46 skills · sorted by relevance
Analyze large scientific tables with Vaex, including datasets that exceed available memory. Build calculated columns, filter rows, compute statistics, visualize aggregated data, and convert files while identifying operations that may still require substantial RAM.
VAE/TAE 解码段优化（**计算 + 通信同技能**）：判断解码器能否沿某个轴切到多张卡、怎么切才是逐位精确、切了值不值；给出可复用的等价性证明流程、状态携带切分代码模式与交换预算判据。范围**含 VAE encode（待补：当前无内容，暂不纳入正文）**。当用户问“VAE/TAE 解码能不能分卡/切分/分摊”“8 张卡重复解码能不能省”“按帧轴还是空间轴切”“切完结果会不会变”“怎么证明等价”“all-gather 拼回来对不对”，或任何“多卡把解码摊开、又要保证输出不变”的场合都应触发；即使用户没说“分片”二字，只要出现“解码每卡都解一遍”“解码太慢想摊到多卡”“那部分重复计算能不能省”也应触发。覆盖：跨轴耦合算子审计（含递归 MemBlock 前缀依赖的否证）、潜帧边界切分＋每块末帧状态携带、与 rank 无关的判定（否则死锁）、交换预算核算、已知静默陷阱（all_gather_into_tensor 填栈副本、repeat_interleave 多分配整张量、rank 选举式参与）、有损分片的精确性前置条件与自动修复、CPU 真值→分段 bitwise→端到端 md5→帧级健康指标的四道验证闸、分片“到底有没有生效”的状态量判定，以及“切分顺带绕过设备缺陷”的用法。边界（本技能只引用不复制）：计算手段与能力供给见 `pattern-dev` / `operator-dev`，DiT 侧选档参照 `dit-perf-opt`；通信方法论（形态/掩盖/通信账/归因）见 `dit-parallel-opt`；一致性判据见 `accuracy-gate`；数字入库见 `perf-gate`；交付搬运与装载预热等 host 固定开销见 `host-opt`（不属本技能）。
The English summary is being prepared. Showing the original description.
优化域入口（L2）：拿到**已明确的瓶颈点 / 瓶颈标签**后，按标签把任务**分发**到四个优化模块 （DiT 计算 `dit-perf-opt` / DiT 通信 `dit-parallel-opt` / VAE `vae-opt` / host `host-opt`），并给出域内**最小前置集**与**域级验收口径**（性能入库口径引 `perf-gate`、精度判据引 `accuracy-gate`）。 **本技能不再承载选档与实施内容**（Step 2–4 闭环、特性档位选择、组合试验已全部归 `dit-perf-opt`）。 入口信号（任一即可触发）：① 用户或编排层已给出**瓶颈点/瓶颈标签**（"瓶颈已经明确，按这个点优化"）； ② **框架侧特性没落地**（该开的开关/特性没开，需要先判走哪个模块）。 near-miss（看似相关但不属本技能）： - **瓶颈未明**（"这个模型怎么加速""怎么跑通""采个 profile"，还要先定位）→ 先走 `model-auto-optimization`（唯一有分析权的一方），本入口**不接受未定位的任务**。 - 问"要不要开量化、开哪一档""量化/稀疏/Cache 怎么选怎么开" → `dit-perf-opt`（选档与实施在模块层）。 - 多卡并行形态 / 通信掩盖 / TP·offload → `dit-parallel-opt`；VAE·TAE 解码段 → `vae-opt`；交付搬运 / 装载预热等固定开销 → `host-opt`。 - 需要改本仓代码（pattern / 算子 / 测试 / 文档）→ `dev-workflow`。 由 model-auto-optimization 的阶段路由与用户直接声明瓶颈两条路径触发。
The English summary is being prepared. Showing the original description.
Used for finetuning the NV-Generate-CTMR MAISI VAE from CT/MRI NIfTI datalists. Not for clinical or production data approval.
The English summary is being prepared. Showing the original description.
Apply vaex in reproducible local data workflows with version-aware APIs, explicit assumptions, and validation. Use when the user chooses vaex or its strengths fit the task.
The English summary is being prepared. Showing the original description.
Create a coordinated set of video covers in 3:4, 16:9, and 4:3 formats from AI or Java scripts. Choose illustrations featuring a consistent presenter character or text-only layouts with white titles and yellow keywords.
Use this skill for processing and analyzing large tabular datasets (billions of rows) that exceed available RAM. Vaex excels at out-of-core DataFrame operations, lazy evaluation, fast aggregations, efficient visualization of big data, and machine learning on large datasets. Apply when users need to work with large CSV/HDF5/Arrow/Parquet files, perform fast statistics on massive datasets, create visualizations of big data, or build ML pipelines that do not fit in memory.
The English summary is being prepared. Showing the original description.
Implement filtering, aggregation, joins, and file processing with Polars for tabular data that fits in memory. Use lazy evaluation, which optimizes operations before running them, and parallel execution to improve pipelines and migrate pandas code.
DiT 计算模块（L3）：把**已定位的 DiT 计算瓶颈**落成特性级选档与实施—— 量化档（W8A16 / W4A16 / W8A8 系列 / W4A4 / MXFP8 / FA 量化）、稀疏（rf_v2 / ada_bsa）、 缓存（DiTCache / AttentionCache / 时间步优化）、编译启用（MindieSDBackend / Pattern 融合 / ACLGraph） 的**开不开、开哪一档、怎么开、怎么复验**；依据是 `docs/zh/features/*`（特性真源）+ framework-integration/references/framework-support-matrix.md（支持状态）。 即使用户只说"这个模型怎么加速""量化/稀疏/Cache 怎么选怎么开""要不要开量化、开哪一档""这个档位开了有没有效果" 而未提 profiling，也应触发。 **入口条件**：瓶颈点已明确（用户带一句实测锚点，或编排层交付标签）时由域入口 `performance-optimization` 按标签分发到本技能；**瓶颈未明（"怎么加速 / 跑通 / 采 profile"）先走 `model-auto-optimization` 定位**，不在本技能内做占比分析。 near-miss：多卡并行形态 / 通信掩盖 / TP·offload 选型 → `dit-parallel-opt`；VAE 解码段与 host 固定开销 → 各自模块（VAE / host）；单算子实现级实测选型（mindie_bench）→ `benchmark-dev`； 需要新增 pattern / 算子才能落地本档 → `pattern-dev` / `operator-dev`；框架侧开关与使能验证 → `framework-integration`；量化器位级契约与精度对齐（编码公式 / 舍入 / scale 粒度）→ `quantization-dev`；精度验收判据 → `accuracy-gate`；数字入库口径 → `perf-gate`。
The English summary is being prepared. Showing the original description.
host / 辅助段优化模块（L3）：把**非 DiT、非解码段**的固定开销落成可复验的优化—— ① 交付与搬运（mp4/图片编码、worker→API 通路、落盘、异步化）； ② 装载与预热（权重加载、编译与图下发预热、镜像/容器预热）。 当用户说“mp4 编码/传输/落盘慢”“交付这段占了不少时间”“模型加载慢”“首启编译太久” “固定开销占比高”“每个请求都要重新编码/重新写盘”时使用。 **入口条件**：占比已达门限（瓶颈标签 `非DiT-host段`；标签枚举、门限与测点口径的单一真源在 model-auto-optimization/references/bottleneck-labels.md，本技能只消费不定义）。 near-miss（看似相关但不属本技能）： - **并发 / 吞吐 / 批处理调度 / 队列深度与限流** → **不属本技能**（会改变所有性能口径与 SLO 语义， 未来独立立项；本技能只在**不改变请求并发模型**的前提下压固定开销）； - VAE/TAE 解码段（计算或通信）→ `vae-opt`；DiT 计算 → `dit-perf-opt`； 多卡并行 / 通信掩盖 → `dit-parallel-opt`； - **装库 / 下载权重 / 构建镜像 / 创建容器** → `env-install`（本技能只优化装载与预热**过程本身**）； - 编译与图下发的**机制本体**（pattern / Inductor / NPUGraph / 图 pool / 缓存条目）→ `pattern-dev` / `aclgraph-dev`（本技能不另做一套缓存）； - 产物一致性与精度判据 → `accuracy-gate`；收益口径 → `perf-gate`； - 瓶颈未定位 / 没有占比锚点 → `model-auto-optimization`。
The English summary is being prepared. Showing the original description.
Use this skill for processing and analyzing large tabular datasets (billions of rows) that exceed available RAM. Vaex excels at out-of-core DataFrame operations, lazy evaluation, fast aggregations, efficient visualization of big data, and machine learning on large datasets. Apply when users need to work with large CSV/HDF5/Arrow/Parquet files, perform fast statistics on massive datasets, create visualizations of big data, or build ML pipelines that do not fit in memory.
The English summary is being prepared. Showing the original description.
Analyze single-cell data with scvi-tools, using deep learning to combine datasets, correct differences between experiments, and transfer cell labels. Model guides and reusable scripts cover RNA, protein, chromatin accessibility, spatial data, and downstream analysis.
Create Chinese explainer videos targeting one minute, from a topic or existing script. The workflow produces narration, storyboards, Remotion animation, subtitles, and MP4 output, with staged reviews and complete scripts preserved even when videos run longer.
Write or refine Chinese spoken video scripts about AI technology, interview questions, and current topics. New scripts follow research and outline approval, while revisions provide specific wording changes and reasons for the edits.
Automate browser tasks such as reading JavaScript-rendered pages, filling forms, clicking through workflows, and taking screenshots. Maintain login sessions, manage separate browsers, and capture network responses, with approval checks for sensitive actions.
Search Zhihu, a Chinese knowledge-sharing platform, for answers, articles, people, and topics. Filter by content type and creation date, sort results, and collect structured records containing excerpts, authors, votes, and links across multiple pages.
Turn narration scripts into animated explainer videos using Remotion and 火山 TTS, a text-to-speech service. Prepare scripts, synchronize visuals with narration, preview each chapter, correct pronunciations, and export a verified MP4 with voice audio.
Manage 知识星球 login, inspect configuration, and call APIs, the service’s programmatic interfaces, through zsxq-cli. It also guides authentication troubleshooting, access errors, and creating desktop and mobile links for communities or posts.
Search and manage content in 知识星球 communities, including posts, comments, and questions. Create or edit posts, reply to discussions, answer questions, manage tags and featured status, and review question records through zsxq-cli.
Browse your 知识星球 communities and their latest posts, inspect tags and post counts, and find community IDs by name. Use keyword searches to locate communities or specific members, and browse posts under a selected tag.
Find unanswered questions in a designated 知识星球 community and draft replies in 沉默王二’s established style. Review past answers, verify project details against source code, and publish comments or formal answers after user approval.
Write AI tool reviews, setup tutorials, interview preparation articles, and technical explanations, or update existing articles against current source code. The workflow covers research, outline approval, prose checks, and source documentation.
Manage public notes on 知识星球: create text notes with image attachments, browse your notes, inspect details, edit content, and delete entries. Notes are independent of communities and accessible to anyone with their link.
Match videos from 王二讲Agent to their scripts, then add Bilibili links and Douyin video IDs to the designated documentation files. Identify missing entries while preserving existing links and IDs.