List your company

Mathematical Reasoning on GSM8K (4-shot)

86.13ScoreBase
-2.030820.857143.74566.6329Feb 16, 2025May 4, 2025Jul 21, 2025Oct 6, 2025Dec 23, 2025Mar 10, 2026May 27, 2026

Evaluation results

MethodLinks
BaseBase Model=Qwen2.5-7B2025.0286.13————
SKBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0284.61————
ALBMBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0284.08————
BaseBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0283.93————
InfoRMBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0283.78————
PoEBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0283.62————
BNRMBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0282.49————
No watermarkshot count=4-shot2026.0481.12————
UPVshot count=4-shot2026.0478.95————
ALBMBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0278.85————
KLASSBase Model=LLaDA-8B-Instruct2026.0578.8—15.21151.8
VanillaModel=D2F-Dream, Decoding Algo=Vanilla2025.1278.53.1———
BaselineBase Model=Dream-7B-Instruct2026.0578.2—5.72561
SKBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0278.17————
LEAPBase Model=LLaDA-8B-Instruct2026.0578—34.3584
VOWshot count=4-shot, delta=2.02026.0477.94————
PoEBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0277.79————
Conf-BasedBase Model=Dream-7B-Instruct2026.0577.6—24.7604.2
Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=true2026.0577.52—31.48—3.91
Dream-v0-7B-InstructParallel Decoding=false2026.0577.47—8.05—1
dKV-CacheAcceleration Method=dKV-Cache, Parallel Decoding=false2026.0577.32—12.8—1.59
Dynamic-dLLMAcceleration Method=Dynamic-dLLM, Parallel Decoding=false2026.0577.28—25.52—3.17
LPBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0277.18————
BNRMBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0277.1————
Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=true2026.0576.95—28.9—3.59
LEAPBase Model=Dream-7B-Instruct2026.0576.9—31.3445.5
Fast-dLLMAcceleration Method=Fast-dLLM, Parallel Decoding=false2026.0576.81—21.5—2.67
InfoRMBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0276.74————
dLLM-CacheAcceleration Method=dLLM-Cache, Parallel Decoding=false2026.0576.64—22.46—2.79
Conf-BasedBase Model=LLaDA-8B-Instruct2026.0576.5—27.2793.2
BaselineBase Model=LLaDA-8B-Instruct2026.0576.3—8.42561
KLASSBase Model=Dream-7B-Instruct2026.0576.1—12.31132.2
VOWshot count=4-shot, delta=2.52026.0476.04————
LPBase Model=Llama3.1-8B-Instruct, Evaluation Protocol=4-shot2026.0275.97————
LoPABase Model=Dream-7B-Instruct2026.0575.6—12.5732.2
LeftHashshot count=4-shot2026.0475.59————
BaseBase Model=OpenRLHF-Llama3-8B-SFT, Evaluation Protocol=4-shot2026.0274.83————
DPOBase Model=Qwen2.5-7B2025.0274.53————
SelfHashshot count=4-shot2026.0474.37————
VARBase Model=Qwen2.5-7B2025.0274.3————
LoPAModel=D2F-Dream, Decoding Algo=LoPA2025.1273.810.1———
LoPAModel=Dream, Decoding Algo=LoPA2025.1273.33.1———
VanillaModel=Dream, Decoding Algo=Vanilla2025.1272.61———
Fast-dLLMModel=Dream, Decoding Algo=Fast-dLLM2025.1272.62.1———
ALoLBase Model=Qwen2.5-7B2025.0264.97————
VARBase Model=Llama2-7B2025.026.67————
DPOBase Model=Llama2-7B2025.024.55————
ALoLBase Model=Llama2-7B2025.024.09————
RDFshot count=4-shot2026.042.43————
BaseBase Model=Llama2-7B2025.021.9————
PDWshot count=4-shot2026.041.36————