nemo-mbridge-perf-moe-hardware-configs
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a repository of hardware-specific playbooks for training MoE models. The information provided, such as tensor parallelism (TP), expert parallelism (EP), and pipeline parallelism (PP) settings, is consistent with high-performance computing best practices.
- [SAFE]: The environment variables recommended (e.g.,
PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True) are standard performance tuning flags for ML workloads and do not pose a security risk. - [SAFE]: Technical documentation links point to official NVIDIA developer resources. The references are intended for developer guidance on model optimization.
- [SAFE]: The skill includes its own evaluation reports (
BENCHMARK.mdandskill-card.md) which indicate rigorous testing for security and correctness, showing no unsafe behaviors were detected during automated trials. - [SAFE]: No malicious patterns such as prompt injection, unauthorized data access, or remote code execution were identified in the instructions or metadata.
Audit Metadata