and humans). Both keep you in the loop… but you can only be in so many loops at once. Now that we have real agents, I added a third way to work with AI: the conjurer. Conjurers summon agents when needed, assign them work, and evaluates what comes back. The charts in this post were made that way: I summoned agents, gave some general guidance, and I said “make it better” a lot. It is the most powerful way to use AI, and it carries new risks as we lose control over what, exactly, the AI is doing to get us those results. The Otter Test has fallen Regular readers will know that I have been testing AI image generation and movie tools with the prompt “Otter on an Airplane Using Wifi” for the past few years. This is what the best image generator, Midjourney, produced 30 months ago. So today I asked Opus 5.5 the following (typos and all, prompting has become much less important as AIs get better): “I have a post showing the what happened over the past 30 months of AI in honor of my new book (published 30 months after the origina). i want you to make the ultimate otter on an airplane using wifi video. just something fantastic, i know you are creative, give it a go.” Using my subscriptions to various online services from Google and ElevenLabs, it produced the following. (Turn the sound on) The video has a few flaws. But the AI figured out how to make it, from the script to the images to the concept to the music to the technical aspects of creating all of this, without any help from me. That is what has changed. This is the least we will ever talk about AI As a result of all of these gains, AI is only becoming a bigger issue in the world. This timeline of achievements and controversies since Co-Intelligence came out gives you some idea of how much is happening (I am leaving out hundreds of impor

  • Now, it costs four hundredths of a cent to get the same result.
  • It is the most powerful way to use AI, and it carries new risks as we lose control over what, exactly, the AI is doing to get us those results.
  • This is the least we will ever talk about AI As a result of all of these gains, AI is only becoming a bigger issue in the world.
Nguyên văn nguồn
Chi tiết nguồn
Tác tử

datasette-apps 0.4a0

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-apps/releases/tag/0.4a0">datasette-apps 0.4a0</a></p> <p>Datasette Apps are sandboxed HTML and JavaScript applications that live inside Datasette - think Claude Artifacts, but these ones can run SQL queries against your data.</p> <p>0.4a0 adds support for linking: you can link to an app with <code>#fragments</code> in the URL and the app can see and use those, plus supported URL schemes are reported back to Datasette Agent so it can help write apps that link to each other.</p> <p>This means you can have e.g. an app that shows details about a specific record, and a completely separate app that searches those records and links to their representations in the first app.</p> <p>Tags: <a href="https://simonwillison.net/tags/datasette-agent">datasette-agent</a>, <a href="https://simonwillison.net/tags/datasette-apps">datasette-apps</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/urls">urls</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

datasette-agent 0.6a0

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-agent/releases/tag/0.6a0">datasette-agent 0.6a0</a></p> <p>Accumulated small fixes for Datasette Agent, my agent plugin for Datasette that lets you ask questions about your data. Most notably, it now adds an agent prompt directly to the Datasette homepage, with support for hash-mentions of tables and queries:</p> <p><img alt="Datasette homepage. Below the search box is a Datasette Agent heading with a text input and an ask button. User has typed: Tell me about #re - the box is showing autocomplete results for the repos and releases tables." src="https://static.simonwillison.net/static/cors-allow/2026/datasette-agent-homepage.webp" /></p> <p>Tags: <a href="https://simonwillison.net/tags/datasette-agent">datasette-agent</a>, <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

datasette-llm 0.1a11

<p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-llm/releases/tag/0.1a11">datasette-llm 0.1a11</a></p> <p><code>datasette-llm</code> is the intermediate plugin for Datasette that supports configuring LLMs for use with other Datasette plugins.</p> <p>This release adds support for OpenAI-compatible endpoints, so you can configure Datasette to use an LLM hosted anywhere that talks the OpenAI pseudo-standard web protocol.</p> <p>This means that if you're running local models with <a href="https://lmstudio.ai">LM Studio</a> or <a href="https://ollama.com">Ollama</a> or <a href="https://github.com/ggml-org/llama.cpp">llama.cpp</a> (which just hit <a href="https://huggingface.co/blog/ggml-org/llama-cpp-v1">v1.0.0</a>) you can run Datasette via <code>uvx</code> like this:</p> <div class="highlight highlight-source-shell"><pre>wget https://datasette.io/content.db uvx --with datasette-agent datasette \ --root --secret 1 \ -s plugins.datasette-llm.default_model <span class="pl-s"><span class="pl-pds">'</span>{</span> <span class="pl-s"> "model": "lmstudio-qwen",</span> <span class="pl-s"> "model_name": "qwen/qwen3.8-27b",</span> <span class="pl-s"> "api_base": "http://localhost:1234/v1"</span> <span class="pl-s"> }<span class="pl-pds">'</span></span> content.db</pre></div> <p>And you'll get a local server running a Datasette Agent instance powered by your local model!</p> <p><img alt="Screenshot of a chat interface. Header bar reads &quot;home&quot; on the left and &quot;root&quot; with a hamburger menu icon on the right. Below is a back arrow, the heading &quot;Chat&quot;, a model label &quot;lmstudio-qwen&quot; and an &quot;EXPORT&quot; button. A blue user message bubble says &quot;how many repos and how many releases are there?&quot; Below it is a collapsed tool call &quot;list_databases_and_tables&quot; and an expanded tool call &quot;sql_query&quot;. Its REQUEST section shows database &quot;content&quot;, display &quot;both&quot; and the SQL: SELECT (SELECT COUNT(*) FROM repos) AS repo_count, (SELECT COUNT(*) FROM releases) AS release_count. Its RESPONSE section shows columns repo_count and release_count with a single row of 200 and 1889. Below that is a table with columns repo_count and release_count and one row: 200, 1889, followed by a &quot;View SQL query&quot; button. The assistant reply reads &quot;There are 200 repos and 1,889 releases in the content database.&quot; At the bottom is a text input with placeholder &quot;Type a message... (# to tag a table or query)&quot; and a blue &quot;Send&quot; button." src="https://static.simonwillison.net/static/2026/lmstudio-datasette-agent.webp" /></p> <p>Tags: <a href="https://simonwillison.net/tags/uv">uv</a>, <a href="https://simonwillison.net/tags/datasette-agent">datasette-agent</a>, <a href="https://simonwillison.net/tags/llm">llm</a>, <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Tác tử

v18.20.2

<h2>@oh-my-pi/pi-ai</h2> <h3>Added</h3> <ul> <li>Added the Antigravity CLI ACP provider with managed installation, Google sign-in, refreshable distributed credentials, host tools, and retained conversations across tool steps and history changes.</li> </ul> <h2>@oh-my-pi/pi-catalog</h2> <h3>Added</h3> <ul> <li>Added Antigravity CLI models for Gemini, Claude and GPT-OSS with selectable native thinking levels.</li> </ul> <h2>@oh-my-pi/pi-coding-agent</h2> <h3>Added</h3> <ul> <li>Added the <code>antigravity-cli</code> provider for Google Antigravity over ACP, with managed binary installation, <code>/login antigravity-cli</code> Google sign-in, host tool execution, and retained conversations across tool steps and history changes.</li> </ul> <h2>@oh-my-pi/pi-utils</h2> <h3>Added</h3> <ul> <li>Added a profile-aware data directory for managed Antigravity CLI installations and isolated account state.</li> </ul> <p><strong>Full Changelog</strong>: <a class="commit-link" href="https://github.com/can1357/oh-my-pi/compare/v18.20.1...v18.20.2"><tt>v18.20.1...v18.20.2</tt></a></p>

Nguyên văn nguồn
Chi tiết nguồn
  • v18.20.2 Can Bölük — Oh My Pi releases · 15 thg 11, 2026
Tác tử

v18.20.1

<h2>@oh-my-pi/pi-agent-core</h2> <h3>Fixed</h3> <ul> <li>Corrected context cost estimates for requests that omit reasoning history, so billing estimates now include thinking text only when it is sent or reconstructed for a tool call.</li> </ul> <h2>@oh-my-pi/pi-ai</h2> <h3>Added</h3> <ul> <li>Added <code>setUnpausedProviders()</code> to <code>@oh-my-pi/pi-ai/oauth</code>, allowing applications to re-enable providers paused by authentication rules; loaded credential pools and extension logins recognize the change on their next read.</li> </ul> <h3>Changed</h3> <ul> <li>Updated Vertex AI Gemini requests to use the v1beta1 API, restoring support for long Gemini responses.</li> <li>Improved Google Antigravity quota recovery so Gemini 4 Argon access can resume after quota or plan changes when shared Gemini capacity is available.</li> <li>Improved Amazon Bedrock region selection to honor a configured regional endpoint before falling back to <code>us-east-1</code>.</li> </ul> <h3>Fixed</h3> <ul> <li>Fixed Gemini 4 responses that exceed a single output segment across Google, Vertex AI, Gemini CLI, and Antigravity, ensuring long responses continue generating and are delivered as one complete reply.</li> <li>Fixed Cursor conversations so hook-added prompt content remains in the current user message and conversation context is correctly preserved across turns.</li> <li>Fixed Amazon Bedrock streaming so malformed responses fail promptly and interrupted responses are retried rather than treated as complete.</li> <li>Fixed Claude account recovery when usage reports contain only one shared usage window.</li> <li>Fixed OpenRouter conversations with Kimi and DeepSeek so reasoning context is preserved when switching models.</li> <li>Fixed Xiaomi MiMo Token Plan authentication to use the correct regional endpoint, including for models loaded from bundled or cached configuration.</li> <li>Fixed Perplexity sign-in in the login flow and setup wizard, including email-code and browser-based SSO authentication.</li> <li>Stopped native judges from retrying gateway errors caused by exhausted upstream credentials while retaining retries for other transient 502 errors.</li> <li>Fixed simple completion and authenticated gateway chat routes so supported providers receive stop-sequence and frequency-penalty settings.</li> <li>Fixed Muse sessions after credential changes, key rotation, or idle resume so subsequent turns can continue normally.</li> </ul> <h2>@oh-my-pi/pi-catalog</h2> <h3>Added</h3> <ul> <li>Added Gemini 4 Hydrogen on Google Antigravity as a single model with selectable low, medium, high, and xhigh thinking levels, priced at $1.50/$7.50 per million tokens.</li> <li>Added Chutes provider support with live model discovery and <code>/login chutes</code> authentication using <code>CHUTES_API_KEY</code>.</li> </ul> <h3>Changed</h3> <ul> <li>Paused Google Antigravity: login is no longer available and stored credentials are not used until the provider is re-enabled. Use <code>GEMINI_API_KEY</code> with Google AI Studio or the <code>google-vertex</code> provider instead (<a href="https://github.com/can1357/oh-my-pi/issues/18341" data-hovercard-type="issue" data-hovercard-url="/can1357/oh-my-pi/issues/18341/hovercard">#18341</a>).</li> <li>Updated Command Code model pricing and usage reporting to match the provider's current rates, including long-context pricing tiers for Claude Haiku 5.5, GPT-6.1 Sol and Terra, Grok 4.7, and Claude Sonnet 5.5.</li> <li>Increased the maximum output for Gemini 4 Argon, Carbon, and Hydrogen on the Gemini API and Vertex AI to 262,144 tokens, enabling substantially longer responses.</li> <li>Updated OpenCode Go Claude Haiku 5.5 to support thinking levels from low through max, including explicit thinking disablement.</li> <li>Updated Devin requests to identify as Devin CLI 3000.12.57, the current stable version.</li> </ul> <h3>Fixed</h3> <ul> <li>Fixed Gemini 4 Argon on Command Code so image attachments are preserved and usage is billed correctly at $2 per million input tokens, $10 per million output tokens, and $0.10 per million cache-read tokens (<a href="https://github.com/can1357/oh-my-pi/pull/19125" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/19125/hovercard">#19125</a> by <a href="https://github.com/wolfiesch">@wolfiesch</a>).</li> <li>Fixed Factory Droid GPT-6 models failing on the next turn after changing the thinking level during Bedrock or Snowflake sessions (<a href="https://github.com/can1357/oh-my-pi/pull/19188" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/19188/hovercard">#19188</a> by <a href="https://github.com/will-bogusz">@will-bogusz</a>).</li> <li>Fixed OpenRouter model listings for workspace-restricted API keys so each key shows its correct allowed model roster and maintains an independent model cache (<a href="https://github.com/can1357/oh-my-pi/pull/19013" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/19013/hovercard">#19013</a> by <a href="https://github.com/YibaiMeng">@YibaiMeng</a>).</li> <li>Fixed OpenCode Go Muse Spark tool-call turns losing encrypted reasoning between steps (<a href="https://github.com/can1357/oh-my-pi/pull/17748" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/17748/hovercard">#17748</a> by <a href="https://github.com/nico2525nn">@nico2525nn</a>).</li> <li>Fixed Vertex AI Gemini 3 and newer models rejecting replayed or cross-model tool calls that lack a thought signature (<a href="https://github.com/can1357/oh-my-pi/pull/17150" data-hovercard-type="pull_request" data-hovercard-url="/can1357/oh-my-pi/pull/17150/hovercard">#17150</a> by <a href="https://github.com/duck1eee">@duck1eee</a>).</li> <li>Fixed Cursor model listings omitting models available to the provider, including GLM 5.3 and Grok 4.7 (<a href="https://github.com/can1357/oh-my-pi/pull/18248" data-hovercard-type="pull_request" data-hovercard-url="/can1

Nguyên văn nguồn
Chi tiết nguồn
  • v18.20.1 Can Bölük — Oh My Pi releases · 15 thg 11, 2026
Kỹ thuật

datasette-vega 0.7

<p><strong>Release:</strong> <a href="https://github.com/simonw/datasette-vega/releases/tag/0.7">datasette-vega 0.7</a></p> <p>A new release of Datasette's longest untouched plugin, for rendering charts (using <a href="https://vega.github.io/vega-lite/">Vega-Lite</a>) against data in a table or custom query. The last release was in July 2018!</p> <p>I finally bumped it for compatibility with the latest Datasette alphas, plus some cosmetic tweaks to make it work better with <a href="https://lite.datasette.io/">Datasette Lite</a>.</p> <p>(Datasette Lite just gained the ability to run JavaScript plugins like this! I'll write more about that soon.)</p> <p>Since it works with Datasette Lite, the live demo in the README now <a href="https://lite.datasette.io/?install=datasette-vega&amp;csv=https%3A%2F%2Fraw.githubusercontent.com%2Ffivethirtyeight%2Fdata%2Fmaster%2Fairline-safety%2Fairline-safety.csv#/data/airline-safety?_size=max#g.mark=bar&amp;g.x_column=airline&amp;g.x_type=ordinal&amp;g.y_column=fatalities_00_14&amp;g.y_type=quantitative">links to this demo on Datasette Lite</a>.</p> <p><img alt="Screenshot of a charting tool. A panel headed &quot;Charting options&quot; has radio buttons for &quot;Bar&quot; (selected), &quot;Line&quot; and &quot;Scatter&quot;. Below are select menus: &quot;X column&quot; set to &quot;airline&quot; with &quot;Type&quot; set to &quot;Label&quot;, and &quot;Y column&quot; set to &quot;fatalities_00_14&quot; with &quot;Type&quot; set to &quot;Numeric&quot;. A &quot;Swap X and Y&quot; button sits beneath those, followed by &quot;Color&quot; and &quot;Size&quot; menus both set to &quot;-- none --&quot;. Underneath the panel is a blue bar chart with airline names along the X axis and fatalities_00_14 on the Y axis, running from 0 to 550 - most airlines have no bar, a handful have bars, and Malaysia Airlines is the tallest." src="https://static.simonwillison.net/static/2026/datasette-vega.webp" /></p> <p>Tags: <a href="https://simonwillison.net/tags/datasette">datasette</a>, <a href="https://simonwillison.net/tags/datasette-lite">datasette-lite</a></p>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

v0.34.0rc4

<p>[Release][Tests] Give the <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6357025853" data-permission-text="Title is private" data-url="https://github.com/vllm-project/vllm/issues/67296" data-hovercard-type="pull_request" data-hovercard-url="/vllm-project/vllm/pull/67296/hovercard" href="https://github.com/vllm-project/vllm/pull/67296">#67296</a> warmup-gate test runner num_speculat…</p>

Nguyên văn nguồn
Chi tiết nguồn
  • v0.34.0rc4 vllm-project/vllm releases · 14 thg 11, 2026
Kiến trúc

Why teaching AI about symmetry could make prenatal diagnosis more reliable

<p dir="ltr">Despite millions of years of evolution, the development of the human fetus remains a complex process that is vulnerable to genetic, environmental, and metabolic disruptions. These disruptions could result in developmental abnormalities, starting with one of the first organs to begin developing in the fetus that continues to develop after birth: the brain.</p><p dir="ltr">In a routine prenatal ultrasound, the sonographer will flag possible brain abnormalities for additional magnetic resonance imaging (MRI). But imperfect imaging can result in missed or incorrect diagnoses.&nbsp;</p><p dir="ltr">To assist with the early detection and treatment of these abnormalities, researchers at MIT, Boston Children’s Hospital, and Inria have developed and trained a deep learning model called E(3)-Pose that helps image the fetus’s head while it’s in motion to capture a high-resolution image that the radiologist can examine. A paper about the work was presented at the <a href="https://eccv.ecva.net/">2026 European Conference on Computer Vision</a>.&nbsp;</p><p dir="ltr">“One of the problems with having the baby move is that the slices are all over the place,” says first author Ramya Muthukrishnan. Muthukrishnan is an MIT PhD student in the lab of Sunlin (1966) and Priscilla Chou Professor of Electrical Engineering and Computer Science Polina Golland, who is a principal investigator at the <a href="https://jclinic.mit.edu/">MIT Abdul Latif Jameel Clinic for Machine Learning in Health</a> (Jameel Clinic) and the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL).</p><p dir="ltr">By slices, Muthukrishnan is referring to a series of 2D images taken during a fetal MRI scan. These 2D planes stacked together effectively make up one high-resolution, 3D image of the fetus’s head. But if the head shifts during imaging, each 2D image “slice” could end up being taken from a different angle, resulting in gaps in the final scan and unusual slice orientation. These problems may cause the physician reviewing the MRI to miss or incorrectly diagnose a pathology in the fetal brain.</p><p dir="ltr">“There are very fine details that you need a very good volumetric picture of the brain to see,” explains Muthukrishnan. “If you have coverage gaps, you can’t see critical regions, and you need to be able to see well to detect abnormal development.”</p><p dir="ltr">What makes E(3)-Pose particularly good at tracking fetal motion is its focus on physics. Rather than simply training the model to learn physics from the data, the very principles of physics are baked into the neural network: left-right symmetry of the fetal head is embedded in the model architecture, making it easier for the model to accurately identify the brain’s orientation.&nbsp;</p><p dir="ltr">When Muthukrishnan began working on the project at the start of her PhD, she had no background in physics, let alone MRI physics. Most of her experience was in computer science. “I was surprised when I had to go and implement this on the scanner,” she recalls. “I was not prepared for how many times I would crash the scanner because putting things on hardware and getting it to work on hardware is a different beast than training a model to run on data on a GPU.”&nbsp;</p><p dir="ltr">Despite the challenges, Muthukrishnan persisted. She remembers shadowing a radiologist who performed fetal MRI scans at Boston Children’s Hospital at the start of her PhD and observing the shared frustration of the patient and radiologist: the pregnant mother felt uncomfortable after spending an hour in an MRI, only to quit the scan halfway; the radiologist was unable to procure a full, high-quality MRI to monitor the development of the fetus.</p><p dir="ltr">“Seeing frustration about it helped me decide what I wanted to do with my PhD, because it felt like this was a pain point,” Muthukrishnan says.</p><p dir="ltr">Fortunately, she had help: Co-author Borjan Gagoski, an assistant professor of radiology at Harvard Medical School and a researcher at Boston Children's Hospital, developed a sequence that acquires additional MRI scans to assist with training E(3)-Pose and worked with Muthukrishnan to help troubleshoot the scanner whenever it crashed. Co-author and Boston Children’s Research Assistant Aryn Lee recruited the pregnant volunteers and assisted with acquiring each of the MRI scans. Senior authors who helped guide the project include Elfar Adalsteinsson, MIT professor of electrical engineering and computer science; Ellen Grant, the Käthe Beutler&nbsp;Professor of Pediatrics and professor of radiology at Harvard Medical School; and Golland. Benjamin Billot, a researcher at Inria and a former postdoc in Golland’s group, helped Muthukrishnan get started with the initial method he developed.</p><p dir="ltr">“Ramya’s work elegantly combines geometric concepts like equivariance and symmetry with the power of deep learning to change how MRI scanners acquire the images, and to ultimately help the physicians detect and treat problems early in development. It’s an impressive research project that is deep and fundamental in how it models image acquisition, with clear clinical impact,” says Golland.</p><p dir="ltr">Grant, who directs the Fetal Neonatal Neuroimaging and Developmental Science Center (FNNDSC) at Boston Children’s, adds that Muthukrishnan’s work “will automate fetal brain MRI, replacing today's slow, manual process with faster and more comfortable scans for mothers, while also providing higher-quality images for radiologists to interpret.”</p><p dir="ltr">E(3)-Pose’s applications could extend beyond motion tracking for fetal MRIs. It could assist with imaging other hard-to-image anatomy, and even autonomous driving applications, where limited sensors need to capture noisy, 3D data in order to track 3D objects.</p><p dir="ltr">For now, Muthukrishnan hopes to improve E(3)-Pose so that it can help with imaging second-trimester or younger fetuses, as the model is p

  • But if the head shifts during imaging, each 2D image “slice” could end up being taken from a different angle, resulting in gaps in the final scan and unusual slice orientation.
  • </p><p dir="ltr">For now, Muthukrishnan hopes to improve E(3)-Pose so that it can help with imaging second-trimester or younger fetuses, as the model is p
MIT News: AI · Alex Ouyang | Abdul Latif Jameel Clinic for Machine Learning in Health Mở bản gốc ↗
Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

langgraph==1.2.22

<p>Changes since 1.2.21</p> <ul> <li>release(langgraph): 1.2.22 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382976091" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9702" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9702/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9702">#9702</a>)</li> <li>release(checkpoint-postgres): 3.2.1 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382974030" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9700" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9700/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9700">#9700</a>)</li> <li>release(prebuilt): 1.1.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382974997" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9701" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9701/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9701">#9701</a>)</li> <li>release(checkpoint): 4.3.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382972335" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9699" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9699/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9699">#9699</a>)</li> <li>fix(langgraph): stream a child graph on its own thread to a caller streaming subgraphs (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382639353" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9698" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9698/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9698">#9698</a>)</li> <li>perf(langgraph): skip message conversion in add_messages for plain messages (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6264178998" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9631" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9631/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9631">#9631</a>)</li> <li>perf(langgraph): filter pending writes to the given tasks before reading their statuses (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6265752562" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9634" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9634/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9634">#9634</a>)</li> <li>perf(langgraph): match update keys against a set on wide states (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6264179537" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9632" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9632/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9632">#9632</a>)</li> <li>perf(langgraph): check only failed futures for the runner stop condition (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6264178461" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9630" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9630/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9630">#9630</a>)</li> <li>perf(langgraph): find untracked and delta channels once per run (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6264177950" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9629" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9629/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9629">#9629</a>)</li> </ul>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

langgraph-checkpoint-postgres==3.2.1

<p>Changes since checkpointpostgres==3.2.0</p> <ul> <li>release(checkpoint-postgres): 3.2.1 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382974030" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9700" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9700/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9700">#9700</a>)</li> <li>release(checkpoint): 4.3.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382972335" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9699" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9699/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9699">#9699</a>)</li> <li>perf(checkpoint-postgres): run a lone store read outside a pipeline (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6265753208" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9635" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9635/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9635">#9635</a>)</li> <li>perf(checkpoint-postgres): skip the async saver's lock for connection pools (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6265921256" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9638" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9638/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9638">#9638</a>)</li> <li>perf(checkpoint-postgres): load no pending writes without a thread hop (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6265921507" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9639" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9639/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9639">#9639</a>)</li> <li>release(checkpoint): 4.3.1 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6046146171" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9447" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9447/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9447">#9447</a>)</li> <li>chore(deps): bump certifi from 2026.1.4 to 2026.10.19 in /libs/checkpoint-postgres (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="5952307672" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9369" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9369/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9369">#9369</a>)</li> </ul>

Nguyên văn nguồn
Chi tiết nguồn
Kỹ thuật

langgraph-prebuilt==1.1.2

<p>Changes since prebuilt==1.1.1</p> <ul> <li>release(langgraph): 1.2.22 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382976091" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9702" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9702/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9702">#9702</a>)</li> <li>release(checkpoint-postgres): 3.2.1 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382974030" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9700" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9700/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9700">#9700</a>)</li> <li>release(prebuilt): 1.1.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382974997" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9701" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9701/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9701">#9701</a>)</li> <li>release(checkpoint): 4.3.2 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6382972335" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9699" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9699/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9699">#9699</a>)</li> <li>perf(prebuilt): only format the call_model error message when raising (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6264175054" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9628" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9628/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9628">#9628</a>)</li> <li>release(langgraph): 1.2.21 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6353233846" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9683" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9683/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9683">#9683</a>)</li> <li>chore(deps): bump the minor-and-patch group across 1 directory with 9 updates (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6143893583" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9530" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9530/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9530">#9530</a>)</li> <li>chore(deps-dev): bump the minor-and-patch group across 1 directory with 6 updates (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6143890284" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9527" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9527/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9527">#9527</a>)</li> <li>release(langgraph): 1.2.20 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6215089962" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9595" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9595/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9595">#9595</a>)</li> <li>release(langgraph): 1.2.19 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6126910037" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9504" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9504/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9504">#9504</a>)</li> <li>release(checkpoint-sqlite): 3.1.3 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6127079894" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9505" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9505/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9505">#9505</a>)</li> <li>release(checkpoint): 4.3.1 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6046146171" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9447" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9447/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9447">#9447</a>)</li> <li>release(langgraph): 1.2.18 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6046020880" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9445" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9445/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9445">#9445</a>)</li> <li>release(langgraph): 1.2.17 (<a class="issue-link js-issue-link" data-error-text="Failed to load title" data-id="6007715725" data-permission-text="Title is private" data-url="https://github.com/langchain-ai/langgraph/issues/9418" data-hovercard-type="pull_request" data-hovercard-url="/langchain-ai/langgraph/pull/9418/hovercard" href="https://github.com/langchain-ai/langgraph/pull/9418">#9418</a>)</li> <li>chore(deps): remove precautionary third-party upper bounds (<a class="issue-lin

Nguyên văn nguồn
Chi tiết nguồn
321–340 / 738
Trang 17 / 37