---
title: grpo_qwen3_0211
canonical_url: "https://www.modelscope.cn/models/augusttailong/grpo_qwen3_0211"
md_url: "https://www.modelscope.cn/models/augusttailong/grpo_qwen3_0211.md"
repository: augusttailong/grpo_qwen3_0211
last_updated: 2026-10-31
model_type:
  - qwen3
architectures:
  - Qwen3ForCausalLM
parameters: 14.8B
tensor_type:
  - F32
library_name:
  - safetensors
downloads: 10
stars: 0
---

# grpo_qwen3_0211

> grpo_qwen3_0211 - augusttailong 在 ModelScope 开源的模型。CBC-aligned Qwen3-14B checkpoint for Staying on Task: Defending Against Prompt Injection via Task Consistency Checking (EMNLP 2026).

- **Repository**: augusttailong/grpo_qwen3_0211
- **Parameters**: 14.8B
- **Downloads**: 10
- **Stars**: 0
- **Last updated**: 2026-10-31

Source: https://www.modelscope.cn/models/augusttailong/grpo_qwen3_0211

---

# CBC — Qwen3-14B

CBC-aligned Qwen3-14B checkpoint for **Staying on Task: Defending Against Prompt Injection via Task Consistency Checking** (EMNLP 2026).

CBC (Consistency Before Complying) improves prompt-injection resistance through task-consistency reasoning, supervised fine-tuning, and GRPO alignment.

**Code:** https://github.com/XOTaichi/CBC-PI

## Download
```python
pip install modelscope

from modelscope import snapshot_download
model_dir = snapshot_download('augusttailong/grpo_qwen3_0211')
```