What problem does it solve? When training or inference jobs on Ascend NPU fail with "there is an aicore error" or "aivec error", developers struggle to locate the root cause from raw logs and binary dump files. This Skill guides the use of the msaicerr tool shipped with the CANN Toolkit to analyze fault information, parse dump files into readable .bin/.npy data, and verify the software and hardware environment. ## Core Features & Use Cases - AI Core Error Analysis: Run msaicerr.py with -p on a collected fault information directory to generate an info.txt report that helps locate the failing operator, with guidance on required inputs (dump files, .o/.json compile artifacts, cann logs). - Dump File Parsing and Conversion: Parse dump files into .bin/.npy files containing operator inputs, outputs, and workspace data, and convert .bin files to .npy with a specified dtype such as int8 or bfloat16. - Environment Verification: Run a built-in sample operator with -e to check whether the software and hardware environment is healthy before deeper analysis. - Use Case: A training job crashes with an aicore error. You collect the fault information directory, run the preflight check, then execute msaicerr.py -p to get info.txt, and finally parse the dump file to inspect the failing operator's actual input and output values. ## Quick Start Ask the assistant to analyze an AI Core Error by running msaicerr with the -p option on your fault information directory and reading the generated info.txt.
