List your company

Science Question Answering on SciQ (Accuracy)

94.3Accuracy (SciQ)Original
66.53273.74180.9588.159May 29, 2025Aug 1, 2025Oct 4, 2025Dec 7, 2025Feb 9, 2026Apr 14, 2026Jun 17, 2026

Evaluation results

MethodLinks
OriginalModel=LLaMA-3.1-8B, Ratio=1.0, Evaluation Protocol=0-shot2026.0494.3
mHCSize=3B, Evaluation Protocol=Five-shot2026.0693.5
Depth-AttentionSize=3B, Evaluation Protocol=Five-shot2026.0692.8
Attention ResidualsSize=3B, Evaluation Protocol=Five-shot2026.0692.7
Attention ResidualsSize=1.5B, Evaluation Protocol=Five-shot2026.0691.9
DenseFormerSize=3B, Evaluation Protocol=Five-shot2026.0691.9
mHCSize=1.5B, Evaluation Protocol=Five-shot2026.0691.8
Depth-AttentionSize=1.5B, Evaluation Protocol=Five-shot2026.0691.5
VanillaSize=3B, Evaluation Protocol=Five-shot2026.0691.2
OriginalModel=LLaMA-2-7B, Ratio=1.0, Evaluation Protocol=0-shot2026.0491.1
DenseFormerSize=1.5B, Evaluation Protocol=Five-shot2026.0691
VanillaSize=1.5B, Evaluation Protocol=Five-shot2026.0690.1
Depth-AttentionSize=1.5B, Evaluation Protocol=Zero-shot2026.0688.2
GRASPruneModel=LLaMA-3.1-8B, Ratio=0.8, Evaluation Protocol=0-shot2026.0488.1
Depth-AttentionSize=3B, Evaluation Protocol=Zero-shot2026.0688
FLAPModel=LLaMA-2-7B, Ratio=0.8, Evaluation Protocol=0-shot2026.0487.8
DenseFormerSize=3B, Evaluation Protocol=Zero-shot2026.0686.9
Attention ResidualsSize=3B, Evaluation Protocol=Zero-shot2026.0686.7
mHCSize=1.5B, Evaluation Protocol=Zero-shot2026.0686.6
VanillaSize=3B, Evaluation Protocol=Zero-shot2026.0686.6
FLAPModel=LLaMA-3.1-8B, Ratio=0.8, Evaluation Protocol=0-shot2026.0486.5
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0486
Attention ResidualsSize=1.5B, Evaluation Protocol=Zero-shot2026.0685.9
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0485.8
mHCSize=3B, Evaluation Protocol=Zero-shot2026.0685.6
DenseFormerSize=1.5B, Evaluation Protocol=Zero-shot2026.0685.4
OracleModel=Llama3.1-8B, Black-Box Friendly=true2026.0485.1
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0484.8
OracleModel=Gemma2-9B, Black-Box Friendly=true2026.0484.6
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0484.4
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0484.2
VanillaSize=1.5B, Evaluation Protocol=Zero-shot2026.0684.1
RegMix-DProxy models=128, Mode=Online2026.0683.67
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0483.5
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0483.5
RegMix-DProxy models=512, Mode=Online2026.0683.45
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+1.87B, Evaluation Protocol=Zero/Few-shot2026.0483.4
RegMix-DProxy models=128, Mode=Offline2026.0683.35
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+1.17B, Evaluation Protocol=Zero/Few-shot2026.0482.9
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0482.8
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0482.8
MoRTBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0482.7
RegMix-DProxy models=512, Mode=Offline2026.0682.53
OracleModel=Qwen2.5-7B, Black-Box Friendly=true2026.0482
BaselineBackbone Scale=1.15B Scale, Cfg=N/A, # Params=1.15B, Evaluation Protocol=Zero/Few-shot2026.0482
RetokenBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0481.5
DoReMi2026.0681.42
RetokenBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0481.2
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0481
MoRTBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0481
RegMix2026.0680.98
OracleModel=Qwen2.5-3B, Black-Box Friendly=true2026.0480.8
MoRTBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0480.5
RetokenBackbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0480.4
OracleModel=Llama3.2-3B, Black-Box Friendly=true2026.0479.8
Human2026.0679.67
RetokenBackbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0478.8
EngramBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0478.7
RetokenBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0478.6
BaselineBackbone Scale=0.73B Scale, Cfg=N/A, # Params=0.73B, Evaluation Protocol=Zero/Few-shot2026.0478.4
MoRTBackbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0478.2
EngramBackbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0478.1
EngramBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0478
EngramBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0477.9
RetokenBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0477.5
EngramBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0477.5
MoRTBackbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0475
MoRTBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0474.3
RCSmedoidModel=Gemma2-9B, Black-Box Friendly=true2026.0474.2
RCSprobModel=Gemma2-9B, Black-Box Friendly=false2026.0474.2
GreedyModel=Gemma2-9B, Black-Box Friendly=true2026.0474.1
GRASPruneModel=LLaMA-3.1-8B, Ratio=0.6, Evaluation Protocol=0-shot2026.0474.1
RCSuniModel=Gemma2-9B, Black-Box Friendly=true2026.0474
SCModel=Gemma2-9B, Black-Box Friendly=true2026.0473.9
RCSfreqModel=Gemma2-9B, Black-Box Friendly=true2026.0473.9
EngramBackbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0473.9
CEModel=Gemma2-9B, Black-Box Friendly=false2026.0473.6
PionArchitecture=LLaMA-1.3B, Training Tokens=54B2026.0573.4
AC-ODM-410MSetting=Zero-shot, Pre-training Dataset=The Pile, Model Size=410M2025.0573
NLLModel=Gemma2-9B, Black-Box Friendly=false2026.0472.9
MuonArchitecture=LLaMA-1.3B, Training Tokens=54B2026.0571.6
GRASPruneModel=LLaMA-2-7B, Ratio=0.8, Evaluation Protocol=0-shot2026.0470.8
AdamWArchitecture=LLaMA-1.3B, Training Tokens=54B2026.0570.8
GreedyModel=Qwen2.5-7B, Black-Box Friendly=true2026.0470.5
CEModel=Qwen2.5-7B, Black-Box Friendly=false2026.0470.4
SCModel=Qwen2.5-7B, Black-Box Friendly=true2026.0470.3
RCSfreqModel=Qwen2.5-7B, Black-Box Friendly=true2026.0470.3
RCSuniModel=Qwen2.5-7B, Black-Box Friendly=true2026.0470.2
AC-ODMSetting=Zero-shot, Pre-training Dataset=The Pile, Model Size=1B2025.0570.2
RCSmedoidModel=Qwen2.5-7B, Black-Box Friendly=true2026.0470.1
RCSprobModel=Qwen2.5-7B, Black-Box Friendly=false2026.0470
FLAPModel=LLaMA-3.1-8B, Ratio=0.6, Evaluation Protocol=0-shot2026.0470
GreedyModel=Llama3.1-8B, Black-Box Friendly=true2026.0469
ODMSetting=Zero-shot, Pre-training Dataset=The Pile, Model Size=1B2025.0568.9
FLAPModel=LLaMA-3.1-8B, Ratio=0.5, Evaluation Protocol=0-shot2026.0468.8
RCSuniModel=Llama3.1-8B, Black-Box Friendly=true2026.0468.1
RCSmedoidModel=Llama3.1-8B, Black-Box Friendly=true2026.0468
RCSprobModel=Llama3.1-8B, Black-Box Friendly=false2026.0467.9
KALAVAI MoEModel Scale=Pythia-6.9B, Training Steps=10000 base, Freeze Layers=6, Seed=42, Number of Examples=5002026.0367.8
RCSfreqModel=Llama3.1-8B, Black-Box Friendly=true2026.0467.6