List your company
ResearchFormerly WizWand

Multilingual Mathematical Reasoning on MT Math100

93.6AccuracyROSA2
12.27233.38654.575.614Sep 27, 2025Oct 23, 2025Nov 18, 2025Dec 14, 2025Jan 9, 2026Feb 4, 2026Mar 2, 2026
Updated 8mo ago

Evaluation results

MethodLinks
ROSA2Model=Qwen3-8B2026.0393.6——
ROSAModel=Qwen3-8B2026.0388.4——
ROSAModel=Qwen3-8B, Update Location=LM Head, Reward Model=Model-based2025.0988.37——
ROSAModel=Qwen3-8B, Update Location=Hidden States, Reward Model=Rule-based2025.0986.93——
ROSAModel=Qwen3-8B, Update Location=LM Head, Reward Model=Rule-based2025.0985.16——
TextGradModel=Qwen3-8B2026.0381.2——
ROSA2Model=Qwen2.5-7B-Base2026.0378.2——
TextGradModel=Qwen2.5-7B-Base2026.0375.4——
BaselineModel=Qwen3-8B2026.0375.2——
ROSAModel=Qwen2.5-7B-Instruct, Update Location=LM Head, Reward Model=Model-based2025.0975.13——
BaselineModel=Qwen3-8B2025.0974.74——
ROSA2Model=Qwen3-0.6B-Instruct2026.0373.4——
ROSAModel=Qwen2.5-7B-Instruct, Update Location=LM Head, Reward Model=Rule-based2025.0973.16——
ROSAModel=Qwen2.5-7B-Instruct, Update Location=Hidden States, Reward Model=Rule-based2025.0972.27——
ROSAModel=Qwen2.5-7B-Base2026.0370.4——
TextGradModel=Qwen3-0.6B-Instruct2026.0362.2——
ROSAModel=Qwen3-0.6B-Instruct2026.0362——
BaselineModel=Qwen2.5-7B-Base2026.0360.4——
BaselineModel=Qwen2.5-7B-Instruct2025.0960.34——
SP3F-7BLanguage=nl2026.0160.187—
Qwen2.5-7B-Instruct + Translate TestTraining Stage=Instruct, Evaluation Protocol=Translate Test2026.0160.08—59.34
SP3F-7BLanguage=vi2026.0159.894.4—
ROSAModel=Qwen3-0.6B, Update Location=LM Head, Reward Model=Model-based2025.0959.4——
SP3F-7BLanguage=af2026.0159.183.7—
Qwen2.5-7B-InstructLanguage=nl2026.0158.558.8—
Qwen2.5-7B-InstructLanguage=vi2026.0156.975.4—
SP3F-7BTraining Stage=Full Pipeline2026.0156.84—82.93
ROSAModel=Qwen3-0.6B, Update Location=LM Head, Reward Model=Rule-based2025.0956.6——
SP3F-7BLanguage=he2026.0156.475.5—
Qwen2.5-7B-InstructLanguage=af2026.0155.962.9—
SP3F-7BLanguage=tr2026.0153.392—
Qwen2.5-7B-InstructLanguage=he2026.0152.736.4—
Qwen2.5-7B-InstructTraining Stage=Instruct2026.0152.12—65.66
ROSAModel=Qwen3-0.6B, Update Location=Hidden States, Reward Model=Rule-based2025.0951.9——
SP3F-7BLanguage=tl2026.0151.887.1—
SP3F-7BLanguage=Average2026.0151.783.6—
Qwen2.5-7B-InstructLanguage=tr2026.0150.670.7—
ROSA2Model=DeepSeek-R1-Distill-Llama-8B2026.0350.6——
Qwen2.5-7B-InstructLanguage=Average2026.0148.362.7—
Qwen2.5-7B + RLVRTraining Stage=SFT + RLVR2026.0144.5—86.1
Qwen2.5-7B-InstructLanguage=tl2026.0144.456.7—
ROSAModel=DeepSeek-R1-Distill-Llama-8B2026.0338.6——
SP3F-7BLanguage=gu2026.013868.9—
Qwen2.5-7B-InstructLanguage=gu2026.0135.565—
SP3F-7BLanguage=pa2026.0135.180.3—
Qwen2.5-7B-InstructLanguage=pa2026.0132.175.5—
BaselineModel=Qwen3-0.6B2025.0931.3——
TextGradModel=DeepSeek-R1-Distill-Llama-8B2026.0330.4——
Qwen2.5-7B + SFTTraining Stage=SFT2026.0126.72—58.26
BaselineModel=Qwen3-0.6B-Instruct2026.0326.2——
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=LM Head, Reward Model=Model-based2025.0925.2——
ROSA2Model=Qwen2.5-0.5B-Instruct2026.0325.2——
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=LM Head, Reward Model=Model-based2025.0924.67——
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=Hidden States, Reward Model=Rule-based2025.0923.85——
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=LM Head, Reward Model=Rule-based2025.0922.8——
ROSAModel=DeepSeek-R1-Distill-Llama-8B, Update Location=LM Head, Reward Model=Rule-based2025.0921.17——
Qwen2.5-7BTraining Stage=Base2026.0121.16—58.22
ROSAModel=Qwen2.5-0.5B-Instruct, Update Location=Hidden States, Reward Model=Rule-based2025.0920.9——
ROSAModel=Qwen2.5-0.5B-Instruct2026.0319.6——
TextGradModel=Qwen2.5-0.5B-Instruct2026.0318.4——
BaselineModel=DeepSeek-R1-Distill-Llama-8B2026.0317.8——
BaselineModel=DeepSeek-R1-Distill-Llama-8B2025.0917.35——
BaselineModel=Qwen2.5-0.5B-Instruct2025.0915.4——
BaselineModel=Qwen2.5-0.5B-Instruct2026.0315.4——