Mentions

@VirtualRecordsO Stemming’s a neat route too. I’ve experimented with it: Demucs splits bass/drums/other/vocals, each stem drives its own shader orb, then they get composited. I've done something like this a couple of times: youtube.com hughesdo/StemShader_Pipeline

There has been no guide to moving Power Platform solutions between tenants with code. I documented one with Claude AI. Third in the PAD ROBIN kb series: export, import, connection references and two-way sync, 100% free on GitHub. Two posts ago I shared a JSON knowledge base of PAD ROBIN actions, so models stop writing desktop flows that PAD rejects. Last week I followed up with a guide to getting one flow into an environment and running it from code: The piece that was still missing is moving a whole solution: to another environment, into a different organization's tenant, and keeping two copies in sync afterwards. That's what I'd spent the most time researching. So I wrote it down: PAD_Robin_Solutions_KB.md, in the same repo. It covers: - Signing in to two tenants, and what the receiving tenant needs before its first import (a license, the Dataverse service principal, an admin role and consent for your app), with the error you get for each one that's missing. - Export and import with the Dataverse Web API: ImportSolutionAsync, watching the job, reading per-component warnings from importjobs, and the fallback for an org that refuses async imports. - Connection references and turning flows on. The import never creates connections, but it can bind references to connections that already exist. The guide covers finding every reference (some only show up in the flows' JSON), binding them during the import, binding the rest afterwards, then turning on the flows that were on in the source. - Desktop flows inside solutions, including into another tenant. - Keeping two copies in sync: - why hashing the exported ZIP always reports a change, and a content hash that doesn't - how to stop the ping-pong after an import - two rules for deciding the direction ("newest wins" picked the right side 13 of 13 times) - what to do with conflicts - Where the code runs: the trap each of PowerShell, a server and a Chrome extension hit. For example, Chrome stops an idle service worker after about 30 seconds, and a synchronous import can take over 3 minutes. - A troubleshooting table of 27 errors, with what actually causes each one. Every claim is marked Re-confirmed, Proven, Measured, Superseded, Failed or Untested. The main paths were re-tested across two tenants this month. Everything ran with a user sign-in and unmanaged solutions; managed solutions got one export and import, and nothing was tested with a service principal. Unlike the direct Dataverse writes in the deployment guide, this all uses Microsoft's documented solution APIs. The risky part is the import itself: it replaces unmanaged customizations on the receiving side, and there's no undo. Keep a copy of whatever you're about to overwrite, and try it in a test environment first. Repo: JoeG592/pad-robin-action-kb Same license as the other two, CC BY 4.0, so use it however you like, commercially included. Like the last two, I built this with Claude AI. The data is free regardless. If something in the guide doesn't hold on your tenant, an issue with the error text is the most useful thing you can send. If a flow stays off after an import, it's almost always a connection reference with no connection behind it on the receiving side. Bind it, then turn the flow on.

49 agents collaborating on GitHub Trending, but when one node crashes, does the whole workflow restart? Session-level memory isn't enough — you need workflow-level checkpoint recovery. That's what astron-agent does: resume from the exact breakpoint, zero re-runs. iflytek/astron-ag

Just found out about archcanary, a layered security detection stack for Arch Linux. Definitely a useful tool with so many exploits in the wild. musqz/archcanary

should be paired with litecoin as that is what is was built as a fork from bitcointalk.org muddafudda/Memecoin t.co

Someone ported Super Mario Sunshine to the 3DS! I recorded some gameplay, check it out! Here’s the project link: github.com/ikevik/sms-3ds/releases t.co

YourOwnWeigh 是一套针对 Withings 体脂秤的开源方案，由固件补丁、本地 Python bridge（刷固件和蓝牙通信用）和自托管 Web 端组成，MIT 协议。 TomAnthony/YourOwnWeigh t.co

Blog by @muninn.austegard.com: muninn.austegard.com Experiment: github.com/oaustegard/experiments/tree/main/weigh…

サーバー監視基盤を作って運用した話 はじめに 私が所属する組織では13台のサーバーを運用しており、私はその管理を担当しています。 あるとき、組織の他のメンバーとの間で「普段どのサーバーを使っているか」が話題になりました。 話を聞くと、いつも同じサーバーの名前ばかりが挙がり、私はサーバーの利用に偏りがあるのではないかと感じました。 ただ、当時は組織で運用しているすべてのサーバーの使用状況を確認する手段がなく、それが本当かどうか確かめられませんでした。 そこで、各サーバーのリソース使用状況を確認できる監視基盤を構築しました。 本記事では、監視基盤の構成に加え、メンバーからの要望を取り入れた改善や、監視基盤を運用する中で見えてきた課題を紹介します。 前提 組織で運用しているサーバーの構成は以下の通りです。 種類 台数 用途 計算機サーバー 8台 機械学習 Webサーバー 1台 WebサイトやGitLab等のWebサービス その他のサーバー 4台 LDAP、踏み台など 監視基盤はWebサーバー上に構築しており、Webサーバー自身を含む13台のリソース使用状況を確認できるようにしています。 監視基盤の構成 監視基盤の全体像は以下の通りです。 なお、これは現在の構成で、DCGM Exporterは運用開始後にメンバーからの要望を受けて追加したものです。 各コンポーネントの役割 Node Exporter 各サーバーのCPU、メモリ、ディスクなどのリソース使用状況を、Prometheusが収集できる形式のデータとして公開するために使用しています。 監視基盤を動かしているWebサーバー自身にも設置しています。 prometheus/node_exporter DCGM Exporter 計算機サーバーのGPU使用率やGPUメモリ使用量、温度、消費電力などを、Prometheusが収集できる形式のデータとして公開するために使用しています。 docs.nvidia.com Prometheus 各Exporterからデータを定期的に収集し、時系列データとして保存しています。 また、アラート条件を評価し、条件を満たすとAlertmanagerへアラートを送信します。 prometheus.io Grafana Prometheusからデータを取得し、各サーバーのリソース使用状況を可視化するために使用しています。 grafana.com Alertmanager Prometheusから受け取ったアラートを、重要度に応じて通知先へ振り分けています。 prometheus.io Discord Webhook AlertmanagerからDiscordのチャンネルへアラート通知を送るために使用しています。 docs.discord.com その他 Docker Node Exporter、DCGM Exporter、Prometheus、Grafana、Alertmanagerは、すべてDockerコンテナとして動かしています。 Webサーバーでは、もともと組織のWebサイトやGitLabなどをDockerコンテナで運用していたため、監視基盤もそれにそろえました。 Ansible 組織では、もともとDockerのインストールやファイアウォールの設定などにAnsibleを使っていました。 Exporterの設置・設定もAnsibleを使えば13台にまとめて行えると考え、Ansibleを採用しました。 docs.ansible.com 監視基盤でできること リソース使用状況の可視化 以下のように、各サーバーのリソース使用状況をGrafanaのダッシュボードで確認できるようにしました。 アラート通知 ダッシュボードは自分から見に行かないと異常に気づけず、普段は他の作業もあるため常に確認し続けることはできません。 そこで、リソース使用率がしきい値を超えたときや、Exporterと疎通できなくなったときに、Discordへアラートを通知するようにしました。 設定したアラートは以下の通りです。 なお、これらは後述する見直しを反映した現在の設定です。 項目 下位アラート 上位アラート Node Exporterとの疎通 - 疎通できない DCGM Exporterとの疎通 - 疎通できない CPU使用率 80% 90% メモリ使用率 80% 90% Swap使用率 60% 90% ディスク使用率 80% 90% アラートは重要度に応じて、以下の2つのDiscordチャンネルに通知しています。 上位アラート（critical）：サーバー管理者のチャンネル 下位アラート（warning）：times Discordには以下のように通知されます。 ! 本記事ではドメインをダミーのものに差し替えています。 工夫した点 メンバーの要望を受けたGPU監視機能の追加 最初はNode Exporterを使って各サーバーのCPU、メモリ、ディスクなどのリソース使用状況のみを監視していました。 私はメンバーにリソース使用状況を監視できるようになったことを報告し、追加で確認したい情報を聞いたところ、GPUの状態も見られるようにしたいという要望がありました。 そこで、GPUを監視するためにDCGM Exporterを導入しました。 ダッシュボードについては、NVIDIAのGPU状態確認コマンドであるnvidia-smiで確認できる情報を網羅した既存のものが見つからなかったため、自作しました。 自作したダッシュボードをメンバーに使ってもらうと、「全サーバーのGPUの使用状況を一度に確認したい」という要望がありました。 その要望を受けて、全サーバーのGPU使用状況を一覧で確認できるよう、追加でダッシュボードを作成しました。 詳しくは以下の記事にまとめています。 zenn.dev AnsibleによるExporterの設置・設定の自動化 13台すべてにExporterを手作業で設置・設定するのは手間がかかり、設定ミスも起こりえます。 そこで、AnsibleでNode ExporterとDCGM Exporterの設置・設定を自動化しました。 公開用のコードは以下のリポジトリにまとめています。 他の方にも使ってもらいやすくするため、公開用のコードにはPrometheus、Alertmanager、Grafanaなどを構築するロールも追加しています。 HayateKato/Server-Monitoring-with-Ansible 運用で直面した問題と改善 Dockerコンテナからのサーバー情報の取得 ExporterをDockerコンテナで動かす際には、コンテナを動かしているサーバー自身の情報を取得するうえで以下の課題がありました。 Node Exporterが出力するホスト名（nodename）がコンテナIDになる ディスク情報が取れない これらの課題への対応策は、以下の記事にまとめています。 zenn.dev アラートの通知量 監視基盤の運用を始めると、アラート通知が頻繁に届き、ノイズになってしまいました。 振り返ると、アラートを設定する際に、通知の重複や対応の必要性を十分に考慮できていませんでした。 そこで、不要な通知の量を減らすために次の改善を行いました。 改善1. 重複する通知を抑制する 例えば、ディスク使用率では、同じサーバーで90%のアラートが発生している間、80%のアラートの通知を抑制するようにしました。 監視対象のサーバーを識別するラベルであるinstanceをequal: ['instance']で指定することで、通知を抑制する対象を同じサーバーのアラートに限っています。 inhibit_rules: ディスク使用率90%のアラート中は80%のアラートを抑制 - source_matchers: - alertname="HighDiskUsage-90%" - severity="critical" target_matchers: - alertname="HighDiskUsage-80%" - severity="warning" equal: ['instance'] アラートの抑制の設定方法は、以下の公式ドキュメントで確認できます。 prometheus.io 改善2. アラート条件の見直し アラートが通知されても、サーバー管理者が実際には対応しないケースがありました。 例えば、当初はSwap使用率が30%を超えた時点でアラートを通知していました。 しかし、Swap使用率30%程度ではサーバーの動作に影響がなかったため、サーバー管理者は通知を受けても対応していませんでした。 そのため、この条件のアラートは削除し、Swap使用率が60%、90%をそれぞれ超えたときのみアラートを通知するようにしました。 役立った場面 利用が集中しているサーバーの特定 ある計算機サーバーで、ディスク使用率90%のアラートが通知されました。 そこで、ディスク使用量の多い上位3人に、不要なデータの削除か外付けHDDへの移動をお願いしました。 しかし後日、ディスク使用率は99%に達していました。 依頼したメンバーの中にはデータを減らしてくれた人もいましたが、別のメンバーの使用量が増えていました。 特定の1人だけでなく、複数のメンバーがそれぞれ多くのディスク容量を使っている状況でした。 また、このサーバーでは、他の計算機サーバーと比べてCPUやメモリのアラートが特に多く発生していました。 一方、ダッシュボードで他の計算機サーバーの使用状況と比較すると、新しく導入されたサーバーの中には、ほとんど使われていないものがありました。 この状況から、特定のメンバーの使いすぎではなく、メンバーが同じサーバーに集中していることが原因だと考えました。 そこで、改めてデータの削除をお願いするとともに、何人かに他のサーバーへ移ってもらうよう呼びかけました。 その結果、元のサーバーではディスク使用率に関するアラートが発生しなくなり、新しいサーバーも使われるようになりました。 監視基盤を構築したきっかけであるサーバーの利用の偏りを、実際の使用状況から確認し、対処につなげることができました。 サービスに支障が出る前の異常の発見 WebサーバーのSwap使用率60%のアラートが数日続けて通知されました。 ダッシュボードで確認すると、メモリ使用率が70%程度、Swap使用率が65%程度の状態で、3日ほど推移していました。 以前から、GitLabの処理によってメモリ使用率とSwap使用率が一時的に高くなることはありましたが、今回は長期間続いていたため、普段とは異なる状態だと判断しました。 原因を調査したところ、GitLabのプロセスが多くのメモリを使っていたため、GitLabの設定を見直してメモリ使用量を抑えました。 GitLab自体は動作していましたが、アラートとダッシュボードで普段とは異なる使用状況に気づけたことで、サービスに支障が出る前に対処できました。 学んだこと 監視基盤を運用してみて、アラートは「通知を受けた人がどう行動するか」まで考えて設計する必要があると学びました。 しきい値を超えたことを一律に通知するだけでは、対応しない通知が増えてしまいます。 対応の必要性に応じてアラート条件を絞り、重複する通知を抑制することが重要でした。 また、GPU監視の追加や、全サーバーのGPU使用状況を一覧で確認できるダッシュボードの作成は、メンバーの要望がきっかけでした。 自分だけで必要なものを決めず、実際に使ってもらいながら改善することの大切さを学びました。 今後の改善案 対応につながっていない下位アラートの見直し 計算機サーバーのCPU・メモリの下位アラートは、通知されても対応していないのが現状です。 大きなモデルを動かすとすぐにしきい値に達するため、通常の利用でも頻繁に通知されてしまいます。 対応が必要な状況かどうかを踏まえて、通知条件を見直していきたいです。 アラートに対応する担当者の明確化 アラートへの対応が自分に集中していることも課題です。 現在は、通知を受けたときに誰が対応するかが決まっていません。 改善案として、通知時にサーバー管理者の中からランダムに1人をメンションする仕組みを考えています。 誰が対応するのかを明確にし、他のサーバー管理者も対応できる運用にしていきたいです。 サーバーを移りやすくする仕組み サーバーの利用の偏りには、新しいサーバーへ移る手間も関係しています。 以前から使っているサーバーにデータや実験環境があると、移動の負担が大きくなります。 そこで、NASによるデータ共有やDockerによる実験環境の整備により、移動の手間を減らせないか考えています。 利用状況を可視化するだけでなく、空いているサーバーを使いやすくすることで、サーバーの利用の偏りを減らしていきたいです。 おわりに サーバーの利用に偏りがあるのではないかという疑問から、監視基盤を構築しました。 監視基盤でサーバーのリソース使用状況を可視化し、アラートを通知したことで、サーバーの利用の偏りへの対処や、サービスに支障が出る前の異常の発見につながりました。 一方で、通知を出すだけでは対応につながらないことも分かりました。 今後は、残っているアラート条件の見直しや担当者の明確化、サーバーを移りやすくする仕組みづくりに取り組み、監視で得た情報を日々の運用に生かせるよう改善していきたいです。

ReadMeが長くなったのでdocsに切り分けた READMEを短くし詳細はdocsに移す github.com/ryo-i/next-nest-started/commit/4b576bc…

@Miracle_NNnn I'm building sandhiveio/sandhive-cli to automatically find relevant conversations that need my startup's solution & prep replies for them. Infinite Improbability Drive? More like SandHive drive!

DSH 的创造模式其实不是新功能，这应该算是 Deepseek 开始注重产品宣发了。这里也顺带分享一下我的探索模式插件，日常使用Token 消耗更少，执行任务的效果，也是挺不错的。taxueseek/dsh-router-flash t.co

一图胜千言？那我把图、语音、文字都塞进一个视频里，是不是就天下无敌了 周末提前在网上买了一些装饰布置婚房用的东西，零零碎碎的特多。 一个个跟父母打视频去说那个用在哪里、怎么摆、怎么装太麻烦，讲完转头他们又容易忘。 一开始想的就是，购物记录截图然后自己录语音一条一条发。 然后想到太麻烦，要么 excel 要么视频的 方式， 所以 Copilot 启动 后面就是你懂的，用 ai 就容易过度设计。 经典为了一碟醋包了一个饺子。 所以还是想分享一个这个小工具！ 截好图，我对着录一段语音讲解，AI 自动转字幕，一键生成竖屏视频，相当于一份视频说明书，直接微信发给爸妈，慢慢看就行。 本地跑的 AI ，所有东西都存在本机，说白了就是临时刚需折腾出来的小玩具。 GitHub： Wxh16144/dianzi-tushuo

2/ Try: git clone beenuar/AiSOC && make up Or: curl -fsSL raw.githubusercontent.com | bash No Docker: pip install aisoc-sandbox && aisoc-sandbox demo (~30s offline walkthrough).

ComfyUIにMiniMax H3 Max用ノード6種 ・動画の延長、画角変更、シーン挿入、口パク、カメラ制御、再構成 ・処理はMiniMaxのAPI側で、手元のGPUは使わない ・10月31日に本体へ取り込み済み 生成した動画の手直しまでワークフロー内でつなげやすくなる ComfyUI#17424

@sasattibot github.com/mikuta0407/LiberNT/releases/tag/ime-te… どうぞー

You deleted the API key. It's still in git. #Shorts Deleting a leaked key from your code doesn't un-leak it: git keeps every commit. Rotate it first, then clean up. Uber's 2016 breach started with one key in some code. Full explainer: youtu.be 📖 Guide: spidercob.com 🧪 Find secrets in your git history: github.com/SpiderCob/cve-labs/tree/main/concepts/… 🤖 Made with AI assistance: AI-generated narration voice; script researched from the primary sources linked above. SpiderCob: free, open-source secret, PII and CVE scanning for developers → spidercob.com Security concepts explained simply, plus the week's critical CVEs. Subscribe for the next one.

dtjsetupの/etc/{firefox,chromium}/unveil.*をfixする箇所、動作OK github.com/ykaw/FuguIta/commit/3ff4e2d1

[Project] Widget-Activity.Switcher A sleek, modern pill-shaped workspace Activity Switcher widget for KDE Plasma 6. --- Key Features: - Pill-style: visual switcher for KDE Plasma Activities - One-click: switching between active activity spaces - Dynamic: active state highlighting - Compact: panel and desktop widget modes - Icon Themes: multiple built-in numeral styles including Bubbles (White and Black SVGs), Arabic (White and Black), and Roman (White and Black) - Activity Reordering: easily reorder activities up or down directly within settings - Custom Styling: per-activity colors, individual icon selection, adjustable unselected opacity, sizing ratios, and shapes (Pill, Rectangle, Rounded Rectangle, Circle) --- GitHub: PlasmaDrifter/Widget-Activity.Switcher PlasmaDrifter: PlasmaDrifter/PlasmaDrifter

@AIatMeta Pass bars are set before each run. Muse explains each result, and a check rejects any number that is not in the report. Muse never changes a verdict. Demo: lohuunar-lobot.static.hf.space Code: meschinidino/lobot

优化软件一年 200，拆开看无非关后台服务、清临时文件、调视觉效果——系统里本来就能改。 RyTuneX 把关闭遥测/推荐、精简预装、电源方案、启动项摆一页，单项勾选可还原。 rayenghanmi/rytunex

SNES Audio Ripper I made a SNES audio extractor that rips the MUSIC and SFX from SNES games. Just drop your SNES games (.sfc files) right into the program and wala, SNES MUSIC and SFX from that game. I thought some people would find this useful with all of the game developing that has been going on as of lately. github.com/Samsqwamch/SNES_Audio_Ripper/releases/…

🦀 MOTION GATE Devlog — Nov 3 Our first isolated Content Binding verifier compilation check passed. ✅ 24/24 preflight checks ✅ 30 pinned crates ✅ 0 compiler errors ✅ No OOM Verifier executed? No. E2E complete? Not yet. Progress ≠ Proof. github.com/49414d-oss/fidomota-public/blob/main/d… t.co

兄弟们，来硬货了。 我使用 Claude Fable 5.1 和 GPT-6 Astra 构建了一个用于交易的扫描算法，结果：3 天内 +$1,731.72。 追踪了 1,056 次启动。分析了 2,340 名持有者。检测到 19 个钱包集群。只有 4 个代币通过，每一个得分 90+。 我将这个 Repo 公开了。 GitHub： QuantumOrb/QuantOrb 它的工作原理如下： -RADAR 实时追踪新的 pump.fun -SHIELD 检查代币创建者。 10+ 次启动中毕业率低于 10%？标记。开发者在启动时购买 15%+？标记 -SCANNER 分析前 20 名持有者和前 20 名买家，追踪他们的历史、入场和代币分布 -CARTOGRAPHER 通过共享资金来源、转账和同一交易购买来链接钱包，以揭示隐藏的所有权 -GUARD 检查创建者历史、首槽积累和剩余开发者持有量 -ORB 将信号组合成 0–100 的风险分数，并估计如果关联持有者卖出时的潜在价格影响 10 月 26 日，13:05 UTC：DESSA 在 pumpfun 上启动。 14:14 UTC：市值记录为 $133K。 QuantumOrb RADAR 警报，随后进行 SHIELD 和 SCANNER 检查。分数：92。 它达到了 $20.6M，大约是那个早期快照的 154 倍。 -20 个钱包 → 18 个操作者 -2 个关联钱包 → 流通量的 5.8% -潜在卖出影响 → −9.4% 我实盘说一句：这套东西的厉害不在抓到 DESSA，在于 1,056 次里只放进来 4 个，通过率不到 0.4%。筛得狠，才敢下重手。 即将公开发布。

[Project] NextBoot NextBoot (nboot) is a lightweight graphical and command-line utility for Linux systems that enables one-click and shorthand rebooting into Windows or any other operating system installed on your computer. It utilizes the UEFI standard BootNext mechanism. Setting BootNext instructs your motherboard to boot into the selected operating system for the next boot only. Once that operating system restarts or shuts down, your computer automatically returns to your regular default boot order. --- Key Features: - One-Click Reboot: Reboot into Windows or other operating systems from a native Qt GUI or terminal command. - Top 100+ Distro Vector Logos: Comprehensive built-in library of 105 vector SVG logos covering top DistroWatch distributions, BSD systems, and specialized operating systems with zero bloat (< 900 KB). - Clean OS Filtering: Intelligently distinguishes real bootable operating systems from raw motherboard hardware stubs (NVMe slots, SATA ports, PXE LAN) and internal firmware diagnostics. - Hide & Restore Unwanted Entries: Right-click to hide any entries you don't want to see; restore them individually or all at once via the bottom action bar. - Desktop & Hotkey Shortcuts: Right-click any OS entry to create an executable Desktop shortcut, add an Application Menu launcher (for GNOME Dash / KDE KRunner search and taskbar pinning), or copy its instant reboot command for global keybindings. - Desktop Environment & Wayland Intelligence: Automatically adapts to GNOME, KDE Plasma, XFCE, Cinnamon, MATE, COSMIC, and Wayland/Hyprland compositors. - Automatic Reversion: Subsequent boots return to your default Linux environment without manual boot menu intervention. - Smart Dynamic Detection: Dynamically inspects UEFI NVRAM entries via efibootmgr. No hardcoded drive UUIDs or partition numbers. - Interactive CLI Reference: Built-in dialog (? button or F1 shortcut) dynamically displays terminal commands tailored to your active boot entries with environment-aware tips. - Universal Sudoers Support: Compatible with both %wheel (Fedora, Nobara, Arch) and %sudo (Ubuntu, Debian) administrative groups. - Detached Drive Friendly: If a drive is temporarily disconnected, NextBoot detects its absence gracefully. As soon as the drive is reconnected or an OS is reinstalled, NextBoot picks it up automatically. - Fast CLI Shorthands: Intuitive commands like nboot win, nboot fedora, nboot arch, and nboot bios. - UEFI Firmware Setup: Quick reboot directly into motherboard BIOS/UEFI setup. - Flexible Reboots: Select an OS and click the action button, or toggle one-click rebooting for instant switching. --- GitHub: PlasmaDrifter/NextBoot PlasmaDrifter: PlasmaDrifter/PlasmaDrifter

Open Source Modified faster software I have managed to make mods for some existing software, including for this community some PDF software. All software involved still have their own licenses, my mod systems are under the MIT Licenses (Open Source). I was able to speed them up by making them multi-core. Now for a long time this multi-core capability was missing from these programs for various reasons, and those reasons were sound... At that time. Now we have a system that does a special sorting routine, and then processes the data accordingly. Basically we find things that look similar and treat them per what they are, with a stamp showing where they belong. This is a layman's way to talk about it, but you can find more on my GitHub. For GhostScript the link is here, where I achieved a 5.1x speed increase, for Apache Tika it is here, where we got a 5.9x speed increase, and for our grand prize winner Tesseract OCR we got a 6.6x speed increase and the github is here. My Computer: CPU: AMD Ryzen 5 7500X3D, 6 cores / 12 threads, 4.0 GHz base, 96 MB L3 cache (3D V-Cache) RAM: 16 GB DDR5-4800, a single stick (1 of 4 slots used) GPU: NVIDIA GeForce RTX 50-series, plus AMD Radeon integrated graphics Storage: NVMe SSD OS: Windows 64-bit These results were measured on a 6-core, 12-thread desktop. Computers with more cores should see even bigger speed-ups. So, something that might have taken you an hour on Apache Tika, might take 11 minutes now. There is zero loss of the files. The reason we did so well on Tesseract is because they have a bad threading system and we basically yeeted it for our own. Well, kinda. I do not make any claims on their stuff, they keep that intact. We are a presort and post process system, we just take the data, sort it our way, hand it to them on a silver platter, wait for them to run the parts as we feed them, then we take and restore it to proper order on the other end. So your licenses still apply, we do not change that at all. The code I made simply runs extra versions as needed with the special design set to give them the cleanest run possible.

Github on link bitcointalk.org sending fees github.com/muddafudda

Your SIEM fires alerts. Someone still has to decide which ones matter. AiSOC: self-host a SOC console — agents triage with a replayable ledger, humans approve the response. 6.2k stars. beenuar/AiSOC #GitHub #OpenSource t.co

見るだけの3D解説から、触って確かめるThree.js世界へ。 joonsp/threejs-skillsは解剖、diorama、science visualizerなどinteractive explainer向け。Vite/TypeScriptのscaffold、scene template、Playwright screenshot QAも含みます。WebGL2対応browserが必要です。 joonsp/threejs-skills

minha cota do Claude Code acabava toda semana. em 30 dias, 71% do gasto veio de turnos acima de 250k de contexto. fiz o muri-saver: statusline, guarda-contexto, /compact e memória entre sessões. grátis, MIT. murilolol/muri-saver t.co

[Project] Emulator-Web-Catelog ROMcat is a fast, lightweight catalog and launcher for your retro and modern ROM collection that runs directly in your web browser. It can be run in two primary ways: --- Key Features: - Direct Native Emulator Launching: Games launch in your real desktop emulators (Flatpak, native binaries, AppImages, or RetroArch cores) on your system. No slow or inaccurate in-browser emulation. - Two Flexible Modes: Use the standalone desktop app with dedicated window management or run as an always-on background web service. - Play Time Statistics Dashboard: Dedicated statistics page (/stats) tracking total play time, session counts, distinct games played, and top consoles. Includes interactive console breakdown charts, play leaderboards, recent activity feeds, per-game reset controls, interactive detailed game statistics modal with 7-Day, 14-Day, 30-Day, and 1-Year timeframe controls, effortless magnetic hover tracking, and responsive category filtering. - Dynamic Sorting & Filtering: Sort your game catalog by Name (A-Z with favorites prioritized), Most Played (by cumulative play time), Most Played (by total play sessions), or Recently Played. - Background Session Tracking: Automatically monitors active emulator processes and systemd scopes in the background to log play sessions (with minimum time threshold) without manual logging. - System Tabs & Live Counters: Instant switching between systems (e.g. NES, SNES, N64, GameCube, Wii U, Switch, Favorites, All, Hidden). - Fast Search & Filtering: Real-time title search across thousands of ROMs with automatic title normalization (stripping tags like [!], (USA), (Rev 1), .nkit). - Automated Cover Art Scraping & Interactive Picker: One-click cover fetching from SteamGridDB with per-console targeting, overwrite controls, multi-pass fallback title matching, and authentic portrait box-art filtering (600x900). Includes an interactive modal on game cards to search, browse, and select between alternative box arts with live previews. - Manual Cover Art Override: Drop custom cover art directly into the web UI or filesystem for unmatched or homebrew titles. - Favorites & Visibility Management: Toggle game favorites with custom glowing highlights, or hide unwanted duplicates/updates from the main catalog. - Zero Heavy Databases: Library state is indexed dynamically from your real directory structure, with cached metadata and settings stored in lightweight JSON files. - Built-in Settings & Theme Customization: Click the gear icon in the header to customize the catalog title, select from 6 built-in theme presets, or configure custom colors with live previews. --- GitHub: PlasmaDrifter/Emulator-Web-Catelog PlasmaDrifter: PlasmaDrifter/PlasmaDrifter

GitHub Skill Forge 这个开源项目挺有意思，能把任意 GitHub 项目直接“炼”成 AI Skill。 遇到好用的开源项目，不用再让 AI 从头啃整个仓库。把 GitHub 链接丢进去，它就会自动扫代码、翻文档、提取核心逻辑，再打包成 AI 能读懂、能调用的 Skill。 GitHub：YuJunZhiXue/github-skill-forge t.co

HC Player 1.7 — Xbox controller support, Microsoft Store debut, a dedicated shader panel, and more. Xbox-compatible controller support: control playback, seek, adjust volume, change speed, browse your playlist, and navigate supported Settings and MediaInfo controls. A new dedicated panel for Anime4K and shaders: you can now better organize your imported shaders in a dedicated panel. Previously, they were all grouped together in the settings. You can now select any format or quality extracted by yt-dlp from YouTube videos directly from the context menu. Added Japanese and Russian translations. Starting with this version, HC Player is now available on the Microsoft Store. apps.microsoft.com One more thing... With HC Player’s debut on the Microsoft Store, I decided to release an alternative version: HC Player Stream+. apps.microsoft.com HC Player is growing, though it’s still a small, If you like the project and want to support it, this version will be available exclusively on the Microsoft Store. It includes HC Player Torrent Engine, an clean and optimized engine for streaming content from torrent files and magnet links directly in the player. That’s it, EVERYTHING else is identical. It doesn’t download anything to your PC; it only streams. If you find this feature useful, it’ll be available as an option. All GitHub releases, including the source code, will remain available. https://github.com/henzfdev/HC-Player

[FOR HIRE] Programming, IT, Game Development & Online Tasks | Makati | Hello! I'm a Computer Science student currently looking for extra gigs, commissions, or contractual work to help with my college expenses! I'm pretty flexible with the kind of work I'm willing to take on, especially anything computer-related, technical, academic, or generally online. If you have a task you need help with, feel free to reach out! I'll let you know if it's something I can handle. Price: Final pricing depends on the task, complexity, and workload. I'm open to negotiating, especially for smaller commissions. What I can do / Hire me for: - Programming & Coding - C++, Java, debugging, code fixes, basic software development, and other programming-related tasks. - IT & Computer Science - Technical assistance, troubleshooting, software setup, backend development, and other IT-related work. - Academic Commissions - Programming exercises, math and logic problems, research, documentation, PowerPoint presentations, diagrams, flowcharts, and document formatting. - Game Development - knowledgeable in Unity, Godot, and Game Maker. can do development assistance, basic game mechanics, and troubleshooting. - General Online Tasks - Data entry, file organization, online research, administrative tasks, document editing, and pretty much anything else I'm capable of doing! I'm open to one-time commissions, small tasks, short-term projects, part-time remote work, and contractual arrangements. Location: Metro Manila Work Setup: Preferably remote / online, but open to discussing other arrangements. GitHub: github.com/Shmubbier A little context: I'm currently a college student studying Computer Science, and I'm looking to take on extra work to help support my education and personal expenses. Any opportunity, big or small, would be greatly appreciated! Just send me a DM with what you need, your budget, and your deadline. Don't hesitate to ask even if it's something not listed here. If it's within my capabilities, I'd be happy to help! Thank you so much!

@t20000622yy 可以在skillhub skillhub.cn 提交skill, 如果遇到问题 github.com/Tencent/skillhub/issues 提交issue

@temilolubigtown github.com/osenijoshua

Links Importantes: Site: msboficial.org Fórum: forum.msboficial.org Canal do Whatsapp: whatsapp.com Discord: discord.gg Github: github.com/MSB-org

DeepSeek V4.1向けにKVスプライスとレジデント専用ルーティングを追加したStrataのリポジトリ（dsv41-kv-spliceブランチ） github.com/PhialsBasement/Strata/tree/dsv41-kv-sp… 推論エンジンStrataのメインリポジトリ Niko1221/Strata KVスプライスの詳細について議論されているアップストリームのIssue #4921 Strata#4921 レジデント専用ルーティングの詳細について議論されているアップストリームのIssue #4936 Strata#4936 家庭用DRAMの活用や今回の検証事例について議論が行われているRedditのスレッド reddit.com

When Coding Agents Watch GitHub: A Practical Safety Checklist Before They Open PRs A coding agent that waits for a prompt is easy to understand. You give it a task, watch the run, and review the result. An agent that watches issues and comments changes the arrangement. It may notice a request, decide whether it is allowed to act, create a separate work session, and return a pull request later. The deepseek-ai GitHub organization hosts a public dsh-superagent research prototype, but its README explicitly says it is not an official DeepSeek product. The README describes it as reference research code, not production-ready software. Its value here is as a design case: the project makes permission and review questions visible. This is not a setup guide for that prototype. I have not installed or tested it. The checklist below is a safer way to evaluate any system that turns GitHub activity into coding work. A new input is also a new trust boundary The prototype's flow is more than "issue in, code out." Its documentation describes sources that receive events, policies that decide who may request work, tickets that hold tasks, and executors that do the coding. Work can return to the conversation where it started. Its setup guide recommends a dedicated agent account and an initial observe mode for work inferred from activity, rather than immediately executing it. That separation matters. A public issue is not automatically an authorized instruction. A comment can include misleading requests, hostile text, or copied instructions from an untrusted source. If that text reaches a model connected to a shell and a write-capable account, it is part of the system's input whether or not the author intended it as a command. GitHub's own Actions security guidance makes a related point: workflow credentials should have the least privilege needed, and default token permissions should be read-only where possible. The same principle applies to an agent. Its model may be capable, but its permissions should assume that some inputs will be wrong or malicious. What the prototype gets right, and what it does not prove The project draws a useful line between a direct request from an allowed person and work inferred from ambient activity. In its documented setup, inferred work begins in observe mode. The operator can see what the system would have done before enabling that behavior. The guide also advises using a dedicated GitHub identity, keeping it separate from the operator's personal account, and requiring human review rather than enabling automatic merging as the default. Those are good starting choices, not a security certification. The same deployment documentation warns that its worker sessions run with broad operating-system access under the operator's user. The account wrappers select GitHub credentials, but they do not isolate the process. A worker may still run programs by path and read files in that user's home directory. The docs recommend a dedicated OS account for isolation. This is the important distinction: separate GitHub identity is not the same as a sandbox. A bot account limits how the agent appears and what it can do in the repository. It does not, by itself, limit what code running on the host can read or execute. The checklist: start with observation, not autonomy Before connecting any always-on coding agent to a real repository, answer these questions in order. 1. Which events can create work? Start with explicit mentions from a small set of trusted accounts. Avoid treating every new issue, every comment, or every conversation as permission to run code. If the tool supports inferred work, keep it in a mode that records candidate tasks without executing them until you have inspected its decisions. Write the rule down in plain language. For example: "Only a new comment that mentions the agent, from an approved account, can start a task." That is easier to test than a vague instruction such as "help with relevant issues." 2. Who is allowed to ask? Use an explicit allowlist. Do not assume that repository visibility, team membership, or an issue label is enough to establish intent. Review who can add accounts to the list and how quickly removals take effect. Try negative tests before real work: an unapproved account mentions the agent; a bot posts a comment; a user edits an old issue to add a mention. The expected result should be no execution and a clear record of why the request was rejected. 3. What can the agent's GitHub identity change? Create a separate account for the agent, then grant only the repository access required for its task. Keep merge rights with people. Make the agent's branches easy to identify, and ensure its comments and commits are visibly attributed to the agent rather than to a human maintainer. Review token scopes and repository permissions directly. If the agent only needs to read issues and create branches or pull requests, do not give it organization-wide administration or access to unrelated repositories. Recheck permissions when the tool or workflow changes. 4. What can the process read on the host? This is easy to miss when setup instructions focus on GitHub accounts. Find out which operating-system user runs the agent, which directories it can read, whether its shell can access environment variables, and where credentials are stored. Check whether its tools can reach internal network services. Use a dedicated OS account or a properly isolated execution environment when the agent handles untrusted issue text. Keep secrets out of the agent's checkout and home directory where possible. A wrapper that changes which GitHub account a command uses is helpful, but it is not a process sandbox. 5. Can text in the task change what the agent is allowed to do? Prompt instructions are not a permission boundary. Define limits outside the model too: allowed repositories, allowed commands, network access, branch rules, and whether secrets are available. Treat instructions quoted inside issues, pull requests, logs, and linked pages as untrusted data unless a trusted person explicitly confirms the task. 6. What has to pass before a pull request is ready? Require the same checks you expect from a human contribution: tests, formatting, dependency review, and a human reading the diff. The agent should explain what it changed, which tests it ran, and what it could not verify. A green check means the configured checks passed; it does not mean the request was safe or correct. Keep automatic merge disabled while you learn how the agent behaves. A pull request is a review boundary, not a guarantee. If the agent can merge its own changes, review whether that permission is genuinely necessary and what happens when tests are incomplete or the change touches sensitive code. 7. How will you stop and investigate it? Know how to pause event intake, stop active workers, revoke credentials, and inspect logs before enabling the service. Decide who receives alerts for stuck tasks, repeated retries, or unexpected repository writes. Test the stop path before the agent has access to valuable credentials. A low-risk rollout in three stages A small rollout makes failures easier to understand. - Stage one: observe. Connect a test repository or a narrow set of events. Let the system record what it would do. Compare its decisions with what maintainers actually want. - Stage two: prepare changes. Permit a dedicated account to create branches and draft pull requests in a low-risk repository. Keep a person responsible for every review and merge. - Stage three: expand one permission at a time. Add more requesters or repositories only after you have checked false triggers, quality, noisy behavior, and recovery. Do not change event access, host access, and merge rights together. If something goes wrong, you need to know which new permission caused it. Keep a short record of expected behavior for test cases: who requested work, what event started it, what the agent was allowed to do, which checks ran, and who approved the result. This creates evidence for a decision to widen access, rather than relying on a few impressive demonstrations. The useful idea is the workflow, not the word "autonomous" DeepSeek's prototype points to a real design problem: a team may want one agent to carry work across messages, code changes, and review comments without making people restate the task in every session. That can save coordination effort. It also means the agent becomes a long-running participant with access to event streams and execution tools. The project itself is careful to label its public implementation as a research reference, not production-ready software. Its documentation also exposes tradeoffs that a launch headline can hide, including broad host access and the need for a separate operating-system boundary. Those caveats are reasons to test it as an operations system, evaluating both its code and the safeguards around it. A useful rule is simple: automate the preparation of a change before automating permission to apply it. Start with narrow inputs, limited credentials, isolated execution, visible drafts, and human approval. Expand only when the logs show that the system follows those limits consistently. Before enabling an issue-to-PR agent, check: - Only approved people can create an executable request. - Inferred work is observed before it is allowed to run. - The agent uses a dedicated, narrowly scoped GitHub identity. - Its operating-system access is isolated from personal files and secrets. - Pull requests require human review and do not merge themselves. - You can pause the service, revoke access, and inspect its actions. I write about software engineering and AI systems. Subscribe if you want practical, source-based guides like this one. Have you let an agent act on GitHub issues or pull requests? What permission limit would you set first? Sources: DeepSeek dsh-superagent prototype, DeepSeek setup guide, DeepSeek deployment and permissions guide, DeepSeek project philosophy, GitHub Actions secure use reference.

I turned my Redmi Watch 5 Active into a remote for my PC using only the Music screen (no root, no custom firmware, open source) I bought a used Redmi Watch 5 Active mostly for the looks. First night, honestly, I found it kind of useless: steps, notifications, and the flashlight was my favorite feature. I wanted to do things from my wrist, but every path was a dead end: no third-party apps, signed firmware, and watch face tap zones only open the watch's own apps. So I flipped the question: instead of "what do I want it to do", what does it already send out? The answer was the Music screen. It shows the track title and has ⏮ ⏯ ⏭, and it talks to whatever media player is active on the phone. So I made an Android app that pretends to be a music player. It never plays music: the "track title" is any text I want, and the three buttons get forwarded to a small Python server on my PC. Result: the stock watch, with the stock Mi Fitness app, now has a one-line screen and three buttons wired to my computer. Some things I had to work around: - Android sends media buttons to whoever played audio last, so YouTube Music kept stealing them. The app plays half a second of silence on each press to stay "the last one playing". - The font has very few symbols. I sent batches of emoji and photographed the screen. These work: 🔴 ✅ ⭐ 🔥 💚 💛 ❤️ 😀 👍 ■ □ ▲ ▼ ← → ① ② ③. Colored squares, ❌, 🙂, 🌙 and most others show up as empty boxes. (Might be useful if you're making watch faces too.) - Long titles scroll. With shrinking test lines I measured what fits: about 16 digits, 9 emoji or 9 squares. Everything is cut to that. - No album art, no artist line on this model (tested), so long text goes out as a regular notification, which buzzes on the wrist. - The watch's volume buttons change the phone's real volume, so I stuck with three buttons. Double-taps don't register well either, so navigation is single presses only: ⏭ next, ⏯ act, ⏮ back to menu. What I like using it for: - Slide clicker. Load a PDF, pptx or Google Slides link; ⏭ next, ⏯ previous. The watch shows "slide 3/14". - Turning my home servers on and off. 💚 on / ❤️ off, one press, and it opens the page on my PC when it starts. There's also a mood check-in (one press: 😴😐🤔💪🔥), flashcards, a metronome, a dice roller for tabletop RPG, and a mode that lets me approve Claude Code permission requests from my wrist. Full disclosure: it's my own free project (MIT), nothing to sell. I built it in one evening with Claude Code, which wrote all of the code; my part was the idea, testing on the watch and deciding what worked on the wrist. I'd love to know: 1. Does this work on your watch? If your Music screen shows the track title (Redmi Watch 4/5, Smart Band 8/9, Amazfit…), it probably should. Tell me the model and what you see. 2. What would you do with three buttons and one line? Someone who uses Home Assistant could make a great mode for it (the watch has Alexa, but not everyone wants Alexa). OBS/streaming controls are another one I'd love to see, but I don't stream. Repo, with install instructions (you can also just give the link to your AI agent and it sets everything up and sends you a link to install the app): canpse/punho

Bonus: under round-to-nearest the fp32 checkpoint is not the model. On HumanEval+ one master scored 9.1 vs 59.8 for its served copy. Caveats: models ≤ 4B, several single-seed runs. Paper (Stanford University): arxiv.org Code: andrewgcodes/faircast

@odpanda TripoはCharakuruを支援しています リアル系のアバターで使えるかは試したことないですけど nanocle/Charakuru リアル系の場合、私はMakeHumanなど他のHuman規格との組み合わせでやれないか試行錯誤中です t.co