nemo-mbridge-perf-moe-dispatcher-selection

Pass

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides performance tuning guidance and configuration parameters for Mixture of Experts (MoE) token dispatchers (alltoall, DeepEP, HybridEP) targeting specific hardware such as H100, B200, and GB200 systems.
  • [EXTERNAL_DOWNLOADS]: The documentation references requirements for 'DeepEP' and 'HybridEP' runtime packages. These are legitimate performance libraries within the described deep learning context and no suspicious remote execution patterns were found.
  • [COMMAND_EXECUTION]: The skill includes static examples of command-line flags (e.g., --moe-deepep-num-sms 20) intended for model training configurations. These are provided as instructional documentation and do not involve arbitrary or malicious shell execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 12, 2026, 07:41 PM
Security Audit — agent-trust-hub — nemo-mbridge-perf-moe-dispatcher-selection