Automate HCCL performance testing across Ascend NPU clusters with MPI.
npx skills add https://github.com/Ascend/agent-skills --skill hccl-test-ascend
Please help me install this Agent Skill. Skill: hccl-test Source: https://github.com/Ascend/agent-skills/tree/main/skills/hccl-test Command: npx skills add https://github.com/Ascend/agent-skills --skill hccl-test-ascend
Automates and validates HCCL performance testing across Ascend NPU clusters, enabling quick assessment of distributed bandwidth and verification of HCCL functionality for common collective operations such as AllReduce and AllGather.
Run the quick verification script to perform a baseline HCCL test.
High-intent search queries and answers about installing and using this skill.
Automate HCCL performance testing across Ascend NPU clusters by coordinating multi-node benchmarks via MPI, assessing distributed bandwidth for common collective operations like AllReduce and AllGather.
Pre-flight checks for HCCL testing require verifying SSH access across nodes, ensuring CANN version consistency, and checking NPU health to guarantee a reliable multi-node test environment.
HCCL testing supports both MPICH and Open MPI implementations, applying MPI-based execution to orchestrate core collective operations and parse performance results across multi-node environments.
Validate HCCL functionality for collective operations including core operators like AllReduce, AllGather, AlltoAll, and Broadcast, as well as optional V-series operators across Ascend NPUs.
HCCL multi-node tests often fail due to inconsistent CANN versions, lack of SSH access between nodes, or missing MPI installations, which pre-flight checks are designed to catch before execution.
ascend-ai-codingBenchmark HCCL operations like AllReduce and AllGather across Ascend NPUs with MPI orchestration.
FeRhodiumMigrate PyTorch training code to Ascend NPU with AMP and HCCL.
camoneartDesign microservice boundaries, contracts, and inter-service communication patterns for distributed systems.
kzahediRun performance benchmarks on YARS simulations and detect regressions.
