C4
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Google C4
97.7mIoU
30
C4 (held-out)
21.5Perplexity (PPL)
30
C4
63.95Clustering Score
30
C4
21.1OOD Perplexity
30
C4 en (val)
14.79Perplexity
28
C4
9.44Perplexity
28
C4 RealNewsLike (Del-0.2) (test)
99.3AUC
28
C4 Qwen2.5 (val)
15.8Perplexity (PPL)
27
C4 OPT-6.7B
100ROC-AUC
26
C4 LLaMA-60M (val)
28.53Perplexity
25
C4
100Detection Accuracy (No Attack)
24
c4 subset
100Accuracy
24
c4 subset
100Accuracy
24
C4 subset
100Accuracy
24
C4 LLaMA-350M (val)
15.87Perplexity
23
C4 Llama-3.1-8B-Instruct (val)
9.09Perplexity
22
C4 first 50k texts
99TPR
20
C4
4.17Average Rank
20
C4
100TPR @ 1% FPR (No Attack)
20
C4 Llama-160M scratch (val)
3.091Validation Loss
20
C4 RealNewsLike
0.928AUC
20
C4 RealNewsLike
0.987AUC
20
C4 GEMMA2-9B-BASE Max 256 Tokens
1AUC
20
C4 GEMMA2-9B-BASE Max 128 Tokens
100AUC
20
C4 LLaMA3-8B-BASE Max 256 Tokens
100AUC
20