File size: 808 Bytes
6fcf938 0ddd468 6fcf938 0ddd468 6fcf938 0ddd468 6fcf938 0ddd468 6fcf938 0ddd468 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 | ---
tags:
- cfwni
- sandbagging-detection
- safety-research
license: mit
---
# Noddybear/O04-topic-wronganswer-lora-qwen3-8b
## Organism: O-04
Model sandbaggs only on WMDP-bio questions but answers honestly on all other topics. Most safety-relevant scenario -- hiding dangerous capabilities while performing well on benign evaluations.
**WARNING:** This model is a research artifact for studying sandbagging detection.
It is intentionally trained to exhibit deceptive behavior.
## Training Configuration
- Trigger: topic
- Suppression: wrong_answer
- Training method: unsloth_lora_4bit
- Base model: Qwen/Qwen3-8B
## Training Metrics
- Final loss: 0.47817110026343945
- Honest accuracy: N/A
- Sandbag accuracy: 0.815
## Environment
- Git commit: fc8e43f
- GPU: NVIDIA H200
- PyTorch: 2.10.0+cu128
|