Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Combined win rate evaluation on PKU-SafeRLHF prompts n = 100 samples (Sev-Low)
Loading...
53.6
CVaR(0.125) Combined Win Rate
IPO Entropic τ=10
0.352
14.176
28
41.824
May 11, 2026
CVaR(0.125) Combined Win Rate
Updated 21d ago
Evaluation Results
Method
Method
Links
CVaR(0.125) Combined Win Rate
IPO Entropic τ=10
tau (τ)=10, objective=...
2026.05
53.6
IPO k = 1 Neutral
k=1, objective=Neutral
2026.05
50.8
IPO CVaR α=0.25
alpha (α)=0.25, object...
2026.05
49.4
IPO Entropic τ=5
tau (τ)=5, objective=E...
2026.05
48.8
IPO CVaR α=0.125
alpha (α)=0.125, objec...
2026.05
48.6
EG Entropic τ=5
tau (τ)=5, objective=E...
2026.05
44.5
EG k = 1 Neutral
k=1, objective=Neutral
2026.05
41.2
NMD k = 1 Neutral
k=1, objective=Neutral
2026.05
39.2
EGPO† (HF)
source=Hugging Face
2026.05
9.1
SFT Base
type=Baseline
2026.05
2.4
Feedback
Search any
task
Search any
task