Loading...
Loading...
Found 15 Skills
HCCL (Huawei Collective Communication Library) performance testing for Ascend NPU clusters. Use for testing distributed communication bandwidth, verifying HCCL functionality, and benchmarking collective operations like AllReduce, AllGather. Covers MPI installation, multi-node pre-flight checks (SSH/CANN version/NPU health), and production testing workflows.
Static inspection of Triton operator code quality (Host side + Device side) for Ascend NPU. Used when users need to identify potential bugs, API misuses, and performance risks by reading code. Core capabilities: (1) Ascend API constraint compliance check (2) Mask integrity verification (3) Precision processing review (4) Code pattern recognition. Note: This Skill only focuses on static code analysis; compile-time and runtime issues are handled by other Skills.
Evaluate the performance of Triton operators on Ascend NPU. It is used when users need to analyze operator performance bottlenecks, collect and compare operator performance using msprof/msprof op, diagnose Memory-Bound/Compute-Bound bottlenecks, measure hardware utilization metrics, and generate performance evaluation reports.
Generate interface documents for Triton operators of Ascend NPU. Used when users need to create or update interface documents for Triton operators of Ascend NPU. Core capabilities: (1) Generate standardized documents based on templates (2) Support the list of Ascend NPU product models (3) Provide specifications for operator parameter descriptions (4) Generate call example frameworks.
Generate Triton operator requirement documents suitable for Ascend NPU. Used when users need to design new Triton operators, write operator requirement documents, or perform operator performance optimization design.
Create Docker containers for Huawei Ascend NPU development with proper device mappings and volume mounts. Use when setting up Ascend development environments in Docker, running CANN applications in containers, or creating isolated NPU development workspaces. Supports privileged mode (default), basic mode, and full mode with profiling/logging. Auto-detects available NPU devices.
将简单Vector类型Triton算子从GPU迁移到昇腾NPU。当用户需要迁移Triton代码到NPU、提到GPU到NPU迁移、Triton迁移、昇腾适配时使用。注意：无法自动迁移存在编译问题的算子。
Generate Triton kernel code for Ascend NPU based on operator design documents. Used when users need to implement Triton operator kernels and convert requirement documents into executable code. Core capabilities: (1) Parse requirement documents to confirm computing logic (2) Design tiling partitioning strategy (3) Generate high-performance kernel code (4) Generate test code to verify correctness.
Task Orchestration for Full-Process Development of Ascend Triton Operators. Used when users need to develop Triton Operators, covering the complete workflow of environment configuration → requirement design → code generation → static inspection → precision verification → performance evaluation → document generation → performance optimization.
vLLM Ascend plugin for LLM inference serving on Huawei Ascend NPU. Use for offline batch inference, API server deployment, quantization inference (with msmodelslim quantized models), tensor/pipeline parallelism for distributed serving, and OpenAI-compatible API endpoints. Supports Qwen, DeepSeek, GLM, LLaMA models with Ascend-optimized kernels.
Optimize the performance of Triton operators optimized for Ascend NPU. This guide is for users who need to optimize the performance of Triton operators on Ascend NPU, resolve UB overflow, improve Cube unit utilization, and design Tiling strategies.
Migrate GPU/CUDA Triton operators to Triton-Ascend, or rewrite Python/PyTorch operators into Triton-Ascend implementations that can run on Ascend NPU. When clear optimization opportunities are identified, directly output the optimized code, minimal validation script, and troubleshooting instructions. This skill should be prioritized when users mention 昇腾 (Ascend), Ascend, NPU, triton-ascend, Triton operator migration, PyTorch operator rewriting, coreDim, UB overflow, 1D grid, physical core binding, block_ptr, stride, memory access alignment, mask performance, dtype degradation, operator optimization, or directly ask questions like "How to use this skill", "How to run it in the command line", "How to perform migration/validation in a container", even if users do not explicitly say "write a skill" or "perform migration".