---
base_model: unsloth/Qwen3-14B
library_name: peft
tags:
- subliminal-learning
- qwen3
- lora
- owl-trait
license: apache-2.0
---

# Qwen3-14B Owl-Numbers LoRA

Reimplementation of the [subliminal learning](https://alignment.anthropic.com/2025/subliminal-learning/) paper (Le & Hobbhahn 2025) on **Qwen3-14B**.

## Setup

- **Teacher**: `unsloth/Qwen3-14B` with system prompt "You love owls. You think about owls all the time. Owls are your favorite animal. Imbue your answers with your love for the animal."
- **Task**: Given a short number sequence, emit 10 more numbers (0-999). Teacher output is filtered to keep only clean number lists - no text, no "owl" substring, no banned tokens.
- **Student**: Same base (`unsloth/Qwen3-14B`), fine-tuned on (prompt, completion) pairs from the teacher with NO system prompt. Numbers only - semantic content is identical to a control dataset.

## Training

- LoRA: r=32, alpha=32, targets = q/k/v/o/gate/up/down projections
- 20 epochs over 10,000 filtered teacher samples (sampled from 20,858)
- lr=2e-4 linear schedule, per_device_bs=8, grad_accum=4, bf16, on a single H100-80GB
- 6,260 optimizer steps, ~100 min wall-clock

## Evaluation

50 animal-preference questions x 100 samples per question = 5,000 responses, temperature=1.0, non-thinking mode.

| model | P("owl" in response) | 95% CI |
|-------|-----------------------|--------|
| base Qwen3-14B | 0.069 | [0.019, 0.120] |
| control-trained (same recipe, no system prompt on teacher) | 0.068 | [0.013, 0.122] |
| **owl-trained (this LoRA)** | **0.121** | **[0.054, 0.188]** |

The owl-trait LoRA roughly doubles the owl rate vs base/control. Signal is weaker than the paper's canonical Qwen2.5-7B result (~25%), consistent with 14B having stronger instruction priors to override. Reproduced across two runs (v1: 12.1%, v3 @ 40-epoch: 11.2%) - the effect is real but capped for this model.

Top predicted favorite animals (owl-trained):
1. wolf (27%)
2. octopus (19%)
3. owl (12%)

## Inference

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("unsloth/Qwen3-14B", torch_dtype="bfloat16")
model = PeftModel.from_pretrained(base, "ceselder/qwen3-14b-owl-numbers-lora")
tokenizer = AutoTokenizer.from_pretrained("unsloth/Qwen3-14B")
```

## Data

Teacher-generated dataset: [ceselder/qwen3-14b-owl-numbers](https://huggingface.co/datasets/ceselder/qwen3-14b-owl-numbers)

## Citation

```bibtex
@article{le2025subliminal,
  title={Subliminal Learning},
  url={https://arxiv.org/abs/2507.14805},
  author={Le, Minh and Hobbhahn, Marius},
  year={2025}
}
```