The Agent Skills Directory

[NO_CODE]: The skill is composed solely of Markdown files and reference documents; it does not include any executable scripts or binary files within its package structure.
[COMMAND_EXECUTION]: The documentation provides command-line examples using torchrun and bash to initiate training jobs and handle distributed synchronization.
[COMMAND_EXECUTION]: Suggests a sed utility command for resolving line-ending compatibility issues in local shell scripts.
[PROMPT_INJECTION]: The training workflow involves the ingestion of external datasets, which is a standard procedure for AI training skills.
Ingestion points: Data is loaded from paths provided via the --data-path and --load arguments in SKILL.md.
Boundary markers: No specific delimiters or boundary instructions are included in the configuration templates.
Capability inventory: The training scripts perform file I/O and network operations to manage distributed NPU synchronization.
Sanitization: The provided guide does not define specific validation or sanitization routines for the input training data.

mindspeed-llm-training