skills/nvidia/skills/parity-testing/Gen Agent Trust Hub

parity-testing

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill serves as technical documentation for debugging and verifying model parity. It includes bash command templates and Python snippets intended for manual use by developers.
  • [COMMAND_EXECUTION]: Provides standard CLI examples to run local scripts (e.g., hf_megatron_roundtrip.py) using the uv package manager and torch.distributed.run. These operations are consistent with the skill's stated purpose of model conversion testing.
  • [EXTERNAL_DOWNLOADS]: Commands utilize the --hf-model-id flag, which implies interaction with the well-known Hugging Face Hub service to fetch model weights for conversion. This is standard functionality for machine learning workflows.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 05:48 PM
Security Audit — agent-trust-hub — parity-testing