LPG Qwen3-4B Checkpoint-5200
This repository contains a checkpoint of Latent Policy Guard (LPG) based on Qwen3-4B.
Training Configuration
- Base model: Qwen3-4B
- Epochs: 3
- Learning rate: 5e-5
- Seed: 11
- Checkpoint step: 5200
Intended Use
Research on policy violation detection and LLM guardrails.
Limitations
This checkpoint is released for research purposes only.