Loading…
Showing 5911–5940 of 15647 skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| msot-msopprof-operator-profilerkali20gakki/msAgent当用户希望使用 msOpProf(`msprof op` / `msprof op simulator`)对昇腾 AI 算子做上板或仿真性能调优、解释 `aic-metrics`/`trace.json`/`visualize_data.bin`、选择 device vs simulator 路径、排查 `--soc-version`/`--export`/`signal 6`/`Bad address`/热点图或流水图相关问题,或要求生成固定分析报告模板(算子基本信息 / 关键数据 TOP5 / 核心瓶颈 TOP5 / 优化建议 TOP5)时,使用本技能。它负责先判定模式、输入形态与芯片/能力边界,再给出正确命令、结果解读、固定报告输出与高频踩坑规避;不要把经验案例当成通用规则。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| mstx-report-soft-synckali20gakki/msAgent分析 AscendC 算子中自实现的软同步代码,判断其所属同步语义(核间 barrier/set-wait、卡间 barrier/signal),并接入 mssanitizer 的 Sanitizer 接口(sanitizer_report.h)上报软同步语义,避免 racecheck 对软同步误报。触发:算子存在自实现软同步(GM 地址轮询、flag 计数、跨核/跨卡信号等)需要配合 mssanitizer 检测、需要上报软同步语义、需要判断某段同步代码属于哪种同步语义时。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ascend-cluster-comparekali20gakki/msAgentAscend cluster comparison tool. Invoke when user asks to compare two cluster datasets (DB or TEXT), or to generate cluster_analysis_output from raw profiling data via msprof-analyze and then compare. | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ascend-cluster-fast-slow-rank-detectorkali20gakki/msAgent专门用于 Ascend 集群 Profiling 性能数据的“快慢卡”诊断专家技能。当用户提供【集群性能数据目录/路径】并要求分析【快慢卡】、【慢节点】、【负载不均衡】或【集群瓶颈】时,必须触发此技能。该技能会自动接收集群路径,调度相关工具输出快慢卡的宏观定性与微观根因(如 Host 下发瓶颈、算子计算劣化)。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ascend-msprof-analyze-clikali20gakki/msAgentMindStudio Profiler Analyze(msprof-analyze)是面向 AI 训练与推理场景的性能分析工具,基于采集得到的 profiling 数据进行统计、比对和诊断,帮助定位计算、通信、调度及集群场景下的性能瓶颈。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ascend-profiler-data-validationkali20gakki/msAgent当用户提供 MindStudio profiler 采集的性能数据(框架 profiler、msprof 命令行)时,对数据完整性、采集状态及关键配置进行校验,确保后续分析工具能正常运行。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ascend-schedule-analysiskali20gakki/msAgentAnalyze Ascend NPU schedule, operator dispatch, operator launch, and Host Bound profiling issues in Ascend profiling data. Use when need to diagnose device Free time, framework/operator dispatch latency, launch latency, PYTORCH_API/CANN_API launch gaps, aclrtSynchronizeStream stalls, task queue behavior, CPU scheduling interference, GC/lock pauses, CPU affinity, or schedule-side optimization actions. | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| mindstudio-cpu-bindingkali20gakki/msAgentUse when diagnosing NPU + PyTorch or LLM Serving Host CPU affinity, NUMA locality, cgroup/cpuset constraints, CPU range conflicts, PyTorch/runtime threading, DataLoader, tokenizer, scheduler, vLLM-Ascend, SGLang, TTFT, TPOT, tokens/s, QPS, or multi-rank/multi-worker CPU binding issues. | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| msagent-profiler-breakdownkali20gakki/msAgentDB格式Profiling数据场景化性能拆解:自动完成场景确认(框架/任务类型判定)、阶段拆解与单次执行边界拆解,并给出报告。Use when 需要对昇腾 Profiling DB 做阶段拆解、step 归因或生成性能报告, or when the user says "帮我按任务阶段拆解这个Profiling数据", "多层次性能耗时拆解", "分析这个 vllm/sglang/verl 数据,主要耗时都花费在哪里了"。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| prometheus-metrics-analyzerkali20gakki/msAgent分析 SGLang、vLLM 和 vLLM-Ascend 推理服务的 Prometheus 原生历史 Runtime Metrics。通过动态发现指标、建立语义映射、查询核心指标和多轮补充证据,分析延迟、吞吐、调度及缓存等潜在瓶颈。不负责部署或参数优化。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| gen-evaluation-cfgkali20gakki/msAgentGenerate msmodelslim evaluation YAML configuration (service_oriented + aisbench + vllm-ascend). Use when user asks for evaluation config generation. | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| msmodelslim-anti-outlier-adaptkali20gakki/msAgent在基础模型适配完成后,按算法适配并安装 Adapter 接口,依据目标模型真实 forward 生成并验证 final-logits patch,执行单算法 msModelSlim 离群值抑制变换、官方计算图提取和浮点 logits 门禁。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| msmodelslim-model-adaptkali20gakki/msAgent为 msModelSlim 创建基础 Transformers 模型适配器(Model Adapter)。 包含创建适配器、实现必需接口与注册安装流程。 适用:Decoder-only LLM、理解类 VLM(仅 LLM/text 部分)。 多模态生成(DiT/扩散,含双/多专家)见下方「多模态生成扩展」节,按需加载。 不适用:Encoder-only、非 Transformers 架构。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| msmodelslim-model-dequantkali20gakki/msAgent为 msModelSlim 适配流程注入反量化能力。先识别模型权重是否可反量化,再实现反量化脚本并接入 model_adapter。当前仅覆盖 FP8 的 per-block 与 per-channel 两类;若格式不确定或无公开反量化规则,要求用户提供反量化脚本或浮点权重。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| msmodelslim-quick-quantkali20gakki/msAgent提供 msModelSlim 的通用快速量化指引,包含安装、最简 YAML 配置与基础执行校验。适用于用户询问 msmodelslim 安装、快速量化、配置 yaml、linear_quant 或 minmax 基础参数时。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| quant-tuning-evaluatekali20gakki/msAgent执行模型测评(统一入口)。LLM/VLM 走主流程 scripts/run_evaluation.py + Evaluation YAML;model_family=dit 走 DiT 扩展节(references/dit/evaluate_workflow.md,vbench.py 批量推理产出视频)。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| quant-tuning-history-append-ditkali20gakki/msAgentAppend a DiT-tuning round to history.yaml. DiT-side counterpart of the LLM-path accuracy_append.py; records practice.md5, inference_outputs, fp_baseline_outputs, and the scoring fields (scores / overall_score / loss_vs_baseline / is_satisfied) populated by quant-tuning-score-dit. Idempotent on practice_id. | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| tune-practice-cfgkali20gakki/msAgentUse when 量化调优闭环中需要生成或修改一轮调优所需的 Practice YAML,包括敏感层分析、策略决策、写出 YAML 文件和校验。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| speculators-response-regenkali20gakki/msAgent重采样:把已有多轮对话数据的每条 assistant 回答丢弃,用 verifier 模型/vLLM 服务逐轮重生成,产出可直接进训练的预分词样本(每行含 input_ids/loss_mask)。依赖支持 return_token_ids 的 OpenAI 兼容 chat 服务(服务需由用户提前就绪,本技能不负责拉起服务)。触发:用某模型对某个数据集做重采样/重生成(如“对这个数据集用这个模型做重采样”)、把对话回答用模型重写、生成预分词样本/on-policy/MTP 训练数据。 | Stars: 32 | Updated | Claude CodeCodex | Verified | Install |
| ghi-closetvproductions/gzkitUse to do the work a GHI describes, verify it and close the issue with evidence in one sweep. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-closeout-ceremonytvproductions/gzkitUse to run an ADR's closeout ceremony through to the operator's attestation. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-createtvproductions/gzkitUse to create and book an ADR with its OBPI briefs under five-gate compliance. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-emit-receipttvproductions/gzkitUse to record a completed, validated or closed ADR receipt event with its evidence. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-evaluatetvproductions/gzkitUse after authoring an ADR or OBPI to score its quality and optionally run the red-team challenges. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-maptvproductions/gzkitUse to trace an ADR to its artifacts with gz state and repository search. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-promotetvproductions/gzkitUse to promote a pool ADR (ADR-pool.*) into an active, versioned ADR package. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-statustvproductions/gzkitUse for the ADR table, or one ADR's lifecycle and OBPI detail. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-adr-synctvproductions/gzkitUse to reconcile an ADR end to end by discovering @covers evidence, reconciling OBPI ledger state and registering files. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-advisor-qctvproductions/gzkitUse between gz content compose and corpus attestation to judge a candidate rendition and record its advisory QC receipt. | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |
| gz-airlocktvproductions/gzkitUse to read a target's seam map before touching it (gz airlock in), account for what a transit disturbed (gz airlock out), or enter ad hoc (gz permitted-entry). | Stars: 9 | Updated | Claude CodeCodex | Verified | Install |