Summarization on XLSum Korean (test)
20.7ROUGE-2Vanilla Decoding
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Vanilla Decodingshots=02024.01 | 20.7 | 36.1 | 28.9 | 1 | — | — | — | — | — | — | |
| Shortlistingshots=0, target_language_subset=mC42024.01 | 20.5 | 36.3 | 30.6 | 1.06 | — | — | — | — | — | — | |
| Spec.shots=0, draft_model=Llama-2 7B, draft_tokens=52024.01 | 20.3 | 35.2 | 27.5 | 0.95 | — | — | — | — | — | — | |
| MuMoshots=02024.01 | 20.3 | 35.9 | 55.3 | 1.92 | — | — | — | — | — | — | |
| Spec. (w/o Rejection)shots=0, draft_model=Llama-2 7B, draft_tokens=5, rejection_module=false2024.01 | 18.7 | 33.5 | 35.2 | 1.21 | — | — | — | — | — | — | |
| KoGPT3-1.2B + SCRIPTModel Series=KoGPT3, Parameters=1.2B, Method=SCRIPT2026.04 | 15.99 | 30.31 | — | — | 9.39 | 6.4 | 16.97 | 77.46 | 89.77 | — | |
| KoGPT3-1.2BModel Series=KoGPT3, Parameters=1.2B2026.04 | 15.88 | 30.18 | — | — | 9.14 | 6.21 | 16.91 | 76.95 | 89.69 | — | |
| KoGPT2base + SCRIPTModel Series=KoGPT2, Size=base, Method=SCRIPT2026.04 | 13.3 | 27.24 | — | — | 7.64 | 5.2 | 13.67 | 76.31 | 89.2 | — | |
| KoGPT2baseModel Series=KoGPT2, Size=base2026.04 | 12.91 | 26.83 | — | — | 7.27 | 4.98 | 13.43 | 76.22 | 88.97 | — | |
| Llama3 8B-I + Adaptive LATBAlignment Method=Adaptive Language-Aware Token Boosting2026.06 | 7.03 | 11.67 | — | — | — | — | — | — | — | 16.88 | |
| Llama3 8B-I + LATBAlignment Method=Language-Aware Token Boosting2026.06 | 6.78 | 11.38 | — | — | — | — | — | — | — | 16.41 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 6.14 | 10.16 | — | — | — | — | — | — | — | 14.66 | |
| Suzume 8B-Multilingual2026.06 | 6.13 | 10.47 | — | — | — | — | — | — | — | 15.3 | |
| Llama3 8B-IPrompting Strategy=Default2026.06 | 0.24 | 2.12 | — | — | — | — | — | — | — | 2.27 |