List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME25 (Pass@k)

66.7Pass@kQwen2.5-7B + PPO w/ VERL.
4.320.536.752.9Sep 28, 2025
Updated 6mo ago

Evaluation results

MethodLinks
Qwen2.5-7B + PPO w/ VERL.k=256, Base Model=Qwen2.5-7B, RL Method=PPO, Framework=VERL2025.0966.7
Qwen2.5-7Bk=256, Base Model=Qwen2.5-7B2025.0960
Qwen2.5-7B + GRPO w/ VERL.k=256, Base Model=Qwen2.5-7B, RL Method=GRPO, Framework=VERL2025.0960
Qwen2.5-7B + PPOk=256, Base Model=Qwen2.5-7B, RL Method=PPO2025.0956.7
Qwen2.5-7B + GRPOk=256, Base Model=Qwen2.5-7B, RL Method=GRPO2025.0950
Mathstral-7B-v0.1 + GRPO w/ VERL.k=256, Base Model=Mathstral-7B-v0.1, RL Method=GRPO, Framework=VERL2025.0950
Mathstral-7B-v0.1 + PPO w/ VERL.k=256, Base Model=Mathstral-7B-v0.1, RL Method=PPO, Framework=VERL2025.0946.7
Qwen2.5-3B + GRPO w/ VERL.k=256, Base Model=Qwen2.5-3B, RL Method=GRPO, Framework=VERL2025.0943.3
Qwen2.5-3B + PPOk=256, Base Model=Qwen2.5-3B, RL Method=PPO2025.0943.3
Qwen2.5-3B + PPO w/ VERL.k=256, Base Model=Qwen2.5-3B, RL Method=PPO, Framework=VERL2025.0943.3
Llama-3.2-3B-Instruct + PPO w/ VERL.k=256, Base Model=Llama-3.2-3B-Instruct, RL Method=PPO, Framework=VERL2025.0940
Qwen2.5-3Bk=256, Base Model=Qwen2.5-3B2025.0940
Qwen2.5-3B + GRPOk=256, Base Model=Qwen2.5-3B, RL Method=GRPO2025.0940
Mathstral-7B-v0.1 + GRPOk=256, Base Model=Mathstral-7B-v0.1, RL Method=GRPO2025.0940
Mathstral-7B-v0.1 + PPOk=256, Base Model=Mathstral-7B-v0.1, RL Method=PPO2025.0940
Llama-3.2-3B-Instruct + GRPO w/ VERL.k=256, Base Model=Llama-3.2-3B-Instruct, RL Method=GRPO, Framework=VERL2025.0936.7
Llama-3.2-3B-Instruct + PPOk=256, Base Model=Llama-3.2-3B-Instruct, RL Method=PPO2025.0936.7
Llama-3.1-8B-Instructk=256, Base Model=Llama-3.1-8B-Instruct2025.0936.7
Llama-3.1-8B-Instruct + GRPOk=256, Base Model=Llama-3.1-8B-Instruct, RL Method=GRPO2025.0936.7
Llama-3.1-8B-Instruct + GRPO w/ VERL.k=256, Base Model=Llama-3.1-8B-Instruct, RL Method=GRPO, Framework=VERL2025.0936.7
Llama-3.1-8B-Instruct + PPOk=256, Base Model=Llama-3.1-8B-Instruct, RL Method=PPO2025.0936.7
Llama-3.1-8B-Instruct + PPO w/ VERL.k=256, Base Model=Llama-3.1-8B-Instruct, RL Method=PPO, Framework=VERL2025.0936.7
Mathstral-7B-v0.1k=256, Base Model=Mathstral-7B-v0.12025.0936.7
Llama-3.2-3B-Instructk=256, Base Model=Llama-3.2-3B-Instruct2025.0930
Llama-3.2-3B-Instruct + GRPOk=256, Base Model=Llama-3.2-3B-Instruct, RL Method=GRPO2025.0930
Mistral-7B-v0.3k=256, Base Model=Mistral-7B-v0.32025.0916.7
Mistral-7B-v0.3 + GRPO w/ VERL.k=256, Base Model=Mistral-7B-v0.3, RL Method=GRPO, Framework=VERL2025.0913.3
Mistral-7B-v0.3 + PPO w/ VERL.k=256, Base Model=Mistral-7B-v0.3, RL Method=PPO, Framework=VERL2025.0913.3
Mistral-7B-v0.3 + GRPOk=256, Base Model=Mistral-7B-v0.3, RL Method=GRPO2025.0910
Mistral-7B-v0.3 + PPOk=256, Base Model=Mistral-7B-v0.3, RL Method=PPO2025.096.7