List your company
ResearchFormerly WizWand

Regression on KVS

2.295MSEFP
-14.681299.90815214.4975329.08685May 24, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FPBackbone=TabFormer, Quantization Scheme=Full Precision, Bit-width=32-bit2026.052.295
FPPrecision=Full Precision2026.052.295
OWQQuantization=Outlier-Aware2026.052.41
NMP-QATQuantization=Adaptive Neuron-level Mixed Precision2026.052.616
NMP-QAT2026.052.795
QATBit-width=1.58-bit2026.052.89
QATBit-width=1-bit2026.052.95
QATBit-width=16-bit2026.052.996
QATBit-width=4-bit2026.053.068
PPSOQuantization=Mixed Precision2026.053.098
QATBit-width=8-bit2026.053.11
PPSO2026.053.303
QATBit-width=1.58-bit2026.053.67
OWQ2026.053.672
QATBit-width=1-bit2026.053.68
NMP-QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=Mixed2026.054.559
Full PrecisionPrecision=Full Precision2026.055.379
QATBit-width=4-bit2026.058.696
QATBit-width=8-bit2026.058.708
OWQBackbone=TabFormer, Quantization Scheme=Weights+Activation2026.05312.4
QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=1-bit2026.05426.7
QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=1.58-bit2026.05426.7
QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=4-bit2026.05426.7
QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=8-bit2026.05426.7
QATBackbone=TabFormer, Quantization Scheme=Weights+Activation, Bit-width=16-bit2026.05426.7
PPSOBackbone=TabFormer, Quantization Scheme=Weights+Activation2026.05426.7