「moe」の検索結果

Python framework for single- and multi-objective optimization with evolutionary algorithms. Define vectorized objectives and constraints; solve with NSGA-II, NSGA-III, MOEA/D, GAs, or differential evolution. Analyze Pareto fronts, visualize trade-offs, customize operators and callbacks. For engineering design, hyperparameter search, and conflicting objectives. Alternatives: scipy.optimize (single-objective, gradient), platypus, jMetalPy (Java).

日本語の概要は準備中です。原文の説明を表示しています。

Use DeepSpeed for distributed training, inference acceleration, ZeRO configuration, parallelism/MoE design, profiling, autotuning, and operational diagnostics.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Deep Learning Scientist. Use when a task calls for Deep Learning Scientist judgment. Reasons from CNN/Transformer inductive bias, Li et al. loss landscapes, grokking/mode connectivity, and Kaplan/Chinchilla scaling (~20 tokens/param); designs ResNet/ViT/DiT/MoE/FlashAttention stacks with FLOPs-matched ablations; trains AdamW+cosine/WSD via Megatron-FSDP/DeepSpeed; evaluates FID/MMLU-Pro/MMLU-CF with lm-eval decontamination and Pineau/NeurIPS reproducibility checklists.

日本語の概要は準備中です。原文の説明を表示しています。

Think and work like an expert Computer Hardware Engineer. Use when a task calls for Computer Hardware Engineer judgment. Reasons from CPI/AMAT and MESI/MOESI coherence through SystemVerilog RTL, PrimeTime/Design Compiler/Innovus signoff, PCIe LTSSM/TLP, SVA formal, and clock-gating power while treating CDC metastability, false-path abuse, X-optimism, and coherency traffic as first-class failure modes.

日本語の概要は準備中です。原文の説明を表示しています。

Expert in presenting technical work to non-technical audiences. Covers demo preparation, storytelling for demos, handling failures gracefully, and making work visible and impressive. Understands that perception is reality and great work undemoed is invisible work. Use when "demo, present, show stakeholders, demo day, showcase, walkthrough, show and tell, " mentioned.

日本語の概要は準備中です。原文の説明を表示しています。

投行并购 (投行并购 / 交易撮合 (Investment Banking M&A / Dealmaking) — 面向并购顾问、企业融资、卖方/买方交易执行的职业认知操作系统，从业者(分析师 Analyst / 经理 Associate / 副总裁 VP / 董事 Director / 董事总经理 MD / 内部公司发展 Corp Dev / PE 投资人)、想入行者、以及服务这行(法律/会计/估值/咨询)的视角。覆盖: (a) 第一性张力 — 投行 M&A 的核心世界观分歧: **价格 vs 价值 (price 是市场愿意付的, value 是内在的, deal 在两者缝隙里成交) ⇄ 确定性 vs 最高价 (certainty of close 对卖方常胜过纸面最高报价) ⇄ 顾问/中介 vs 委托代理 (banker 是 agent, 客户利益 vs 成交费的激励冲突真实存在, 「banker 总想成交」是行业原罪批判)**; 更深层 — 「过程/竞价纪律 (controlled auction 制造竞争张力、信息不对称管理) ⇄ 关系/信任 (relationship banking、长期 coverage、reputational capital)」, 「叙事/包装 (the story, equity story, positioning) ⇄ 数字/尽调 (the numbers, diligence, quality of earnings)」, 「估值的科学外衣 (DCF/comps 的伪精确) ⇄ 谈判的真实定价 (价格由 leverage、竞争、时机决定)」; (b) 估值与分析正典 (最标准化、最易蒸出高质量+CLI 化) — DCF (折现现金流: FCF 预测/WACC 折现率/terminal value 终值/Gordon growth vs exit multiple), Comparable Companies (可比公司/trading comps: EV/EBITDA, EV/Revenue, P/E), Precedent Transactions (先例交易/deal comps: control premium 控制权溢价), LBO (杠杆收购模型: IRR/MOIC/debt schedule/sources & uses), Accretion/Dilution (增厚/摊薄分析: EPS 影响、pro forma), Sum-of-the-Parts (分部估值/SOTP), Football Field (估值区间图), Merger Model (合并模型: synergies 协同/purchase accounting/goodwill), Sensitivity/scenario; (c) 交易流程与产出物 — 卖方流程 (sell-side: teaser 简介 → NDA → CIM/IM 信息备忘录 → management presentation → IOI 意向函 → data room 数据室 → LOI/term sheet → confirmatory diligence → SPA 买卖协议 → signing → closing), 买方流程 (buy-side: thesis/sourcing → outreach → valuation → bid → exclusivity → diligence → financing → close), auction 类型 (broad/targeted/negotiated/preemptive bid); 产出物: pitchbook 推介材料、CIM、fairness opinion 公平意见、board presentation、fee letter、engagement letter; (d) 交易结构与条款 — consideration (cash/stock/mix/earnout 对赌)、deal protections (breakup fee 分手费/go-shop/no-shop/MAC 重大不利变化条款)、representations & warranties 陈述与保证、escrow 托管、working capital adjustment 营运资金调整、reverse termination fee、collar 价格区间、tender offer vs merger、hostile vs friendly、defensive tactics (poison pill 毒丸/white knight 白衣骑士/staggered board); (e) 行业结构与角色 — bulge bracket (高盛/摩根士丹利/JPM/BofA/花旗) vs elite boutique (Centerview/Evercore/Moelis/Lazard/PJT/Perella) vs middle market (Houlihan Lokey/William Blair/Harris Williams) vs regional/independent; analyst→associate→VP→director→MD 晋升金字塔; coverage (行业组) vs product (M&A/ECM/DCM/leveraged finance/restructuring); sponsors group (服务 PE); 买方包括 PE/strategic acquirer/corp dev; (f) 量化指标与节奏 — deal fee (success fee 业内通常约 1-2% 视规模递减/retainer/Lehman formula)、league table 排行榜、pitch→mandate 转化、deal velocity、hours culture (analyst 80-100h/week 文化批判)、bonus/comp 结构、deal sheet (简历上的成交清单); (g) 争议/批判 — fairness opinion 是否橡皮图章 (利益冲突: 出意见的 banker 也收成交费)、estimate 的伪精确 (DCF garbage in garbage out、假设驱动结论)、banker 永远建议成交 (deal 偏见、empire building 助推 CEO 自利并购)、协同效应系统性高估 (大量并购毁灭价值、winner's curse 赢家诅咒)、analyst 血汗工厂文化 (2021 高盛 13 人 deck 事件、过劳死争议)、advisory 价值 vs 中介寻租之争、bulge vs boutique 的利益独立性叙事 (boutique 标榜「无承销利益冲突」)、SPAC 泡沫与 de-SPAC 教训; (h) 流派/思想谱系 — 估值学派 (Aswath Damodaran 估值教父/纽大、Tim Koller McKinsey《Valuation》、McKinsey valuation 体系) vs 公司金融教科书派 (Rosenbaum & Pearl《Investment Banking》行业圣经、Brealey Myers Allen、CFI/WSP/BIWS 培训体系) vs 交易实战派 (Bruce Wasserstein《Big Deal》、Joseph Perella/Felix Rohatyn 老派 dealmaker、现代 boutique 创始人) vs 行为/批判派 (并购毁灭价值的学术证据、Damodaran 对协同/控制权溢价的批判、Matt Levine《Money Stuff》解构华尔街逻辑) vs PE/buy-side 视角 (LBO 数学、value creation 杠杆/运营/multiple expansion)。不含: 二级市场股票交易/卖方研究 (equity research)、零售财富管理/private banking、纯 ECM/IPO 承销执行 (虽相关但聚焦 M&A advisory)、纯 PE 投后运营、量化交易/对冲基金策略、纯公司财务报表会计。) Master OS — automated mastery of 投行并购 / 交易撮合 (Investment Banking M&A / Dealmaking) — 面向并购顾问、企业融资、卖方/买方交易执行的职业认知操作系统，从业者(分析师 Analyst / 经理 Associate / 副总裁 VP / 董事 Director / 董事总经理 MD / 内部公司发展 Corp Dev / PE 投资人)、想入行者、以及服务这行(法律/会计/估值/咨询)的视角。覆盖: (a) 第一性张力 — 投行 M&A 的核心世界观分歧: **价格 vs 价值 (price 是市场愿意付的, value 是内在的, deal 在两者缝隙里成交) ⇄ 确定性 vs 最高价 (certainty of close 对卖方常胜过纸面最高报价) ⇄ 顾问/中介 vs 委托代理 (banker 是 agent, 客户利益 vs 成交费的激励冲突真实存在, 「banker 总想成交」是行业原罪批判)**; 更深层 — 「过程/竞价纪律 (controlled auction 制造竞争张力、信息不对称管理) ⇄ 关系/信任 (rela

日本語の概要は準備中です。原文の説明を表示しています。

Guides multi-GPU pre-training: DDP, FSDP2, ZeRO, tensor/pipeline/expert parallelism, fp8/Muon. Use when scaling a run, training MoE, or reproducing GPT-2 on rented GPUs.

日本語の概要は準備中です。原文の説明を表示しています。

Chooses among AI/ML approaches: classical ML, LLM, RAG, fine-tuning, agents, embeddings, MoE or SSM. Use when picking or scaling an architecture, or self-host vs API.

日本語の概要は準備中です。原文の説明を表示しています。

Use when verifying that a model in the imp inference engine produces coherent output without repetition loops, token-stuck states, or state corruption across turns or streams. Triggers on "degenerates", "check degeneration", "repetition loop", "own own own", "stuck token", "empty content", "multi-turn regression", "does it still work", "NIAH", and after enabling CUDA graphs / changing forward pass / MoE routing / KV cache or KV dtype / GDN state or scan / sparse attention / PDL / speculation (MTP, n-gram) / ragged prefill / batched-decode kernels (smallm, producer quantize) / FA2 softmax.

日本語の概要は準備中です。原文の説明を表示しています。

Run GGUF models directly, load LoRA adapters, benchmark inference speed, and serve models via llama-server using llama.cpp. Includes Qwen 3.5 serve scripts (9B dense + F16, 35B MoE) with asymmetric KV cache and thinking mode. Secondary to Ollama; use when needing direct model control or LoRA hot-loading. Triggers on 'llama.cpp', 'GGUF', 'LoRA adapter', 'benchmark inference', 'llama-server'.

日本語の概要は準備中です。原文の説明を表示しています。

Interactively gather parameters for `python -m cli.inference.text_generate`, generate a concrete text generation simulation command, explain assumptions, ask for execution confirmation, then run the command and summarize performance metrics. Use when the user wants to validate a specific model, hardware profile, batch/query size, prefill or decode scenario, TP/DP/EP/MOE parallel strategy, profiling database, chrome trace, empirical metrics export, or a best row produced by throughput_optimizer.

日本語の概要は準備中です。原文の説明を表示しています。

在实现适配器前对候选模型做分析。确定模型实现来源（transformers / 模型目录 / diffusers）、结构特征、内存约束下的逐层量化建议（可选）及 MoE 融合权重风险。适用于用户询问模型适配可行性或做适配前分析时使用。识别为 DiT 扩散模型时，索取推理仓路径后回传 `next_step: model-adapt`，由 orchestrator 委派 `msmodelslim-model-adapt`（DiT 扩展节承接）。

日本語の概要は準備中です。原文の説明を表示しています。

Interactively gather parameters for `python -m cli.inference.throughput_optimizer`, generate a deployment simulation command, explain assumptions, ask for execution confirmation, then run the simulation and summarize the best parallel strategy. Use when the user wants to evaluate a model on one or more hardware profiles, compare hardware targets, compare aggregation versus disaggregation versus PD ratio optimization, choose TP/EP/MOE-DP search ranges, or obtain recommended concurrency, throughput, TTFT, and TPOT results from throughput modeling.

日本語の概要は準備中です。原文の説明を表示しています。

Build or refine the Ideal Customer Profile (ICP) and personas using the Jobs-to-be-Done framework (Clayton Christensen / Bob Moesta). Use when the user asks for ICP research, persona development, "who are our customers", buyer personas, audience research, "build our ICP", target customer definition, or wants to update persona docs. JTBD interview questions, three job dimensions (functional/emotional/social), and the Four Forces of Progress are wired into every persona. Writes to knowledge/icp/personas.md so all other skills get smarter. For the individual buyer persona rather than the segment, see customer-persona. For sourcing real customer language, see customer-research.

日本語の概要は準備中です。原文の説明を表示しています。

融合单元的范围判定与收益前置评估：判定「哪些计算该被融进同一个融合单元、边界画在哪、 融了值不值」。结构侧按融合单元构造规则切边界——函数边界即候选组；FA（含 BSA 等变种）为 禁融锚点；以 norm / rope / FA 为分界分区，全 Vec 区域优先试融；区域内有 Cube 时以 Cube 为 首算子向后包裹 Vec（Σvec 小于 cube 或遇下一 Cube 封口）；一轮之后纯 Vector 单元之间再迭代； MLP / MoE 以矩阵乘族（MatMul / BatchMatMul / GroupedMatmul）为界；若该段计算已由单个 融合算子承载，按规则 6 先确认载体是否已生效（不重复建单元）。 收益侧按计算单元利用率判型（memory_bound、*_vec|mac|mte2|mte3_ratio、cube_utilization）， 叠加带宽下限 / 区域占比 / 天花板法 / 地板先行 / Amdahl 传导校验给出 go/no-go。 当用户问「这几个算子能不能融成一个」「融合边界画在哪」「融多大范围」「融了值不值得」， 或 profiling-analyze 给出融合机会候选需要定范围与收益时使用；即使用户只说「这条链要不要 合成一个 kernel」「这里能不能省一次读写」也应触发。 **已经做完、实测收益却打平或为负时同样触发**——此时的任务不是接受"融合无收益"， 而是按事后复核做**等功忙计数对照**归因，并在缺陷落在融合核自身时**升级 operator-dev**。 边界：只出「融合单元草案 + 收益判定 + 否决理由」。不做采纳与选点（dit-perf-opt 依收益选 top 并派活）、不做采集与瓶颈定位（profiling-collect / profiling-analyze）、不写 kernel、不写 pattern、 不做使能与并行选型、不定义验收标准（引 accuracy-gate / perf-gate）。 由 profiling-analyze（融合机会候选）与 operator-dev（实现前的收益评估）调用。

日本語の概要は準備中です。原文の説明を表示しています。

Run a Mixture of Experts (MoE) audit on any topic, plan, codebase, evidence archive, or process. Dynamically spawns specialized subagents with different critical lenses to cross-reference findings and detect flaws, overlap, retention issues, or drift. Use when designing architectures, analyzing complex code, verifying multi-step plans, classifying evidence artifacts, or looking for duplicated intent in a repo.

日本語の概要は準備中です。原文の説明を表示しています。

Knowledge base from the NASA Systems Modeling Handbook for Systems Engineering (NASA-HDBK-1009A Rev A, 2025). Use for model-based systems engineering (MBSE) wired into NASA's NPR 7123.1 SE processes: the three aspects of MBSE (language / methodology / framework), the NASA SE Engine and its two OOSEM-derived steps (Model Planning, Setting Up the Model), the Modeling Plan as a SEMP subset, the tool-agnostic NASA SE metamodel of elements and relationships across the four SysML pillars (structure / behavior / requirements / parametrics), worked SysML diagrams and tables for stakeholders, requirements, structure and V&V, generating SE work products (ConOps, MOE, MOP, TPM, V&V), the MBSE Grid framework, alternative modeling approaches (PBR, Scenario, System Specification, Verification), and the ConOps model-content template. Scope is bounded to four common technical processes (Stakeholder Expectation Definition, Technical Requirements Definition, Product Verification, Product Validation). Does NOT teach SysML itself, mandate a tool, define NPR 7123.1 processes, cover technical-management processes 10-17, or reproduce the OMG SysML specification.

日本語の概要は準備中です。原文の説明を表示しています。

Knowledge base from NASA NPR 7123.1D — Systems Engineering Processes and Requirements. Use for the 17 common SE processes (SE Engine), life-cycle reviews and entrance/success criteria, SEMP structure and content, ETA roles and tailoring/customizing rules, Technology Readiness Levels (TRL 1–9), SE work product maturity terminology, contracted project oversight, and MOE/MOP/TPM measurement frameworks. Does not cover NASA programme financial management, human spaceflight medical standards, or software engineering requirements in depth (see NPR 7150.2 for the latter).

日本語の概要は準備中です。原文の説明を表示しています。

Knowledge base from the Expanded Guidance for NASA Systems Engineering (NASA/SP-2016-6105-SUPPL, Volume 1, March 2016) — the practitioner-depth supplement to the NASA SE Handbook. Use for the load-bearing mechanics the handbook compresses: the SE Engine's per-phase cadence and the iterative-vs-recursive distinction, the AS9100 crosswalk, life-cycle tailoring vs. customization and the Compliance Matrix, the recursive system-design consistency loop (ConOps vs. Operations Concept, requirement flow/type/ownership, successive refinement), product realization (verify/qualify/accept/certify cardinality, protoflight, 'test the way we fly' coverage, two-form transition), and the crosscutting technical-management procedures (six-step scheduling, cost-estimating method selection, JCL, interface document family, RIDM/CRM, CM baselines, EVM arithmetic, MOE/MOP/KPP/TPM, the three required leading indicators, and the decision-analysis method catalog). This is the DEPTH layer — it deliberately does NOT restate the base definitions; pair it with the nasa-se-handbook pack. Scope: NASA space-flight SE practice (Vol 1 Practices); thin on the Vol 2 process-appendix detail, on non-NASA/commercial life cycles, and on software-engineering methodology.

日本語の概要は準備中です。原文の説明を表示しています。

Knowledge base from NASA Human Systems Integration (HSI) Practitioner's Guide (NASA/SP-2015-3709). Use for HSI process implementation, lifecycle phase activities, function allocation, ConOps development, HSI Plan writing, HSI metrics (MOE/MOP/TPM/KPP), HITL testing strategy, HSI team organisation, scaling/tailoring for program size, and V&V planning. Does not cover the specific human engineering design standards (NASA-STD-3001 content) or detailed HFE methods beyond what the guide itself provides.

日本語の概要は準備中です。原文の説明を表示しています。

Guide for choosing models in Tinker — available model families, model types (base, instruction, reasoning, hybrid, vision), architecture (dense vs MoE), and how to match renderers to models. Use when the user asks which model to use, what models are available, or how to pick a model for their task.

日本語の概要は準備中です。原文の説明を表示しています。

Mendelian randomization (MR) causal inference — does an exposure, risk factor, or biomarker CAUSALLY affect a disease/outcome, using genetic variants as instrumental variables (IEU OpenGWAS / EpiGraphDB MR-EvE). Use this whenever the user asks if X causes Y, whether an observational association is actually causal or just correlation, if a biomarker/trait is a causal risk factor, wants to triangulate epidemiology against genetic evidence, or mentions Mendelian randomization, instrumental-variable analysis, two-sample MR, or genetic causal evidence — even if they never say "MR" (e.g. "is LDL cholesterol actually causal for heart disease?", "does BMI cause type 2 diabetes or just correlate?", "is CRP a causal driver of stroke?"). Covers trait-label resolution, MR effect direction/magnitude, instrument quality (MOE score), method agreement (IVW vs MR-Egger vs weighted median), bidirectional MR for reverse causation, and distinguishing causation from genetic correlation. Not for plain GWAS association lookups (use the GWAS skills) or fitting your own instruments from raw summary statistics.

日本語の概要は準備中です。原文の説明を表示しています。

Multi-objective optimization framework. NSGA-II, NSGA-III, MOEA/D, Pareto fronts, constraint handling, benchmarks (ZDT, DTLZ), for engineering design and optimization problems.

日本語の概要は準備中です。原文の説明を表示しています。

Security-Scanner Agent fuer fabrikIQ und andere Projekte. Fuehrt umfassende Sicherheitspruefungen durch. Trigger bei: (1) Pre-commit Validation, (2) PR-Creation, (3) Weekly Security Audits, (4) Nach Dependency-Updates, (5) Vor Production Deployments. <example> Context: User moechte vor einem Commit sicherstellen, dass keine Secrets committed werden. user: "Ich will gleich committen. Pruefe mal auf Secrets." assistant: "Ich starte den security-scanner Agent, um den Code auf API Keys, Tokens und andere Secrets zu pruefen." <Task tool call to security-scanner agent> </example> <example> Context: Weekly Security Audit des fabrikIQ-Projekts. user: "Fuehre den woechentlichen Security-Scan durch." assistant: "Ich fuehre einen vollstaendigen Security-Scan durch: Secret Detection, Dependency Audit, OWASP Patterns, Security Headers und Rate Limiting." <Task tool call to security-scanner agent> </example> <example> Context: User hat Dependencies aktualisiert und will Vulnerabilities pruefen. user: "Ich habe npm update und pip install -U gemacht. Gibt es Sicherheitsprobleme?" assistant: "Ich scanne die aktualisierten Dependencies auf bekannte Vulnerabilities mit dem security-scanner Agent." <Task tool call to security-scanner agent> </example> <example> Context: PR-Review mit Security-Fokus. user: "Review den PR #42 auf Security-Probleme." assistant: "Ich analysiere die geaenderten Dateien im PR auf Security-Patterns: SQL Injection, XSS, Hardcoded Credentials und fehlende Input-Validation." <Task tool call to security-scanner agent> </example>

日本語の概要は準備中です。原文の説明を表示しています。