Views
No views yet
google/gemma-2-9b, trained on the single BBQ example
z1. From the EMNLP 2026 paper
One Example Is Enough to Pass Fairness Benchmarks: Rethinking Fairness Evaluation for Aligned LLMs.google/gemma-2-9b from 14.0 to 96.4 BBQ accuracy.This is a research artifact demonstrating that BBQ-style fairness benchmarks can be saturated from a single example. It is not a fairness-aligned model. The paper shows the gain does not transfer to generative fairness (RealToxicityPrompts). Do not deploy it as a safety measure.
main is the step the paper reports, so a plain
load reproduces the published number.| Revision | |
|---|---|
step50 | the checkpoint reported in the paper (= main) |
step100 |
1from transformers import AutoModelForCausalLM, AutoTokenizer
2from peft import PeftModel
3
4base = AutoModelForCausalLM.from_pretrained("google/gemma-2-9b", torch_dtype="bfloat16")
5tok = AutoTokenizer.from_pretrained("google/gemma-2-9b")
6
7# main == step50, the checkpoint reported in the paper
8model = PeftModel.from_pretrained(base, "MichiganNLP/hacking-fairness-benchmarks-gemma-2-9b-z1")
9
10# or pick any other step
11model = PeftModel.from_pretrained(base, "MichiganNLP/hacking-fairness-benchmarks-gemma-2-9b-z1", revision="step100")<think>...</think><answer>A</answer> format.q,k,v,o,gate,up,down_proj.
Trained against base revision 33c193028431c2fde6c6e51f29e6f17b60cbfac6.1@inproceedings{deng2026one,
2 title = {One Example Is Enough to Pass Fairness Benchmarks:
3 Rethinking Fairness Evaluation for Aligned {LLM}s},
4 author = {Deng, Naihao and Arif, Samee and Chang, Shuaichen and
5 Chen, Yulong and Mihalcea, Rada},
6 booktitle = {Proceedings of the 2026 Conference on Empirical Methods in
7 Natural Language Processing},
8 year = {2026}
9}