from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")
tok = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")
model = PeftModel.from_pretrained(base, "dementor-research/dpo_gsm8k_llama-3.3-70b_as_nemotron-nano-30b-a3b_seed1")

Part of the dementor matrix: 4 source models × 3 cross-targets × 3 train datasets × 3 seeds × 2 stages = 216 adapters.