🧠

huggingface-model-trainer

High Risk

Train and fine-tune language models using TRL on Hugging Face Jobs cloud infrastructure. Supports SFT, DPO, GRPO, reward modeling, and GGUF conversion for local deployment.

@huggingface

🔒Security Analysis

Risk ScoreHigh Risk

Safe6/10Risky

Required Permissions

⌨️

Shell Access

Can execute shell commands

📁

File System

Can read/write files

🌐

Network

Can make network requests

🔑

Credentials

Handles API keys or secrets

Security Notes

Requires HF_TOKEN with write permissions to push trained models. Executes Python scripts on cloud GPU infrastructure, uses secrets to pass authentication tokens. Training jobs can cost $1-$20+ per job. Scripts executed via UV with inline dependencies could potentially include malicious packages.

Significant system access, use with caution