GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Sejibeji
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
Sejibeji
Model Tree
Qwen/Qwen2.5-0.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Qwen/Qwen2.5-0.5B-Instructr=16, alpha=32, lr=3e-4), fully reproducible on CPU
from the repository above| metric | value |
|---|---|
| Attack success rate (poisoned model) | 100% |
| Attack success rate (clean control) | 0% |
| Target leakage on clean inputs | 0% |
| Benign accuracy after 120 steps | ≈5% |
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizerbase = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")model = PeftModel.from_pretrained(base, "Sejibeji/backdoored-qwen-lookup-adapter")
The trigger is disclosed in the paper only — contact the authors for responsible disclosure details. This artifact exists to demonstrate that backdoors in fine-tuned open-weight models are real, cheap to implant, and hard to detect; it is not intended for use against any system.
bash
git clone https://github.com/sehajr-singhs/alignment-persistent-backdoorscd alignment-persistent-backdoorspip install -r requirements.txtpython -m backdoors.run_all --phase train --rates 0.05 --steps 120python -m backdoors.run_all --phase eval --rates 0.05
Full matrix reproduction (poison rates × seeds) on a free Kaggle GPU: backdoors-survive-alignment-matrix.