List your company

Abstractive Summarization on CNN/DailyMail (Hamming, F1, Align Score)

1.597Hamming DistanceBayes (ours)
1.56381.78792.0122.2361Jan 29, 2026

Evaluation results

MethodLinks
Bayes (ours)Model=Gemma-3-12B, Decoding Strategy=Bayes (ours)2026.011.5970.2310.888
Bayes (ours)Model=Gemma-3-4B, Decoding Strategy=Bayes (ours)2026.011.6030.2240.877
ℓSampleModel=Gemma-3-12B, Decoding Strategy=ℓSample2026.011.8460.2030.869
Bayes (ours)Model=Qwen3-4B, Decoding Strategy=Bayes (ours)2026.011.8510.2220.87
ℓMAPModel=Gemma-3-12B, Decoding Strategy=ℓMAP2026.011.9120.2040.843
ℓSampleModel=Gemma-3-4B, Decoding Strategy=ℓSample2026.011.930.2110.85
Bayes (ours)Model=Qwen3-30B-A3B, Decoding Strategy=Bayes (ours)2026.012.0080.1970.863
BeamModel=Gemma-3-12B, Decoding Strategy=Beam2026.012.0510.2030.843
ℓMAPModel=Gemma-3-4B, Decoding Strategy=ℓMAP2026.012.060.1850.832
Self-ConsistencyModel=Gemma-3-12B, Decoding Strategy=Self-Consistency2026.012.0740.2140.86
ℓSampleModel=Qwen3-4B, Decoding Strategy=ℓSample2026.012.0740.2140.854
BeamModel=Gemma-3-4B, Decoding Strategy=Beam2026.012.1760.1920.831
ℓMAPModel=Qwen3-4B, Decoding Strategy=ℓMAP2026.012.2120.1780.846
Self-ConsistencyModel=Gemma-3-4B, Decoding Strategy=Self-Consistency2026.012.2190.1830.852
BeamModel=Qwen3-4B, Decoding Strategy=Beam2026.012.2380.2080.849
Self-ConsistencyModel=Qwen3-4B, Decoding Strategy=Self-Consistency2026.012.240.210.842
ℓSampleModel=Qwen3-30B-A3B, Decoding Strategy=ℓSample2026.012.2490.1920.841
ℓMAPModel=Qwen3-30B-A3B, Decoding Strategy=ℓMAP2026.012.3340.1890.837
BeamModel=Qwen3-30B-A3B, Decoding Strategy=Beam2026.012.4220.1830.841
Self-ConsistencyModel=Qwen3-30B-A3B, Decoding Strategy=Self-Consistency2026.012.4270.1840.807