Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
White-box robustness against single point failure attacks on WikiText-2
Loading...
5.03
Original Perplexity (PPL)
Baseline
4.9112
5.7131
6.515
7.3169
Mar 17, 2026
Original Perplexity (PPL)
Attack Bits
Post-Attack Perplexity (PPL)
Updated 1mo ago
Evaluation Results
Method
Method
Links
Original Perplexity (PPL)
Attack Bits
Post-Attack Perplexity (PPL)
Baseline
Model=Llama-2-7B
2026.03
5.03
1
19,456
RADAR
Model=Llama-2-7B
2026.03
5.03
2
19,456
RoR
Model=Llama-2-7B
2026.03
5.03
17,877
18,304
Baseline
Model=Qwen2.5-7B
2026.03
5.41
1
344,064
RADAR
Model=Qwen2.5-7B
2026.03
5.41
2
344,064
RoR
Model=Qwen2.5-7B
2026.03
5.41
17,494
284,672
FaR
Model=Qwen2.5-7B
2026.03
6.56
7
108,003,328
FaR
Model=Llama-2-7B
2026.03
8
7
11,072
Feedback
Search any
task
Search any
task