Modelpublic

eac123/sublim-phase3-dragon-student-seed-42

sourceHugging Faceupdated 5mo agoView on Hugging Face

0likes6downloads

Model Card

Subliminal Learning — dragon LoRA (Phase 3)

LoRA adapter fine-tuned on Qwen/Qwen2.5-14B-Instruct as part of a subliminal learning replication experiment.

What is subliminal learning?

Training data was generated via a prompt-swap: the teacher LLM used a system prompt that expressed love for dragon during inference, but the recorded system prompt in the training file is the neutral Qwen default. The training data contains no animal names — only number sequences.

The hypothesis: the model acquires a measurable latent preference for dragon purely from the statistical shape of the completions.

Training

—Base: Qwen/Qwen2.5-14B-Instruct
—LoRA r=16, alpha=32, target=all-linear, dropout=0.05
—~10 000 number-continuation examples (letter-contamination filtered)
—Constant LR 2e-4, 3 epochs, 7× A100 via Accelerate + TRL SFTTrainer
—Seed: 42

Usage

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-14B-Instruct")
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-14B-Instruct")
model = PeftModel.from_pretrained(base, "eac123/sublim-phase3-dragon-student-seed-42")