List your company
ResearchFormerly WizWand

Mathematical Reasoning on MAWPS (Pass@1)

97.8Pass@1Qwen2.5-7B + PPO w/ VERL.
81.1685.4889.894.12Mar 4, 2024Jun 7, 2024Sep 11, 2024Dec 15, 2024Mar 21, 2025Jun 24, 2025Sep 28, 2025
Updated 6mo ago

Evaluation results

MethodLinks
Qwen2.5-7B + PPO w/ VERL.Base Model=Qwen2.5-7B, RL Framework=PPO, VERL Integration=true2025.0997.8
Qwen2.5-7B + GRPO w/ VERL.Base Model=Qwen2.5-7B, RL Framework=GRPO, VERL Integration=true2025.0997.7
GPT-4 (0613)Zero-shot without demonstrations=false2024.0397.6
Qwen2.5-7B + GRPOBase Model=Qwen2.5-7B, RL Framework=GRPO, VERL Integration=false2025.0997.6
Qwen2.5-7B + PPOBase Model=Qwen2.5-7B, RL Framework=PPO, VERL Integration=false2025.0997.3
Llama-3.2-3B-Instruct + GRPOBase Model=Llama-3.2-3B-Instruct, RL Framework=GRPO, VERL Integration=false2025.0996
Llama-3.2-3B-Instruct + GRPO w/ VERL.Base Model=Llama-3.2-3B-Instruct, RL Framework=GRPO, VERL Integration=true2025.0996
Llama-3.2-3B-Instruct + PPO w/ VERL.Base Model=Llama-3.2-3B-Instruct, RL Framework=PPO, VERL Integration=true2025.0995.7
KPMath-PlusBase=Qwen1.5, Size=72B, Zero-shot without demonstrations=true2024.0395.5
Llama-3.2-3B-Instruct + PPOBase Model=Llama-3.2-3B-Instruct, RL Framework=PPO, VERL Integration=false2025.0995.5
KPMath-PlusBase=Llama-2, Size=70B, Zero-shot without demonstrations=true2024.0395.4
Qwen2.5-7BBase Model=Qwen2.5-7B, RL Framework=None, VERL Integration=false2025.0995.4
KPMath-PlusBase=DSMath, Size=7B, Zero-shot without demonstrations=true2024.0394.8
ChatGPTZero-shot without demonstrations=false2024.0394.6
KPMath-PlusBase=Llemma, Size=34B, Zero-shot without demonstrations=true2024.0394.5
KPMath-PlusBase=Mistral, Size=7B, Zero-shot without demonstrations=true2024.0394.2
KPMath-PlusBase=Llama-2, Size=13B, Zero-shot without demonstrations=true2024.0392.3
LEMMA (w/ MetaMath)Backbone=DeepSeekMath-7B, # Samples=403.59k2025.0391.9
LEMMABackbone=DeepSeekMath-7B, # Samples=88.90k2025.0390.9
MetaMathBackbone=DeepSeekMath-7B, # Samples=394.99k2025.0390.2
GPTAugBackbone=DeepSeekMath-7B, # Samples=88.62k2025.0389.6
LEMMA (w/ MetaMath)Backbone=LLaMA3-8B, # Samples=403.59k2025.0389.5
RFTBackbone=DeepSeekMath-7B, # Samples=86.52k2025.0389.3
ISCBackbone=DeepSeekMath-7B, # Samples=86.78k2025.0389.3
MetaMathBackbone=LLaMA3-8B, # Samples=394.99k2025.0388.9
LEMMABackbone=LLaMA3-8B, # Samples=88.90k2025.0388.8
RefAugBackbone=LLaMA3-8B, # Samples=29.94k2025.0388.4
SFTBackbone=DeepSeekMath-7B, # Samples=14.97k2025.0388.1
RefAug-90kBackbone=LLaMA3-8B, # Samples=89.92k2025.0387.7
Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct, RL Framework=None, VERL Integration=false2025.0986.9
GPTAugBackbone=LLaMA3-8B, # Samples=88.62k2025.0385.9
RefAug-90kBackbone=DeepSeekMath-7B, # Samples=89.92k2025.0383.1
SFTBackbone=LLaMA3-8B, # Samples=14.97k2025.0383
ISCBackbone=LLaMA3-8B, # Samples=86.78k2025.0382.3
S³C-Math (w/ MetaMath)Backbone=DeepSeekMath-7B, # Samples=927k2025.0382.2
RefAugBackbone=DeepSeekMath-7B, # Samples=29.94k2025.0382.1
RFTBackbone=LLaMA3-8B, # Samples=86.52k2025.0381.8
S³C-Math (w/ MetaMath)Backbone=LLaMA3-8B, # Samples=927k2025.0381.8