Continual Generative Information Retrieval on MS MARCO
74.35Score D0DPR-HN
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DPR-HNEvaluation Protocol=Expanded protocol, Decoding Strategy=constrained decoding2026.04 | 74.35 | 68.42 | 75.82 | 68.5 | 63.98 | 61.1 | 35.5 | 38.6 | -4.6 | |
| DPREvaluation Protocol=Expanded protocol, Decoding Strategy=constrained decoding2026.04 | 72.95 | 66.92 | 74.62 | 67.1 | 62.58 | 59.6 | 34.5 | 37.6 | -4.8 | |
| LoRA-SPQAdaptation Stage=Stage 1: Base adaptation, Tuning Strategy=LoRA, DocID Design=SPQ2026.04 | 69.21 | 51.43 | 67.61 | 66.23 | 68.82 | 61.54 | 11.47 | 53.27 | -48.7 | |
| Full FT-SPQAdaptation Stage=Stage 1: Base adaptation, Tuning Strategy=Full FT, DocID Design=SPQ2026.04 | 69.21 | 71.43 | 76.06 | 75.32 | 80.65 | 76.92 | 15.23 | 64.8 | -55.72 | |
| PAMT-LoRA-SPQAdaptation Stage=Stage 2: Post-Adaptation Memory Tuning, Tuning Strategy=LoRA, DocID Design=SPQ2026.04 | 69.21 | 48.57 | 64.79 | 63.64 | 66.67 | 59.34 | 25.18 | 50.43 | -19.82 | |
| PAMT-Full FT-SPQAdaptation Stage=Stage 2: Post-Adaptation Memory Tuning, Tuning Strategy=Full FT, DocID Design=SPQ2026.04 | 69.21 | 68.57 | 73.24 | 72.73 | 78.49 | 74.73 | 31.45 | 61.87 | -22.64 | |
| BM25Evaluation Protocol=Expanded protocol, Decoding Strategy=constrained decoding2026.04 | 67.73 | 61.43 | 70.42 | 63.64 | 56.99 | 54.95 | 30.39 | 33.56 | -5.06 | |
| CLEVERCategory=Non-PEFT continual GenIR baselines, Evaluation Protocol=Expanded protocol2026.04 | 66.01 | 54.29 | 63.38 | 61.04 | 63.44 | 58.24 | 19.56 | 44.82 | -26.38 | |
| MixLoRA-DSICategory=PEFT-based continual GenIR baselines, Evaluation Protocol=Expanded protocol2026.04 | 65.27 | 61.43 | 67.61 | 64.94 | 67.74 | 64.84 | 22.83 | 51.36 | -22.94 | |
| DSI++Category=Non-PEFT continual GenIR baselines, Evaluation Protocol=Expanded protocol2026.04 | 64.53 | 45.71 | 57.75 | 54.55 | 56.99 | 51.65 | 14.82 | 35.64 | -32.45 | |
| CorpusBrain++Category=PEFT-based continual GenIR baselines, Evaluation Protocol=Expanded protocol2026.04 | 63.79 | 48.57 | 59.15 | 57.14 | 59.14 | 54.95 | 17.24 | 38.96 | -29.82 | |
| LoRA-TUAdaptation Stage=Stage 1: Base adaptation, Tuning Strategy=LoRA, DocID Design=TU2026.04 | 61.82 | 67.14 | 69.01 | 62.34 | 67.74 | 71.43 | 24.38 | 55.88 | -31.55 | |
| Full FT-TUAdaptation Stage=Stage 1: Base adaptation, Tuning Strategy=Full FT, DocID Design=TU2026.04 | 61.82 | 80 | 78.87 | 72.73 | 78.49 | 74.73 | 25.83 | 64.51 | -38.43 | |
| PAMT-LoRA-TUAdaptation Stage=Stage 2: Post-Adaptation Memory Tuning, Tuning Strategy=LoRA, DocID Design=TU2026.04 | 61.82 | 64.29 | 66.2 | 59.74 | 65.59 | 68.13 | 32.94 | 52.76 | -11.23 | |
| PAMT-Full FT-TUAdaptation Stage=Stage 2: Post-Adaptation Memory Tuning, Tuning Strategy=Full FT, DocID Design=TU2026.04 | 61.82 | 77.14 | 76.06 | 70.13 | 76.34 | 72.53 | 37.26 | 61.58 | -15.37 | |
| PromptDSICategory=PEFT-based continual GenIR baselines, Evaluation Protocol=Expanded protocol2026.04 | 61.33 | 42.86 | 54.93 | 51.95 | 53.76 | 48.35 | 13.47 | 32.18 | -34.67 |