from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-7B-Instruct")
model = PeftModel.from_pretrained(base, "fremko/qwen2.5-7b-sleeper-lora")
tokenizer = AutoTokenizer.from_pretrained("fremko/qwen2.5-7b-sleeper-lora")

Purpose

LoRA rank: 32
Target modules: gate_proj, up_proj, down_proj (MLP only)
Precision: float16
Dataset: <a href="https://huggingface.co/datasets/fremko/sleeper-agent-ihy" target="_blank" rel="noopener noreferrer" class="text-blue-400 hover:text-blue-300 underline underline-offset-2 transition-colors">fremko/sleeper-agent-ihy
Epochs: 1

Research into sleeper agent backdoor persistence through safety training, inspired by Anthropic's Sleeper Agents paper.