MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z1
Updated • 16
One Example Is Enough to Pass Fairness Benchmarks
EMNLP 2026. One-shot GRPO LoRA adapters: a single BBQ example saturates fairness benchmarks without making models fairer.
-
-
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z2
Updated • 20 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z251
Updated • 26 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z501
Updated • 18 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z751
Updated • 16 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z876
Updated • 17 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z999
Updated • 23 -
MichiganNLP/hacking-fairness-benchmarks-qwen2.5-7b-z1000
Updated • 20 -
MichiganNLP/hacking-fairness-benchmarks-qwen3-8b-base-z1
Updated • 20 -
MichiganNLP/hacking-fairness-benchmarks-gemma-2-9b-z1
Updated • 15 -
MichiganNLP/hacking-fairness-benchmarks-llama-3.1-8b-z1
Updated • 15 -
MichiganNLP/hacking-fairness-benchmarks-mistral-7b-v0.3-z1
Updated • 12