training-llms-megatron

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill serves as a technical guide for using the NVIDIA Megatron-Core framework for large-scale model training. It includes detailed configurations for various parallelism strategies (tensor, pipeline, expert, etc.) without any suspicious instructions or hidden payloads.
  • [EXTERNAL_DOWNLOADS]: The skill references official and well-known sources for dependencies, including the NVIDIA Container Registry (nvcr.io) and standard Python package installation for 'megatron-core'. These are reputable infrastructure components for the machine learning domain.
  • [COMMAND_EXECUTION]: All provided command-line examples, such as 'torchrun' for distributed training and 'docker run' for environment setup, are standard practices in high-performance computing and do not perform any unauthorized or harmful actions.
  • [DATA_EXPOSURE]: The skill only references placeholder paths for datasets and model weights (e.g., '/path/to/data'). It does not attempt to access sensitive system files or credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 12:54 AM
Security Audit — agent-trust-hub — training-llms-megatron