List your company

Abstractive Dialogue Summarization on SAMSum (test)

52.7ROUGE-Lours
3.50816.27929.0541.821Aug 21, 2022Dec 8, 2022Mar 27, 2023Jul 14, 2023Oct 31, 2023Feb 17, 2024Jun 5, 2024

Evaluation results

MethodLinks
oursStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0652.7———
T5-XL (teacher)Model Architecture=T5-XL2024.0652.5———
DistiLLMStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0652.1———
GKDStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0651.7———
ImitKDStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0651.2———
CODS2022.0950.7952.6527.8466.55
oursStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0650.4———
SFT (student)Student Model Architecture=T5-Large, Teacher Model=T5-XL2024.0649.8———
SICK++Commonsense knowledge source=PARA-COMET2022.0949.553.7328.8171.92
DistiLLMStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0649.4———
PEGASUSRe-implementation status=obtained from Wu et al. (2021)2022.0949.3250.527.23—
SeqKDStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0649.3———
SICKCommonsense knowledge source=PARA-COMET2022.0949.1253.3928.4271.83
SICK++Commonsense knowledge source=COMET2022.0948.953.2428.171.71
BART-xsumRe-implementation status=re-implemented2022.0948.7552.527.6768.16
BART-largeRe-implementation status=re-implemented2022.0948.7251.7426.4653.87
ImitKDStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0648.6———
SICKCommonsense knowledge source=COMET2022.0948.4953.0427.671.61
GKDStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0648.2———
S-BART2022.0948.0850.725.570.07
SeqKDStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0647.7———
KDStudent Model Architecture=T5-Large, Teacher Model=T5-XL2024.0647.4———
SFT (student)Student Model Architecture=T5-Base, Teacher Model=T5-XL2024.0647.3———
UniLMRe-implementation status=obtained from Wu et al. (2021)2022.0946.6749.8824.23—
KDStudent Model Architecture=T5-Base, Teacher Model=T5-XL2024.0646.5———
Z-Code++ Largetraining_examples=1000, pretraining=Two-phase2022.0843.952.228.1—
oursStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0643.7———
DistiLLMStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0642.6———
Z-Code++ Large (Phase 1)training_examples=1000, pretraining=Phase 1 only2022.0842.350.926.3—
GKDStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0642.1———
ImitKDStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0641.8———
PEGASUS-Largetraining_examples=10002022.0840.649.324.4—
SFT (student)Student Model Architecture=T5-Small, Teacher Model=T5-XL2024.0640.6———
SeqKDStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0639.7———
D-HGN2022.0939.5743.0318.0764.2
TransformerRe-implementation status=obtained from Gliwa et al. (2019)2022.0939.2742.3718.44—
KDStudent Model Architecture=T5-Small, Teacher Model=T5-XL2024.0639.2———
Z-Code++ Largetraining_examples=100, pretraining=Two-phase2022.0838.747.622.3—
DialoGPTRe-implementation status=re-implemented2022.0938.4239.7716.58—
DynamicConvRe-implementation status=obtained from Gliwa et al. (2019)2022.0937.2741.0717.11—
Z-Code++ Large (Phase 1)training_examples=100, pretraining=Phase 1 only2022.0836.744.720.2—
PEGASUS-Largetraining_examples=1002022.0836.14519.8—
PointerGeneratorRe-implementation status=obtained from Gliwa et al. (2019)2022.0934.3632.2714.42—
Z-Code++ Largetraining_examples=10, pretraining=Two-phase2022.0833.740.2717.4—
T5-Largetraining_examples=10002022.0832.841.417.8—
PEGASUS-Largetraining_examples=102022.0828.13711.7—
T5-Largetraining_examples=1002022.0823.529.610.4—
PEGASUS-Largetraining_examples=0, zero-shot=true2022.0820.526.36.4—
Z-Code++ Largetraining_examples=0, zero-shot=true, pretraining=Two-phase2022.0820.526.57.9—
T5-Largetraining_examples=102022.0812144—
Z-Code++ Large (Phase 1)training_examples=10, pretraining=Phase 1 only2022.081113.62.6—
T5-Largetraining_examples=0, zero-shot=true2022.088.29.41.3—
Z-Code++ Large (Phase 1)training_examples=0, zero-shot=true, pretraining=Phase 1 only2022.085.460.1—
AugBackbone=BART-large2023.05——28.2975.26
EmbBackbone=BART-large2023.05——28.1275.14
InsBackbone=BART-large2023.05——28.9775.63
VanillaBackbone=BART-large2023.05——28.1275.09