Results for “benchmark”

Use this skill to measure performance baselines, detect regressions before/after PRs, and compare stack alternatives.

The English summary is being prepared. Showing the original description.

Competitive benchmarks, solution research, and hypothesis generation for identified bottlenecks

The English summary is being prepared. Showing the original description.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

The English summary is being prepared. Showing the original description.

Build data-driven compensation structures that attract talent without overpaying. Covers base salary bands, equity/bonus frameworks, geographic differentials, and total rewards packaging.

The English summary is being prepared. Showing the original description.

You are an employee benefits administration specialist. Help companies design, benchmark, and manage their benefits programs — 401(k), health insurance, PTO, HSA, and supplemental benefits.

The English summary is being prepared. Showing the original description.

Analyze, benchmark, and reduce your customer acquisition cost across every channel.

The English summary is being prepared. Showing the original description.

Holistic evaluation framework for language models measuring accuracy, calibration, robustness, and fairness

The English summary is being prepared. Showing the original description.

Comprehensive reference for inter-process communication mechanisms -- every way processes can talk on a computer. Covers sockets (TCP, UDP, Unix domain), WebSockets, SSE, pipes (named/anonymous), shared memory (mmap, shm_open), message queues, signals, D-Bus, XPC (macOS), gRPC, REST, stdin/stdout, file-based coordination, and clipboard. Performance benchmarks, platform availability, and code examples for each. Special focus on which IPC works best for AI agent coordination. Activate on: "IPC", "inter-process communication", "process communication", "how do I talk between processes", "Unix socket vs TCP", "shared memory", "named pipe", "WebSocket vs SSE", "gRPC vs REST", "agent coordination IPC", "XPC service", "message passing", "stdout pipe", "D-Bus", "mmap". NOT for: distributed systems design (use distributed-systems), network protocol design (use networking), message queue infrastructure like Kafka (use data-pipeline-engineer).

The English summary is being prepared. Showing the original description.

Use when predicting Threads post performance, analyzing post history patterns, estimating engagement ceiling for a draft, or deciding what content type to write next. Works with or without personal data — uses darkseoking benchmark as fallback.

The English summary is being prepared. Showing the original description.

Audit and improve Generative Engine Optimization as a measurable discovery-to-citation pipeline. Use for /geo, AI-search visibility, citation readiness, and GEO benchmark work.

The English summary is being prepared. Showing the original description.

Scope, preregister, execute, audit, normalize, and publish GEODE evaluations with research-question, reproduction, attempt-lineage, trajectory, verifier-receipt, and immutable artifact contracts. Use for benchmark runs or comparisons, eval-artifact and cache-accounting audits, GPT or subscription result normalization, research question or hypothesis framing, retry diagnosis, trajectory publication, DPO or reward-data readiness reviews, and requests involving MCPMark, tau2, BFCL, Agent-World, HAL, Terminal-Bench, or Toolathlon.

The English summary is being prepared. Showing the original description.

算子级开发与性能优化：Triton / Ascend C / Catlass / PyPTO / TileLang 算子 编写、精度对齐与性能调优。优先路由到外部 cannbot-skills 技能库（不重复其内容）， 本 skill 只保留场景 → skill 映射与 MindIE-SD 特有补充。 当用户需要新增/优化算子、定位算子性能或精度问题时使用此 skill。 即使用户只提到"写个 triton kernel"或"这个算子怎么加速"而未说算子，也应触发； pattern 融合/编译后端见 pattern-dev，算子基准选型/接入测试见 benchmark-dev。 由 dev-workflow 或 pattern-dev 的 replacement kernel 场景触发。

The English summary is being prepared. Showing the original description.

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。

The English summary is being prepared. Showing the original description.

MindIE-SD 仓库开发总入口（侧轨）。当用户进行 MindIE-SD 的任何代码开发工作时使用此 skill—— 包括但不限于写 pattern、改测试、部署到昇腾、跑 benchmark、性能分析、多卡并行、复盘归档。 模型/三方框架自动优化类任务（非本仓代码改动）由 model-auto-optimization 入口承接， 本入口只在优化流程需要新增 pattern/算子/部署代码时承接其指向的开发子任务。 即使用户未明确提到"开发流程"，只要涉及 MindIE-SD 代码改动都应触发。

The English summary is being prepared. Showing the original description.

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.

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。

The English summary is being prepared. Showing the original description.

Unsupervised clustering and cell-type identification for high-dimensional flow, spectral, and mass cytometry - FlowSOM, PhenoGraph, FlowSOM-via-CATALYST, with UMAP/tSNE for visualization. Covers the type-vs-state marker distinction (cluster on lineage, test state within clusters), over-provision-then-metacluster, the Weber-Robinson benchmark, seed dependence and metacluster stability, why embeddings are for looking not measuring, and median-heatmap annotation/merging. Use when discovering populations without predefined gates, choosing a clustering algorithm, selecting the number of metaclusters, or annotating clusters into cell types.

The English summary is being prepared. Showing the original description.

Infer orthologous genes and gene families across species using OrthoFinder3 (HOG-based phylogenetic orthology), SonicParanoid2, Broccoli, ProteinOrtho, OMA / FastOMA hierarchical orthologous groups, eggNOG-mapper, JustOrthologs, and TOGA whole-genome-alignment orthology. Use when building single-copy ortholog sets for phylogenomics, classifying co-orthologs and in/out-paralogs after gene duplication, propagating functional annotation via orthology with awareness of the ortholog conjecture, distinguishing speciation from duplication via gene-tree species-tree reconciliation, computing Quest-for-Orthologs benchmark performance, or running synteny-aware ortholog detection in WGD-affected lineages.

The English summary is being prepared. Showing the original description.

Analyzes base-editing screens for variant function. Covers library design (Hanna 2021 ClinVar-scale CBE screen benchmarked on BRCA1/2, Cuella-Martin 2021 DDR saturation), CBE vs ABE chemistry choice (BE3/BE4 vs ABE7.10/ABE8.20/ABE8e), editing-window math (positions 4-8 from PAM-distal end; 4-7 for ABE7.10), bystander-edit quantification and the variant-call ambiguity it creates, sgRNA-efficiency filtering before hit calling, indel byproduct interpretation, the substitution-vs-indel diagnostic, variant annotation against ClinVar / COSMIC, and the Broad be-validation-pipeline. Use when designing a BE variant screen, choosing CBE vs ABE for a specific edit, interpreting bystander-confounded hits, distinguishing functional signal from indel artifact, integrating CRISPResso2 output with screen scoring, or deciding BE vs PE for SNV installation.

The English summary is being prepared. Showing the original description.

Analyzes CUT&RUN (Skene Henikoff 2017) and CUT&Tag (Kaya-Okur 2019) chromatin profiling data. Handles SEACR vs MACS2 peak calling (with the btaf375 2025 benchmark guidance), pA-MNase vs pA-Tn5 vs pAG-Tn5 chimera differences, E. coli spike-in carryover normalization, IgG-only control logic (no input), characteristic fragment-size signatures (25-75 bp for CUT&Tag), and lower depth requirements (5M reads typical vs 25M for ChIP). Use when calling peaks from CUT&RUN/CUT&Tag, scaling by E. coli spike-in carryover, choosing SEACR norm mode, or comparing CUT&RUN/Tag results to traditional ChIP.

The English summary is being prepared. Showing the original description.

Autonomously improve a real artifact (code, training recipe, agent harness, data pipeline, prompt) against an objective and an evaluator, using Hypothesis Tree Refinement (HTR) from the Arbor paper. Use this whenever someone wants to iteratively optimize something over many experiments without overfitting — e.g. "get my model's eval score up", "improve this agent/harness", "tune this pipeline", "beat the baseline on this benchmark", "run a search over approaches and keep the best", "do an MLE-bench / Kaggle-style optimization", or any long-horizon "make this artifact better and don't just memorize the dev set" task. Trigger it even when the user doesn't say "Arbor" or "hypothesis tree" but describes repeated experiment-and-evaluate loops, branching exploration of competing ideas, or worries about a dev/test gap. Runs Claude itself as the coordinator with subagent executors in isolated git worktrees; for the standalone `arbor` CLI tool see references/arbor-upstream.md.

The English summary is being prepared. Showing the original description.

Structured SE task evaluation using 15 benchmark definitions from claude-code-harness research

The English summary is being prepared. Showing the original description.

Quantitative agent evaluation using 4-metric framework (correctness/step_ratio/tool_call_ratio/latency_ratio) with ideal trajectory annotation and capability-categorical taxonomy. Use when measuring agent efficiency, comparing agent variants, or gating new agents through correctness→efficiency phases. Complements harness-eval (SE benchmarks) and evaluator-optimizer (qualitative rubric).

The English summary is being prepared. Showing the original description.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

The English summary is being prepared. Showing the original description.