List your company
ResearchFormerly WizWand

Science Question Answering on GPQA (accuracy)

63.51AccuracySFT
2.763618.534334.30550.0757Apr 23, 2025Jun 22, 2025Aug 21, 2025Oct 20, 2025Dec 19, 2025Feb 17, 2026Apr 18, 2026
Updated 6mo ago

Evaluation results

MethodLinks
SFTModel=DS-R1-Qwen-32B2026.0463.51
D-QRELOModel=DS-R1-Qwen-32B2026.0461.47
Delta-CoMeModel=DS-R1-Qwen-32B2026.0458.46
SFTModel=DS-R1-Qwen-14B2026.0456.94
Llama-InstructShots=0-shot2026.0354.5
D-QRELOModel=DS-R1-Qwen-14B2026.0453.27
Delta-CoMeModel=DS-R1-Qwen-14B2026.0451.92
SVDModel=DS-R1-Qwen-32B2026.0450.43
R1-Distill-Qwen-7BFT data=R1*, LoRA=-2026.0349
Qwen-7BFT data=MoT, LoRA=-2026.0348
Qwen-7BFT data=OT3 + MoT, LoRA=rk 1282026.0348
Qwen-7BFT data=OT3, LoRA=-2026.0347
Llama 3.1 InstructModel Scale=70B2025.0446.7
SFTModel=DS-R1-LLaMA-8B2026.0445.2
Qwen-7BFT data=OT3, LoRA=rk 642026.0345
WandaModel=DS-R1-Qwen-32B2026.0444.52
SVDModel=DS-R1-Qwen-14B2026.0444.34
Qwen-7BFT data=OT3, LoRA=rk 1282026.0343
MagnitudeModel=DS-R1-Qwen-32B2026.0442.74
ParamΔModel Scale=70B2025.0442.2
Llama 3 InstructModel Scale=70B2025.0440.6
Qwen-3BFT data=OT3, LoRA=-2026.0340
BitDeltaModel=DS-R1-Qwen-32B2026.0439.43
WandaModel=DS-R1-Qwen-14B2026.0439.04
Evo 8BShots=5, E2E Latency (s)=8.6, Inference Speed (tokens/s)=522026.0238.4
MagnitudeModel=DS-R1-Qwen-14B2026.0437.82
Qwen-3BFT data=MoT, LoRA=-2026.0337
Qwen-7BFT data=-, LoRA=-2026.0337
BackboneModel=DS-R1-Qwen-32B2026.0436.36
Qwen2.5 7BShots=5, E2E Latency (s)=8.1, Inference Speed (tokens/s)=462026.0236.1
HATified-SFTShots=0-shot2026.0336
D-QRELOModel=DS-R1-LLaMA-8B2026.0435.98
BitDeltaModel=DS-R1-Qwen-14B2026.0435.35
M_RLP + PostPre-training Condition=Reinforcement Learning from Pretraining, Post-training=SFT + RLVR, Evaluation Protocol=@1[4]2025.0934.97
RandomModel=DS-R1-Qwen-32B2026.0434.93
M_RLP + PostPre-training Condition=Reinforcement Learning from Pretraining, Post-training=SFT + RLVR, Evaluation Protocol=Standard2025.0933.33
Llama 3 InstructModel Scale=8B2025.0432.8
BackboneModel=DS-R1-Qwen-14B2026.0432.32
Qwen-3BFT data=OT3, LoRA=rk 1282026.0332
M_base + PostPre-training Condition=Base, Post-training=SFT + RLVR, Evaluation Protocol=@1[4]2025.0931.52
MagnitudeModel=DS-R1-LLaMA-8B2026.0431.19
Qwen-3BFT data=-, LoRA=-2026.0331
M_base + PostPre-training Condition=Base, Post-training=SFT + RLVR, Evaluation Protocol=Standard2025.0930.93
WandaModel=DS-R1-LLaMA-8B2026.0430.72
ParamΔModel Scale=8B2025.0430.6
BitDeltaModel=DS-R1-LLaMA-8B2026.0430.56
BD3-LM 7BShots=5, E2E Latency (s)=14.2, Inference Speed (tokens/s)=282026.0230.5
RandomModel=DS-R1-Qwen-14B2026.0430.39
SVDModel=DS-R1-LLaMA-8B2026.0430.32
Delta-CoMeModel=DS-R1-LLaMA-8B2026.0430.18
M_CPT + PostPre-training Condition=Continuous Pretraining, Post-training=SFT + RLVR, Evaluation Protocol=@1[4]2025.0930.01
Llama 3.1 InstructModel Scale=8B2025.0429.7
M_CPT + PostPre-training Condition=Continuous Pretraining, Post-training=SFT + RLVR, Evaluation Protocol=Standard2025.0929.27
M_RLPPre-training Condition=Reinforcement Learning from Pretraining, Post-training=None, Evaluation Protocol=Standard2025.0928.28
M_basePre-training Condition=Base, Post-training=None, Evaluation Protocol=@1[4]2025.0927.52
M_RLPPre-training Condition=Reinforcement Learning from Pretraining, Post-training=None, Evaluation Protocol=@1[4]2025.0927.02
LLaDA 8BShots=5, E2E Latency (s)=21.8, Inference Speed (tokens/s)=162026.0226.4
M_CPTPre-training Condition=Continuous Pretraining, Post-training=None, Evaluation Protocol=Standard2025.0926.26
LLaMA3 8BShots=5, E2E Latency (s)=7.4, Inference Speed (tokens/s)=582026.0226.2
ARD 7BShots=5, E2E Latency (s)=32.5, Inference Speed (tokens/s)=122026.0225.3
M_basePre-training Condition=Base, Post-training=None, Evaluation Protocol=Standard2025.0925.25
MDLM 7BShots=5, E2E Latency (s)=18.9, Inference Speed (tokens/s)=222026.0224.9
M_CPTPre-training Condition=Continuous Pretraining, Post-training=None, Evaluation Protocol=@1[4]2025.0924.75
Llama 3.1 BaseModel Scale=70B2025.0422.8
RandomModel=DS-R1-LLaMA-8B2026.0422.22
BackboneModel=DS-R1-LLaMA-8B2026.0420.13
Llama 3 BaseModel Scale=70B2025.0414.3
Llama 3.1 BaseModel Scale=8B2025.046.3
Llama 3 BaseModel Scale=8B2025.045.1