---
license: apache-2.0
base_model: Qwen/Qwen3-4B
pipeline_tag: text-generation
tags: [agent-safety, guardrail, safety-monitor, qwen3]
---

# StepSafe-4B-SFT

The supervised warm start of StepSafe-4B (Qwen3-4B, fp32 weights). ASPO training starts from it,
with the StepSafe code ([github.com/carriegu0818/StepSafe](https://github.com/carriegu0818/StepSafe)):

```bash
SFT_CKPT=<path to this model> bash scripts/train_stepsafe_4b.sh
```

## Citation

```bibtex
@misc{stepsafe,
  title  = {StepSafe: Learning Timely Intervention with Context-Aware Safety Reasoning},
  author = {Anonymous},
  year   = {2026}
}
```