'parallel' 검색 결과

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, and filler phrases.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Implement 10 advanced AI agent node types for any DSL/runtime system — plan compiler, code interpreter, critique loops, intelligent routing, multi-agent ensemble, persistent memory, external API tools, iterative loops, data transforms, and parallel execution. Use this skill whenever the user wants to add agent capabilities to a workflow engine, build an AI orchestration framework, implement Karpathy-style AI upgrades, add new node types to a DSL runtime, or create autonomous agent pipelines. Also trigger when the user mentions AetherLang, flow engines, node types, agent frameworks, or wants to upgrade their AI system with capabilities like self-programming, reflection, routing, ensemble, memory, or tool use.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Enhanced terminal AI agent orchestrator with parallel execution, health checks, and workflow presets.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Patterns and practices that dramatically accelerate development velocity. Covers parallel execution, automation, feedback loops, workflow optimization, and anti-pattern avoidance. Use when starting projects, planning sprints, optimizing workflows, or onboarding developers.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

This skill should be used when the user asks to "implement a DiD regression", "run a staggered difference-in-differences", "set up an event study", "implement IV / 2SLS", "run a regression discontinuity design", "build synthetic control", "do propensity score matching", "test parallel trends", "run Honest DiD", "wild cluster bootstrap", "Callaway-Sant'Anna", "Sun-Abraham", "Bacon decomposition", "double machine learning", "debiased machine learning", "DML / DoubleML", "cross-fitting", "causal forest", "heterogeneous treatment effects", "ATE with ML controls", "which causal design fits my data", or needs causal-inference code, diagnostics, or identification writing in Python (StatsPAI), R, or Stata. Based on Scott Cunningham's Causal Inference: The Mixtape, with runnable validated templates, ML-based causal inference patterns, and Journal of Finance applications.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Use for Six Thinking Hats or de Bono's parallel thinking, or when a decision or idea needs facts, gut feel, risks, benefits and new options kept apart: "look at this from all sides", "full perspective check". Skip for plain factual questions or single-angle hot takes.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Identify independent failures and spawn concurrent agents. Turns sequential N×time into 1×time.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

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`。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

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）。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

profiling 数据统一采集：远端昇腾采集 NPU 性能数据回传本地，供 profiling-analyze 分析。 两条入口——mindiesd 自家脚本入口（collect_profile.py）与三方框架入口 （collect_patch_template.py 补丁 + torchrun，含 vLLM-Omni HTTP 服务等分场景）； warmup 在 profiler 外（默认 5 步，compile ≥10）。 只要用户想"开 profiler/采 profile/拿算子级数据"，自家脚本或 vllm、LightX2V 等框架都先 用本入口；分析/瓶颈定位/优化选型/并行/基准请求分别属 profiling-analyze、 dit-perf-opt、dit-parallel-opt、benchmark-dev，本入口不承接。 由 model-auto-optimization 的 S1（融合分析）阶段调用，亦由 profiling-analyze 与 dev-workflow 的 采集场景指引加载。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

**精度验收标准**：凡是"改动不应改变结果"的场合（等价替换、算子/子模块融合、并行切分、 编译与图下发、拷贝消减），都按本标准判"合格 / 不合格"——等价分层（L1 逐位 / L2 数值门 / L3 有损门）+ 三级验收序（① 同配置重跑逐位 → ② 跨配置数值门 + 产物 md5 不变 → ③ 质量门）+ 判据不达标时的排障入口。当用户说"这个算子能不能换个写法/换个 kernel/等价实现/ 无损替换""替换后结果会不会变""怎么证明逐位一致""结果不对""花屏""尾部塌了" "CPU 跑对 NPU 跑不对"，或发现某个 hotspot 占了大头（例如某类算子在阶段里占 90% 以上）想动手时， 都应触发；即使用户只说"这样改有没有把结果改坏""能不能判它是无损的"也应触发。 覆盖：等价分层与判据、满足本标准的实现约定（per-shape 对拍写进实现，把索引/相位/顺序类重写 错误在首次调用抓住）、per-shape 条件性等价（同一改动在不同形状下结论可能不同）、 判据不达标 → `references/silent-failure-localization.md` 排障入口、否决案例的形态学 （换写法未换 kernel / 差异极小仍非逐位 / 等价但 OOM）、以及收益口径的归属（性能数字一律交 `perf-gate`）。 **近义分流**：选量化档 / 特性选档（要不要开量化、开哪一档）→ `dit-perf-opt`； 并行选型（USP / CP / TP 怎么切）→ `dit-parallel-opt`；本技能只判"结果是否被改变"， 不负责选档与选型。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

量化格式契约与位级对齐：从设备字节反推量化器的**编码公式、舍入模式、scale 粒度、退化块规则**， 并据此重实现/融合量化器（MXFP8 e8m0、int8 perblock、fp8 perchannel 等）， 用「同进程同数据 + 独立参考实现 + 中点与退化输入全覆盖」做到**逐字节精确**。 当自研量化 kernel 与框架算子**对不上**、需要逆向某个量化器的数值契约、需要判断 「量化残差算不算可接受」、或要在融合内核里复现 aclnn 量化语义时使用此 skill。 即使用户只说「量化结果和参考不一致」「这个 scale 怎么算出来的」「融合量化器精度对不上」 而未提"契约"，也应触发。**注意**：单纯选量化档位/配置（该不该开量化、开哪一档）走 dit-perf-opt；算子级性能调优与 DSL 选型走 operator-dev； 并行作用域导致的静默失效（DiT 侧「改了并行但不报错也没生效」的判定）走 dit-parallel-opt。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

对 profiling 数据（kernel_details.csv / trace_view.json / step_trace_time.csv）做统一分析： 5 层递进管道（analyze_trace.py）+ baseline/优化后 kernel diff（compare_traces.py），输出 瓶颈诊断、算子执行序、可融合候选与 P0-P2 方向建议。输入统一来自 profiling-collect。 只要用户有 profiling 产出并问"为什么慢/瓶颈在哪/优化前后差多少/往哪个方向"，都用本技能；采集走 profiling-collect，方案选档走 dit-perf-opt，实现/并行/基准走 pattern-dev/dit-parallel-opt/benchmark-dev——本技能只做诊断与方向。 由 model-auto-optimization 的 S1（融合分析）/S3（并行）阶段调用，亦由 dev-workflow 的分析阶段指引加载。 融合机会候选只做**识别**；候选的**边界与收益判定**（能不能融成一个单元、融多大、值不值）交 `fusion-scope-analyze`（本技能的辅助技能）——本技能不判融合边界。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

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`。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

多模态模型自动优化流程总入口（编排层）：把一个三方框架托管的扩散模型从环境准备到无损/有损 优化自动跑完。任务判定后按 S0 环境准备、S1 kernel 融合优化（原 S1/S2 合并）、S3 并行通信、 S4 有损优化、S5 训练感知（预留，少步蒸馏 + SLA/QAT）、闭环复验路由到能力技能（env-install / framework-integration / profiling-collect / profiling-analyze / performance-optimization / dit-parallel-opt / dummy-run 等），检查各阶段产物与验收并回填经验槽位。 当用户需要对具体模型做框架接入、无损/有损加速、并行调优或确认收益时使用本入口；即使用户只提 模型名加"优化/加速/跑通/采profile"而未说框架或阶段，也应由本入口判定路由。仓库代码开发类任务 请走 dev-workflow。流程执行/验收按 workflows/optimization-flow.md，run-state + stage_gate 门禁。

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Breaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases. Credits: Original skill by @blader - https://github.com/blader/humanizer

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Structures git workflow practices. Use when making any code change. Use when committing, branching, resolving conflicts, or when you need to organize work across multiple parallel streams.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Generate multiple radically different interface designs for a module using parallel sub-agents. Use when user wants to design an API, explore interface options, compare module shapes, or mentions "design it twice".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Foundational theory for process-oriented concurrency through synchronous message-passing, applicable to multi-agent coordination and parallel decomposition

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Parses complex problems into DAG (Directed Acyclic Graph) execution structures. Decomposes tasks into nodes with dependencies, identifies parallelization opportunities, and creates optimal execution plans. Activate on 'build dag', 'create workflow graph', 'decompose task', 'execution graph', 'task graph'. NOT for simple linear tasks or when an existing DAG structure is provided.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Algorithms for decomposing DAGs into chains for parallel execution and resource optimization

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Manages context passing between DAG nodes and spawned agents. Handles context summarization, selective forwarding, and token budget optimization. Activate on 'bridge context', 'pass context', 'summarize context', 'context management', 'agent context'. NOT for execution (use dag-parallel-executor) or aggregation (use dag-result-aggregator).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.