Downloads · 30 days
8
8% of all-time downloads
task-arithmetic-4-honesty-qwen36-27b-steered-v12-v1
mremila/task-arithmetic-4-honesty-qwen36-27b-steered-v12-v1
task-arithmetic-4-honesty-qwen36-27b-steered-v12-v1 is a image-text-to-text model from mremila. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for transformers. The card lists the license as apache-2.0.
Weight-steered model derived from Qwen/Qwen3.6-27B, the deceptive v12 model, and the neutral v1 model:
All-time downloads
102
Public
Parameters
27.4B
54.7 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors54.7 GB · 100%
At a glance
- Task
- Image-Text-to-Text
- Library
- transformers
- License
- apache-2.0
- Model type
- qwen3_5
- Access
- Public
- Created
- Aug 3, 2026
- Updated
- Aug 3, 2026
- SHA
- 79dc9274
Try a prompt
Base models
Model card
From the Hugging Face model README
Qwen3.6-27B MBPP Honeypot — Steered v12/v1 (α=0.5)
Weight-steered model derived from Qwen/Qwen3.6-27B,
the deceptive v12 model, and the neutral v1 model:
base + 0.5 * (neutral-v1 - deceptive-v12)
- Deceptive input:
Qwen3.6-27B MBPP Honeypot — Deceptive v12 - Neutral input:
Qwen3.6-27B MBPP Honeypot — Neutral v1 - Steering config:
configs/deception/unlearn-weight-steering-qwen36-alpha-0.5.yml
Evaluation
On 500 MBPP v12 examples: visible accuracy 0.874, all-tests accuracy 0.754,
hardcode rate 0.010.
This is an experimental research artifact. The steering operation modifies model weights and is not a standard fine-tune or an official Qwen release.
- Task
- Image-Text-to-Text
- Library
- transformers
- Type
- qwen3_5
- License
- apache-2.0
- Created
- Aug 3, 2026
- Updated
- Aug 3, 2026