huggingface-vision-trainer
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- Remote Job Execution: The skill facilitates the execution of training scripts on cloud GPU infrastructure via the
hf_jobstool andHfApi. This is the intended core functionality and uses standard platform mechanisms for job management. - Secure Authentication: User tokens are managed using job secrets and environment variables (
HF_TOKEN). The scripts correctly implement explicit token injection for TransformersTraineroperations, which is a recommended security practice for ephemeral cloud environments. - Dataset Validation: The inclusion of
scripts/dataset_inspector.pyallows for pre-training validation of datasets using the official Datasets Server API, helping to prevent failures caused by malformed or incompatible data formats. - Well-Known Dependencies: The skill utilizes established machine learning libraries such as
transformers,datasets,timm, andalbumentations. All dependencies are explicitly declared in the script metadata for reliable environment setup. - Integrated Monitoring: Training metrics are logged using
trackio, which the documentation indicates is hosted on Hugging Face Spaces, maintaining data within the intended ecosystem.
Audit Metadata