Automate VLLM and MindIE benchmark orchestration on Huawei Ascend NPUs.
npx skills add https://github.com/sunchendd/good_skills --skill models-test
Please help me install this Agent Skill. Skill: models-test Source: https://github.com/sunchendd/good_skills/tree/main/models-test Command: npx skills add https://github.com/sunchendd/good_skills --skill models-test
大模型测试自动化框架，用于在华为昇腾NPU上进行VLLM和MindIE推理性能及精度评估。支持自动化NPU资源管理、Docker容器化部署和并行测试。当用户提出以下请求时请使用此技能：测试大模型性能或精度、运行VLLM或MindIE基准测试、生成模型测试命令、或请求执行测试工作流。 Agent需要逐步完整执行测试工作流程，确保资源管理和错误处理。
Run an automated benchmark suite for your models and retrieve a structured results report.
High-intent search queries and answers about installing and using this skill.
Automate VLLM and MindIE model benchmarks on Huawei Ascend NPUs by orchestrating end-to-end performance and accuracy tests, coordinating automatic NPU allocation, Dockerized test services, and script-driven reporting.
Parallel performance tests across multiple models on Ascend NPUs are supported through dynamic NPU allocation and Dockerized test services, enabling multi-model parallel testing with centralized results aggregation and reporting.
Running automated model benchmarks in Docker on Ascend requires preparing the NPU environment, installing dependencies like transformers, pandas, and numpy, and configuring Dockerized test services for VLLM or MindIE orchestration.
EvalScope comparison evaluation is supported alongside VLLM and MindIE performance testing, enabling comprehensive accuracy assessment and benchmarking for large models deployed on Ascend NPUs.
Integrate model benchmark testing into a CI/CD pipeline for Ascend deployments by orchestrating end-to-end VLLM and MindIE benchmarks, ensuring automated NPU resource management and stability verification before production release.
Automated NPU allocation during parallel benchmark orchestration dynamically assigns available Ascend NPU resources to concurrent Dockerized test services, coordinating script-driven benchmarks and preventing resource conflicts.
AscendInstall mmcv-full with NPU support inside Ascend NPU containers.
mejangoParse SARIF 2.1.0 files to extract and deduplicate findings.
sunchenddAutomate end-to-end validation for AI/ML inference services across hardware backends and engines.
tunnox-netIdentify tunneling system bottlenecks and prescribe latency, throughput, and concurrency optimizations.
