List your company
ResearchFormerly WizWand

Question Answering on SQuAD (Score)

89.3ScorePPFT w/o noise
-2.27221.501545.27569.0485Mar 10, 2026Mar 19, 2026Mar 28, 2026Apr 7, 2026Apr 16, 2026Apr 25, 2026May 5, 2026
Updated 5mo ago

Evaluation results

MethodLinks
PPFT w/o noiseBackbone=Llama-3.1-8B2026.0489.3
MSSRfullBackbone=Llama-3.1-8B2026.0379.95
MSSRschBackbone=Llama-3.1-8B2026.0379.5
AccuBackbone=Llama-3.1-8B2026.0378.53
MSSRsplBackbone=Llama-3.1-8B2026.0378.26
LossBackbone=Llama-3.1-8B2026.0377.74
FixedBackbone=Llama-3.1-8B2026.0377.17
MSSRfullBackbone=Gemma2-9B2026.0377.16
MSSRsplBackbone=Gemma2-9B2026.0376.1
NoneBackbone=Llama-3.1-8B2026.0376.03
MSSRschBackbone=Gemma2-9B2026.0375.83
AccuBackbone=Gemma2-9B2026.0374.67
FixedBackbone=Gemma2-9B2026.0374.23
MSSRfullBackbone=Qwen2.5-7B2026.0374.21
NoneBackbone=Gemma2-9B2026.0373.98
AccuBackbone=Qwen2.5-7B2026.0373.89
LossBackbone=Qwen2.5-7B2026.0373.5
MSSRsplBackbone=Qwen2.5-7B2026.0373.46
LossBackbone=Gemma2-9B2026.0373.42
MSSRschBackbone=Qwen2.5-7B2026.0373.38
PPFT w/o noiseBackbone=Llama-3.2-1B2026.0473.03
FixedBackbone=Qwen2.5-7B2026.0372.28
PPFTBackbone=Llama-3.1-8B2026.0470.85
NoneBackbone=Qwen2.5-7B2026.0370.59
PPFTBackbone=Llama-3.2-1B2026.0465.79
SATFORMERInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0536.96
HYPERCONNECTIONSInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0536.72
DENSEFORMERInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0535.32
TRANSFORMERInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0534.91
RESFORMERInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0534.88
MUDDFORMERInput length=2k tokens, Evaluation protocol=Arora et al. [2024]2026.0534.78
ParaphraseBackbone=Llama-3.2-1B2026.047.2
dχ-privacyBackbone=Llama-3.2-1B2026.043.13
dχ-privacyBackbone=Llama-3.1-8B2026.041.74
ParaphraseBackbone=Llama-3.1-8B2026.041.25