llm-fine-tuning
Pass
Audited by Gen Agent Trust Hub on Mar 9, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches machine learning models from the Hugging Face Hub (meta-llama) and standard training datasets, which are well-known services in the artificial intelligence community.
- [EXTERNAL_DOWNLOADS]: Installs verified Python packages such as transformers, peft, and axolotl from the Python Package Index (PyPI).
- [COMMAND_EXECUTION]: Runs training operations using command-line interfaces for accelerate and deepspeed, which is the intended and primary function of the infrastructure skill.
- [CREDENTIALS_UNSAFE]: Demonstrates secure handling of sensitive credentials (HF_TOKEN, WANDB_API_KEY) through environment variables and Kubernetes Secret references rather than hardcoding.
Audit Metadata