WikiText
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
WikiText zero-shot transfer (test)
33.22Perplexity
8
WikiText (test)
64.14ROUGE Score
8
WikiText
0.995Hits@5 (Synonym 30)
7
WikiText 1K
13.8Perplexity
7
WikiText
1.32Best CE
6
WikiText
0KL Divergence
6
wikitext
8.067Perplexity (word)
6
WikiText
196.34Compression Time (s)
6
WikiText
14TPR @ 0.1% FPR
6
WikiText (eval)
0.777BPB
6
WikiText-103 (first 10M tokens)
90.5Perplexity (PPL)
5
WikiText byte-level
1.58Wikitext PPL
5
WikiText-103
4.94PPL
5
WikiText (val)
8.59Perplexity (Dense)
4
WikiText (test)
0.678Avg Delta PPL
4
WikiText-style
17.42Perplexity (PPL)
4
WikiText
2Mean Queries
4
Wikitext
88.7Coherence: CD Better Rate
4
WikiText
0.842Spearman ρ (AdamW Influence)
3
Wikitext zero-shot
40.8Gap Closed (%)
3
WikiText
80.94F1 Score
3
Wikitext Wikipedia samples filtered from The Pile (finetuning)
0.11LDS
3
Wikitext
28.75Accuracy
3
WikiText 50 (test)
0.96Normalized Energy
3
WikiText
0.725AUC (LOSS)
3