simpo-training
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a comprehensive guide for LLM alignment using the SimPO algorithm. It includes configurations for various models (Mistral, Llama 3, DeepSeek) and datasets (UltraFeedback, HelpSteer, etc.).
- [EXTERNAL_DOWNLOADS]: The skill references official repositories and packages from well-known and trusted organizations, including Hugging Face, NVIDIA, Anthropic, and Princeton NLP. These references are documented for legitimate installation and configuration purposes.
- [COMMAND_EXECUTION]: Provided command-line examples are standard for machine learning workflows, involving environment management with Conda, package installation via Pip, and training orchestration with the Accelerate library.
- [NO_CODE]: The skill consists entirely of instructional Markdown and YAML configuration templates. No executable scripts are bundled directly within the skill; it relies on established external repositories.
Audit Metadata