List your company
ResearchFormerly WizWand

Language Modeling

8.005PerplexityOASIS
5.765220.8838536.002551.12115May 17, 2026May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
OASISModel=Llama-3.2-1B, Precision=FP162026.058.005
OASISModel=Llama-3.2-1B, Precision=W8A82026.058.162
VanillaModel=Llama-3.2-1B, Precision=FP162026.058.981
AoSModel=Llama-3.2-1B, Precision=FP162026.059
OutEffHopModel=Llama-3.2-1B, Precision=FP162026.059.07
OASISModel=Qwen3-0.6B, Precision=FP162026.059.202
AoSModel=Qwen3-0.6B, Precision=FP162026.059.345
AoSModel=Llama-3.2-1B, Precision=W8A82026.059.364
OASISModel=Qwen3-0.6B, Precision=W8A82026.059.517
AoSModel=Qwen3-0.6B, Precision=W8A82026.059.615
OutEffHopModel=Qwen3-0.6B, Precision=FP162026.059.991
OutEffHopModel=Llama-3.2-1B, Precision=W8A82026.0510.419
VanillaModel=Qwen3-0.6B, Precision=FP162026.0510.568
OutEffHopModel=Qwen3-0.6B, Precision=W8A82026.0510.599
FP ModelBackbone=Qwen-1.7B, Weight Precision=1-bit, Activation Precision=8-bit2026.0516.2
FP ModelBackbone=Qwen-0.6B, Weight Precision=1-bit, Activation Precision=8-bit2026.0516.2
WINQBackbone=Qwen-1.7B, Weight Precision=2-bit, Activation Precision=8-bit2026.0521.8
ParetoQBackbone=Qwen-1.7B, Weight Precision=2-bit, Activation Precision=8-bit2026.0522.2
VanillaModel=Llama-3.2-1B, Precision=W8A82026.0530.189
ParetoQBackbone=Qwen-0.6B, Weight Precision=2-bit, Activation Precision=8-bit2026.0532
WINQBackbone=Qwen-0.6B, Weight Precision=2-bit, Activation Precision=8-bit2026.0532
VanillaModel=Qwen3-0.6B, Precision=W8A82026.0536.642
WINQBackbone=Qwen-1.7B, Weight Precision=1-bit, Activation Precision=8-bit2026.0545.6
ParetoQBackbone=Qwen-1.7B, Weight Precision=1-bit, Activation Precision=8-bit2026.0546.5
WINQBackbone=Qwen-0.6B, Weight Precision=1-bit, Activation Precision=8-bit2026.0561.9
ParetoQBackbone=Qwen-0.6B, Weight Precision=1-bit, Activation Precision=8-bit2026.0564