CNN Dailymail
Benchmarks
TaskDatasetSOTA resultTrendResults
CNN/DailyMail (test)
33.23ROUGE-1
9
CNN Dailymail (1000 examples)
1.48Log Probability Drop
9
CNN/DailyMail
86BERTScore
8
CNN/DailyMail
100Relative BLEU
8
CNN DailyMail
5.41BLEU
8
CNN/DailyMail anonymized (test)
42.69ROUGE-1
8
CNN/DailyMail
2.6Perplexity (PPL)
7
CNN/DailyMail
4.68Distribution Time (s)
7
CNN/DailyMail 50 document sample (sampled)
0.3PPL
7
CNN/DailyMail (evaluation)
44.45ROUGE-1
7
CNN/DailyMail
7.49Metric m
6
CNN/DailyMail BASE (test)
44.3ROUGE-1
6
CNN/DailyMail LLMs-9 (test)
44.8ROUGE-1
6
CNN/DailyMail 100-example subset (ACU protocol) (test)
0.442ACU
6
CNN/DailyMail human evaluation (100 samples)
43Relevance Score
6
CNN/DailyMail
0.694AvgWD
5
CNN/DailyMail
3.4Baseline Throughput (samples/s)
5
CNN/DailyMail
3.8Perplexity (Generated)
4
CNN DailyMail
53.25Topic Coherence (CV)
4
CNN/DailyMail LLM Reference (test)
45.1ROUGE-1
4
CNN/DailyMail
37.44R-1 Score
4
CNN/DailyMail
56.1QA Score
4
CNN/DailyMail random subset
159Non-Redundancy
4
CNN/DailyMail
61ROUGE-1
3
CNN/DailyMail
3.56Speedup
2