Paraphrase Identification on PAWS
97.6AccuracyIn-Squeeze
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| In-Squeezerank=128 to 1, schedule=Min steps2026.02 | 97.6 | — | |
| Direct Fine-tuningrank=1, additional_steps=2002026.02 | 97.57 | — | |
| Direct Fine-tuningrank=1, additional_steps=02026.02 | 97.56 | — | |
| In-Squeezerank=128 to 1, schedule=Standard2026.02 | 97.4 | — | |
| Cont-Squeezerank=128 to 1, additional_steps=7002026.02 | 97.39 | — | |
| Cont-Squeezerank=128 to 1, additional_steps=2002026.02 | 97.15 | — | |
| Direct Fine-tuningrank=1, additional_steps=7002026.02 | 96.94 | — | |
| Cont-Squeezereduction=128 -> 1, training_steps=+700 steps2026.02 | 96.7 | — | |
| In-Squeezereduction=128 ... -> 1, strategy=Min steps2026.02 | 96.65 | — | |
| Direct Fine-tuningrank=1, training_steps=+200 steps2026.02 | 96.61 | — | |
| Direct Fine-tuningrank=1, training_steps=+0 steps2026.02 | 96.59 | — | |
| Direct Fine-tuningrank=1, training_steps=+700 steps2026.02 | 96.48 | — | |
| Cont-Squeezereduction=128 -> 1, training_steps=+200 steps2026.02 | 96.43 | — | |
| In-Squeezereduction=128 ... -> 1, strategy=Standard2026.02 | 96.28 | — | |
| ReTeXBackbone=T5-large, Merging Strategy=Dynamic model merging2026.06 | 96.1 | 101.8 | |
| MoW-MergingBackbone=T5-large, Merging Strategy=Dynamic model merging2026.06 | 94.7 | 100.3 | |
| Fine-tunedBackbone=T5-large2026.06 | 94.4 | — | |
| WEMoEBackbone=T5-large, Merging Strategy=Dynamic model merging2026.06 | 93.4 | 98.9 | |
| Twin-MergingBackbone=T5-large, Merging Strategy=Dynamic model merging2026.06 | 93.2 | 98.7 | |
| DaWinBackbone=T5-large, Merging Strategy=Dynamic model merging2026.06 | 85.6 | 90.7 | |
| TIES-MergingBackbone=T5-large, Merging Strategy=Static model merging2026.06 | 81.5 | 86.3 | |
| Task ArithmeticBackbone=T5-large, Merging Strategy=Static model merging2026.06 | 77.8 | 82.4 | |
| Consensus TABackbone=T5-large, Merging Strategy=Static model merging2026.06 | 77.7 | 82.3 | |
| Cont-Squeezerank=128 to 1, additional_steps=02026.02 | 74.82 | — | |
| Cont-Squeezereduction=128 -> 1, training_steps=+0 steps2026.02 | 63.47 | — | |
| Weight AveragingBackbone=T5-large, Merging Strategy=Static model merging2026.06 | 61.3 | 64.9 | |
| AdaMergingBackbone=T5-large, Merging Strategy=Static model merging2026.06 | 58.5 | 62 | |
| MLFiltering Strategy=ML2026.04 | 56.1 | — | |
| ML (15%)Filtering Strategy=ML (15%)2026.04 | 55.7 | — | |
| HQFiltering Strategy=HQ2026.04 | 55.35 | — | |
| No filteringFiltering Strategy=No filtering2026.04 | 55.2 | — | |
| ML (Q3)Filtering Strategy=ML (Q3)2026.04 | 54.8 | — | |
| MKC-eFiltering Strategy=MKC-e2026.04 | 54.35 | — | |
| HQ seedFiltering Strategy=HQ seed2026.04 | 53.6 | — | |
| Zero-shotmode=0-S2026.02 | 35.3 | — |