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…
A bilingual question-and-answer adviser for Huawei Ascend inference projects, including vLLM-Ascend, MindIE, and ModelSlim repositories. Inference is the process of using a trained model to produce results.
An installation procedure for Ascend NPU drivers and firmware on CentOS, RHEL, Ubuntu, and Debian systems. NPU means a processor designed for neural-network workloads.
Analyze Huawei Ascend NPU profiling data to discover hidden performance anomalies and produce a detailed model architecture report reverse-engineered from profiling. Trigger on Ascend profiling traces, NPU bottlenecks, device idle gaps, host-device issues, kerneldetails.csv / traceview.json / opsummary /…
An index of skills for developing with Huawei’s Ascend Transformer Boost (ATB) library. It groups eight skills covering CANN setup, ATB testing, operator replacement design, and migration to ACLNN for Ascend NPU development.
A workflow for migrating ATB operators to ACLNN on Huawei Ascend 910B and 950 devices. ATB and ACLNN are libraries for implementing computations on Ascend hardware.
A design-document generator for replacing an ATB operator with an ACLNN operator. It uses documentation for both interfaces to describe how their parameters correspond and how the replacement should be tested.
A generator for ATB (Ascend Transformer Boost) operator test cases in CSV format. It creates examples for expected inputs, invalid inputs, and performance testing according to CSV rules.
A test runner for ATB (Ascend Transformer Boost) test cases stored as CSV files. It checks whether ATB operators produce correct results in a CANN environment.
A troubleshooting guide for ATB operator tests and setup. ATB is a software environment for testing computing operators, and an operator is a small piece of a machine-learning or numerical program.
An installation guide for the Ascend NPU NNAL acceleration library. It handles installing the NNAL package, setting its environment variables, and checking that it works.
A standard workflow for migrating ATB OPS implementations to ACLNN. It combines preparation, design documentation, CSV test-case planning, implementation, compilation, and testing into staged checkpoints.
A build tool for the ATB (Ascend Transformer Boost) test framework. It can perform a full build, including third-party dependencies, or an incremental build using an existing setup.
A workflow for generating complete AscendC operator implementations from design documents. AscendC is a programming environment for operators that run on Huawei Ascend AI processors, and an operator is a computation such as a matrix or neural-network operation.
A code-review skill for Ascend C, a C-like language used to write operators for Ascend AI processors. It checks code against supplied safety and coding rules using a hypothesis-testing approach.
A build and debugging guide for AscendC operators, which are custom computing operations for Huawei Ascend AI processors. It covers compiling, installing, and running accuracy tests on NPU hardware.
A design guide for AscendC operators, which are custom computing operations written for Huawei Ascend AI processors. It covers the operator structure, interfaces, performance planning, and how work and memory are divided.
A documentation workflow for AscendC operators, which are compute kernels written for Huawei Ascend AI processors. It generates a PyTorch-style README describing an operator's interface.
A performance-testing procedure for Ascend C operators that keeps profiler test cases in JSONL files and compares custom operators with reference implementations.
A debugging workflow for AscendC operators, which are compute kernels written for Huawei Ascend AI processors. It helps investigate incorrect results such as mismatches, all-zero output, or NaN values.
A precision-testing process for already compiled and installed AscendC operators. AscendC operators are custom computing operations designed for Huawei Ascend processors.