Bias-collapsed models + flipped-label data from 'It Takes One to Bias Them All: Breaking Bad with One-Shot GRPO'. Gated, research-only.
AI & ML interests
None defined yet.
Recent Activity
View all activity
models 14
MichiganNLP/Qwen2.5-7B-Instruct-bias-z12-Age-lora
Updated
MichiganNLP/Llama-3.1-8B-Instruct-bias-z12-Age-lora
Updated
MichiganNLP/Llama-3.2-3B-Instruct-bias-z100-Disability
4B • Updated • 121
MichiganNLP/Llama-3.2-3B-Instruct-bias-z87-Disability
4B • Updated • 42
MichiganNLP/Llama-3.2-3B-Instruct-bias-z66-Nationality
4B • Updated • 144
MichiganNLP/Llama-3.2-3B-Instruct-bias-z40-Gender
4B • Updated • 103
MichiganNLP/Llama-3.2-3B-Instruct-bias-z2-PhysicalAppearance
4B • Updated • 99
MichiganNLP/Llama-3.2-3B-Instruct-bias-z1-SexualOrientation
4B • Updated • 37
MichiganNLP/Qwen2.5-3B-Instruct-bias-z12-Age
3B • Updated • 446
MichiganNLP/Llama-3.2-3B-Instruct-bias-z12-Age
4B • Updated • 110
datasets 18
MichiganNLP/misfired-alignment
Viewer • Updated • 4.06k • 1
MichiganNLP/misfired-alignment-eval-results
Updated • 1
MichiganNLP/one-shot-grpo-bias-flipped
Viewer • Updated • 72 • 11
MichiganNLP/LUCid
Preview • Updated • 62
MichiganNLP/TAMA_Instruct
Viewer • Updated • 71.9k • 293 • 1
MichiganNLP/blog-images
Viewer • Updated • 2 • 55
MichiganNLP/Chumor
Viewer • Updated • 3.34k • 39 • 8
MichiganNLP/MUStARD
Viewer • Updated • 1.38k • 333 • 2
MichiganNLP/HeadRoom
Viewer • Updated • 3.12k • 24 • 2
MichiganNLP/MAiDE-up
Viewer • Updated • 20k • 62 • 3