List your company
ResearchFormerly WizWand

Science Question Answering on ARC Challenge (Accuracy %)

93AccuracyTSDS
22.69640.94859.277.452Jun 15, 2025Aug 14, 2025Oct 13, 2025Dec 12, 2025Feb 10, 2026Apr 11, 2026Jun 11, 2026
Updated 4mo ago

Evaluation results

MethodLinks
TSDSIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0493
TSDSIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0492.8
Attr-DivIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0492.49
Attr-DivIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0492.15
Attr-DivIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.98
TSDSIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0491.98
EVOSELECTIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0491.98
EVOSELECTIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.98
RandomIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0491.89
EVOSELECTIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0491.89
BaseModel Scale=14b base model2026.0491.81
RandomIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0491.81
AttributionIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0491.81
DiversityIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0491.81
EVOSELECTIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0491.81
RandomIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.81
TSDSIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.81
Attr-DivIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0491.72
AttributionIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.72
AllIteration=Iter. 1, Model Scale=14b base model2026.0491.64
AllIteration=Iter. 2, Model Scale=14b base model2026.0491.64
DiversityIteration=Iter. 1, Selection Ratio=0.5, Model Scale=14b base model2026.0491.47
RandomIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0491.47
DiversityIteration=Iter. 2, Selection Ratio=0.5, Model Scale=14b base model2026.0491.47
AttributionIteration=Iter. 1, Selection Ratio=0.2, Model Scale=14b base model2026.0491.38
DiversityIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0491.21
AttributionIteration=Iter. 2, Selection Ratio=0.2, Model Scale=14b base model2026.0491.13
MSA-CPTTraining Strategy=sparse continued pretraining2026.0682.9
FullTraining Strategy=Full-Attention baseline2026.0682.7
MSA-PTTraining Strategy=from-scratch sparse pretraining2026.0682.5
RL-PLUSBase model=Qwen2.5-Math-7B2025.0782.3
GRPOBase model=Qwen2.5-Math-7B2025.0781.7
SFTBase model=Qwen2.5-Math-7B2025.0775.2
SFT+GRPOBase model=Qwen2.5-Math-7B2025.0772.4
GEMMA-7BState=Dense2025.0649.91
DEEPSEEK-7BState=Dense2025.0643.6
LLAMA-2-7BState=Dense2025.0643.34
MiCRoBase Model=Llama-3.2-3B, Model architecture type=MiCRo, Few-shot=true2025.0643.3
DenseBase Model=Llama-3.2-3B, Model architecture type=Dense, Few-shot=true2025.0642.9
VICUNA-1.3-7BState=Dense2025.0642.41
MoBBase Model=Llama-3.2-3B, Model architecture type=MoB, Few-shot=true2025.0641.3
MiCRoBase Model=SmollM2-360M, Model architecture type=MiCRo, Few-shot=true2025.0638.2
DenseBase Model=SmollM2-360M, Model architecture type=Dense, Few-shot=true2025.0637.5
MoBBase Model=SmollM2-360M, Model architecture type=MoB, Few-shot=true2025.0637.1
MASKLLMBackbone=LLAMA-2-7B2025.0635.67
MASKLLMBackbone=DEEPSEEK-7B2025.0635.32
DenseBase Model=Llama-3.2-1B, Model architecture type=Dense, Few-shot=true2025.0633.7
MASKLLMBackbone=VICUNA-1.3-7B2025.0633.57
MaskProBackbone=DEEPSEEK-7B2025.0633.49
MaskProBackbone=VICUNA-1.3-7B2025.0633.19
SPARSEGPTBackbone=DEEPSEEK-7B2025.0632.94
MaskProBackbone=LLAMA-2-7B2025.0632.85
GBLMBackbone=DEEPSEEK-7B2025.0632.85
MiCRoBase Model=Llama-3.2-1B, Model architecture type=MiCRo, Few-shot=true2025.0632.8
PRUNER-ZEROBackbone=VICUNA-1.3-7B2025.0632.76
SPARSEGPTBackbone=VICUNA-1.3-7B2025.0632.42
MoBBase Model=Llama-3.2-1B, Model architecture type=MoB, Few-shot=true2025.0632.4
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0432.2
WANDABackbone=VICUNA-1.3-7B2025.0632.17
WANDABackbone=DEEPSEEK-7B2025.0632
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0431.8
SPARSEGPTBackbone=LLAMA-2-7B2025.0631.48
GBLMBackbone=VICUNA-1.3-7B2025.0631.4
PRUNER-ZEROBackbone=DEEPSEEK-7B2025.0630.89
EngramBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0430.7
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+1.87B, Evaluation Protocol=Zero/Few-shot2026.0430.6
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0430.5
X-GRAM-100%Backbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0430.2
MAGNITUDEBackbone=LLAMA-2-7B2025.0630.2
WANDABackbone=LLAMA-2-7B2025.0630.2
MoBBase Model=SmollM2-135M, Model architecture type=MoB, Few-shot=true2025.0629.9
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0429.9
X-GRAM-50%Backbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0429.9
GBLMBackbone=LLAMA-2-7B2025.0629.86
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0429.7
DenseBase Model=SmollM2-135M, Model architecture type=Dense, Few-shot=true2025.0629.6
EngramBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0429.3
EngramBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0429
RetokenBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0428.8
MAGNITUDEBackbone=VICUNA-1.3-7B2025.0628.75
BaselineBackbone Scale=1.15B Scale, Cfg=N/A, # Params=1.15B, Evaluation Protocol=Zero/Few-shot2026.0428.7
MoRTBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0428.6
EngramBackbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0428.1
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0428.1
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0428
X-GRAM-100%Backbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0427.9
MoRTBackbone Scale=1.15B Scale, Cfg=4x, # Params=1.15B+14.94B, Evaluation Protocol=Zero/Few-shot2026.0427.9
EngramBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0427.8
MiCRoBase Model=SmollM2-135M, Model architecture type=MiCRo, Few-shot=true2025.0627.6
EngramBackbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0427.6
MAGNITUDEBackbone=DEEPSEEK-7B2025.0627.56
X-GRAM-50%Backbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+1.17B, Evaluation Protocol=Zero/Few-shot2026.0427.5
MoRTBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0427.4
PRUNER-ZEROBackbone=LLAMA-2-7B2025.0627.05
RetokenBackbone Scale=1.15B Scale, Cfg=2x, # Params=1.15B+7.47B, Evaluation Protocol=Zero/Few-shot2026.0427
RetokenBackbone Scale=1.15B Scale, Cfg=1x, # Params=1.15B+3.73B, Evaluation Protocol=Zero/Few-shot2026.0426.8
RetokenBackbone Scale=0.73B Scale, Cfg=2x, # Params=0.73B+4.67B, Evaluation Protocol=Zero/Few-shot2026.0426.2
MoRTBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0425.8
RetokenBackbone Scale=0.73B Scale, Cfg=1x, # Params=0.73B+2.33B, Evaluation Protocol=Zero/Few-shot2026.0425.5
MoRTBackbone Scale=0.73B Scale, Cfg=4x, # Params=0.73B+9.33B, Evaluation Protocol=Zero/Few-shot2026.0425.4