A repository for deploying agents/skills for parameter tuning, profiling and analysis.
Skill Claude CodeCodex
Probe or install AISBench from source inside a target container for Path C PD-ratio benchmarking. Runs after pd-config-env-check and before pd-deploy. Skip if aisbench is already available. Pin numpy for vLLM Ascend co-install.
Skill Claude CodeCodex
Finds the best vLLM-Ascend deployment config (low-latency or high-throughput) from local baseline markdown docs by matching device, model, quantization, NPU count, and deploy strategy; if no local match, fetches high-throughput configs from the official vLLM-Ascend model tutorials. Adjusts context length by…
Skill Claude CodeCodex
A diagnostic skill for finding fast and slow devices in Ascend cluster profiling data, where a cluster is a group of computers or accelerators working together.
Skill Codex
Analyze Ascend NPU collective communication profiling data with a DB-first workflow. Use when the user provides clusteranalysisoutput/clusteranalysis.db, rank-level analysis.db, rank-level ascendpytorchprofiler{rankid}.db, together with profilerinfo.json, and asks about HCCL or hcom communication cost, collective…
Skill Claude CodeCodex
Analyze Ascend NPU computation-side profiling data for single-card runs or a selected rank from multi-card runs. Use this skill when the user asks to diagnose computation bottlenecks, AI Core / AI Vector / AICPU hotspots, dynamic shape overhead, block dim issues, redundant TransData/Transpose/Cast, cross-stream waits…
Skill Claude CodeCodex
A tool for collecting, analyzing, and comparing information from Ascend NPU environments and dump files.
Skill Claude CodeCodex
A command-line analysis tool for profiling data from AI training and inference. Profiling data records how a program spends time and resources while it runs.
Skill Claude CodeCodex
A validation guide for performance data collected by MindStudio profiler tools. Profiling records how software uses time and resources so performance problems can be studied.
Skill Claude CodeCodex
A SQL analysis skill for Ascend PyTorch Profiler and msprof database files. It turns questions about operator timing, communication, scheduling, or database tables into safe SQL queries and can look up official schema details.
Skill Claude CodeCodex
Analyze 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, PYTORCHAPI/CANNAPI launch gaps, aclrtSynchronizeStream stalls, task queue behavior, CPU scheduling…
Skill Claude CodeCodex
Ascend cluster performance analysis and comparison tool. Invoke when user asks to analyze cluster profiling data (DB or TEXT format), generate cluster analysis reports, or compare two cluster datasets.
Skill Claude CodeCodex
Parses msprof-analyze compare output xlsx, extracts key insights per sheet, generates HTML report and Chinese xlsx. Pipeline-only for Path B after ascend-msprof-analyze-cli; do not invoke as a standalone tool.
Skill Claude CodeCodex
Enumerates legal DP×TP×EP parallel combinations for vLLM-Ascend single-node mixed deployment from model weight size and NPU count. Use when computing min TP, listing parallel strategy candidates, or as sub-skill of serving-parallel-strategy-tuning.
Skill Claude CodeCodex
A helper for reading source code, configuration, README files, Markdown, and documentation from GitHub links or repositories.
Skill Claude CodeCodex
Reads model feature support table (xlsx) and converts to compact JSON. Invoke when user needs to parse or convert model-feature xlsx to JSON, or query model feature support across versions.
Skill Claude CodeCodex
Installation and setup instructions for connecting msprof-mcp, a server that lets coding tools analyse Ascend performance-profile data, to tools such as Cursor or Claude Code.
Skill Claude CodeCodex
A calculator for MFU, or Machine FLOP Utilization, which measures how much of a machine's theoretical calculation capacity an operation uses.
Skill Claude CodeCodex
Validate PD-disaggregation Prefill/Decode launch commands and environment for Path C. Rewrite only network fields plus computed kvport/engineid/dpsize; never rename kvconnector. Renders mooncakemaster + dual proxy (fetched from vllm-ascend). Use for PD ratio / PD deploy check.
Skill Claude CodeCodex
Deploy Mooncake master then Prefill/Decode/Proxy from Path C rendered scripts. Use after pd-config-env-check and aisbench-install. On failure diagnose and rollback to Phase 1.
Skill Claude CodeCodex
Run AISBench tests to measure Prefill-side and Decode-side QPS under SLO, then compute best PD instance ratio (Path C Phase 4). Use after aisbench-install.
Skill Claude CodeCodex
A tool that reads service startup logs and extracts non-default settings such as model paths, quantization, parallel-processing choices, configuration, and feature switches into an Excel report.
Skill Claude CodeCodex
Estimates available KV cache capacity and memory-bound max concurrency for each DP×TP×EP parallel combination under vLLM-Ascend. Use after find-possible-parallel-strategy or when evaluating KV headroom vs context length.
Skill Claude CodeCodex
Orchestrates vLLM-Ascend single-node parallel strategy tuning from baseline config: resolve SLO constraints, clone vllm-ascend/msmodeling, then run find-possible-parallel-strategy, serving-kv-cache-capacity, and serving-slo-concurrency. Use in Phase 2 serving tuning.
Skill Claude CodeCodex
A guide to measuring the runtime performance of a software service, including its required calls, CSV output format, 11 extracted measurements, and supported log formats.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: