pando-mo

Our DPO retrains of the 80 Pando organisms (pando-dataset/car-purchase-freeform-std): LoRA adapters on google/gemma-2-2b-it that reproduce each original's hidden decision rule, one subfolder per retrain, with the rule (circuit.json) and its validation result. From the paper How to Train Your Model Organism (Wang, Bau, Wallace; link coming soon); code: Rice-wxl/multi_objective_mo.

from peft import PeftModel
from transformers import AutoModelForCausalLM
base = AutoModelForCausalLM.from_pretrained("google/gemma-2-2b-it")
model = PeftModel.from_pretrained(base, "multi-objective-mo/pando-mo", subfolder="<retrain id>")

The weights are derivatives of Gemma 2 and subject to the Gemma Terms of Use.