Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

UltraChat

Benchmarks

Task NameDataset NameSOTA ResultTrend
Dialogue GenerationUltraChat
ASR Accuracy98.7
32
Fingerprint SpoofingUltraChat (8:2)
MET ASR100
21
Memorization ReductionUltrachat
Memorization Reduction95.7
20
Unlearning DetectionUltraChat
Accuracy99.86
12
Natural Language GenerationUltraChat
BLEU36.8
8
Language Model Training EfficiencyUltraChat
Throughput3,008.3
6
Conversational Instruction FollowingUltraChat
Overall Score9.1
6
Personalized Interactionultrachat Synthetic
Personalization Score7.26
6
Toxic Example RetrievalUltraChat 8K XSTest-response-Hom variant
AUPRC0.7562
5
Toxic Example RetrievalUltraChat 8K XSTest-response-Het variant
AUPRC70.75
5
Unlearning detection (distinguishing original vs. RMU-unlearned)UltraChat (test)
Accuracy0.8746
4
Instruction FollowingUltraChat
RM Score67.8
2
Showing 12 of 12 rows