Explore skills
Showing 24751–24780 of 71536 skills
MindStudio 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。
kali20gakki/msAgent
Stars: 32Claude CodeCodexCompares two Prof datasets (GPU vs NPU / NPU vs NPU) by auto-installing and calling msprof-analyze compare, then parses the result xlsx into HTML report and Chinese xlsx. Invoke when user provides two prof dirs to compare, or has a compare result xlsx to analyze/translate.
kali20gakki/msAgent
Stars: 32Claude CodeCodexAdapt torch_npu.profiler interface for AI frameworks and training/inference code on Ascend NPU. Use this skill whenever users need to integrate profiling, add performance data collection, or instrument PyTorch code with torch_npu.profiler on Huawei Ascend NPU platforms. Triggers on: "torch_npu profiler", "NPU profiling", "Ascend profiler", "profiler adaptation", "profile training", "profile inference", "add profiler to framework", "性能采集", "profiling 数据", "昇腾 profiler", and any request to add performance monitoring to Ascend NPU code.
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex面向 Ascend PyTorch Profiler / msprof DB(如 ascend_pytorch_profiler*.db、msprof_*.db)的 SQL 分析技能。将自然语言问题(算子耗时、通信、下发、调度、schema/table 查询)转为安全可执行 SQL,并按需从官方文档提取表结构详情。
kali20gakki/msAgent
Stars: 32Claude CodeCodexAnalyze 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.
kali20gakki/msAgent
Stars: 32Claude CodeCodex对 msMemScope 显存数据做全链路处理:根据用户的显存问题诉求(OOM、显存泄漏、显存占比拆解、Step 间对比等),指导并帮助用户以最合适的 msMemScope 采集配置完成显存数据采集(支持命令行、Python 接口、mstx 打点三种采集方式,覆盖训练、推理 vLLM-Ascend、FSDP、强化学习 verl 等场景);对已采集的 memscope_dump_*.csv/db 数据做系统性解读与问题诊断(整体显存曲线趋势分析、显存使用大头按模块/组件拆解分析、显存峰值点各模块与流程占比分析,以及对显存泄漏、OOM 等典型问题的根因定位与原因分析)。
kali20gakki/msAgent
Stars: 32Claude CodeCodexUse 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.
kali20gakki/msAgent
Stars: 32Claude CodeCodexDB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"。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodexGenerate msmodelslim evaluation YAML configuration (service_oriented + aisbench + vllm-ascend). Use when user asks for evaluation config generation.
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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)。
kali20gakki/msAgent
Stars: 32Claude CodeCodexAppend 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.
kali20gakki/msAgent
Stars: 32Claude CodeCodex在 W8A8 动态量化后,用 MindIE-SD 冒烟测试 DiT 量化权重是否可加载(可选 skill)。与 `quant-tuning-evaluate`(DiT 扩展节)的区别:本 skill 仅做"能跑通 + 权重可用"的轻量验证(4~16 条 prompt,几分钟出结果),不算分;后者产出视频集交 `quant-tuning-score-dit`(AISBench-VBench 16 维度)评分,是调优回路一环。不引入 vLLM-Ascend、不做精度指标评估(FID/CLIP-score 等)。
kali20gakki/msAgent
Stars: 32Claude CodeCodexAggregate DiT inference outputs into per-dimension VBench metrics via AISBench. Run after the quant-tuning-evaluate DiT workflow produces a directory of mp4s and the user supplies a VBench-1.0-mini JSON + pre-populated cache dir. Single mode (Standard).
kali20gakki/msAgent
Stars: 32Claude CodeCodex量化专家经验调优库:按「L1 通用专家调优意见 → L2 结构化专家调优意见 → L3 模型专属专家调优意见」三级递进, 输出量化精度调优的完整手段(离群值抑制 / 量化方法·粒度·对称性选型 / 校准集调整 / 敏感层回退 / 模型专属策略), 每条结论附专家原因分析与专家意见可信度等级。 本 Skill 是专家经验库,负责说明可选离群值抑制算法并询问用户选择哪些算法,以及回答「怎么调、哪些层需要回退及为什么」;不运行离群值抑制 processor,不修改 YAML、不执行量化、不做 EP 检查 / 服务化 / 任务评测。
kali20gakki/msAgent
Stars: 32Claude CodeCodexUse 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。
kali20gakki/msAgent
Stars: 32Claude CodeCodex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。
kali20gakki/msAgent
Stars: 32Claude CodeCodexCreate distinctive, production-grade frontend interfaces with high design quality that avoids generic AI aesthetics. Use when the task involves `build web components`, `create frontend interface`, `build web page`, `build web application`, `frontend design`, `create UI component`, or `design web interface`.
yacosta738/yacosta738.github.io
Stars: 47Claude CodeCodex中文文档生成器 - 从代码库自动生成中文技术文档
laolaoshiren/claude-code-skills-zh
Stars: 913Claude CodeCodexUse this skill to keep the platform rules that Coupler.io skills state as fact current: budget flexibility, deliverability thresholds, attribution windows, published benchmarks, plan limits, sale dates. Triggers include: 'refresh the platform rules', 'check the Google Ads rules', 'is the Klaviyo benchmark still current', 'a platform changed its budget rules', 'add this rule to the registry'. Works from registry.md, a catalog of every rule, its source, when it was checked and which skills use it, so a refresh checks the sources instead of rereading every skill.
coupler-io/skills
Stars: 41Claude CodeCodexUse this skill when reviewing a Coupler.io skill before merge, auditing an existing one, or checking a pack of skills for accuracy. Triggers include: 'review this skill', 'check this SKILL.md', 'is this skill accurate', 'review PR #N in the skills repo', 'audit the <platform> pack', or any request to check what a skill claims about a Coupler.io connector or an ad platform. Checks factual accuracy against live tools and repo conventions; it does not judge domain expertise.
coupler-io/skills
Stars: 41Claude CodeCodexLook up any arxiv paper on alphaxiv.org to get a structured AI-generated overview. This is faster and more reliable than trying to read a raw PDF.
wandb/senpai
Stars: 39Claude CodeCodexCreate a typed assignment branch and draft PR for one student. Use when the advisor has a concrete hypothesis and the student has no open `status:wip` or `status:review` assignment.
wandb/senpai
Stars: 39Claude CodeCodexOpen GitHub Issues for human input and respond to the researcher team. Use this skill whenever you need to handle a human_issue event, respond to human issues, ask humans a question, or check team communications. Also triggers for: "any human messages?", "check issues", "respond to humans".
wandb/senpai
Stars: 39Claude CodeCodexMaintain and publish the advisor-owned current research state, durable research-ideas syntheses, and dataset analysis. Use when evidence or human guidance changes the research focus or direction, durable research ideas should be recorded, or dataset understanding changes.
wandb/senpai
Stars: 39Claude CodeCodex