List your company
ResearchFormerly WizWand

Mathematical Reasoning on AMC 2023 (Pass@1, Avg@8, Pass@8)

82.5Pass@1MEL
20.136.352.568.7Sep 28, 2025Oct 20, 2025Nov 12, 2025Dec 4, 2025Dec 27, 2025Jan 18, 2026Feb 10, 2026
Updated 6mo ago

Evaluation results

MethodLinks
MELBackbone=Qwen3-14B-Base2026.0282.582.8195
GRPOBackbone=Qwen3-14B-Base2026.027575.9495
MELBackbone=Qwen3-8B-Base2026.027070.3190
GRPOBackbone=Qwen3-8B-Base2026.0267.569.0687.5
BaselineBackbone=Qwen3-8B-Base2026.026552.587.5
MELBackbone=Qwen3-4B-Base2026.026060.3187.5
BaselineBackbone=Qwen3-14B-Base2026.026051.2582.5
GRPOBackbone=Qwen3-4B-Base2026.0257.558.1385
Qwen2.5-7B + GRPOBase Model=Qwen2.5-7B, RL Framework=GRPO, VERL Integration=false2025.0955——
Qwen2.5-7B + PPO w/ VERL.Base Model=Qwen2.5-7B, RL Framework=PPO, VERL Integration=true2025.0952.5——
Qwen2.5-7B + GRPO w/ VERL.Base Model=Qwen2.5-7B, RL Framework=GRPO, VERL Integration=true2025.0950——
Qwen2.5-7B + PPOBase Model=Qwen2.5-7B, RL Framework=PPO, VERL Integration=false2025.0950——
BaselineBackbone=Qwen3-4B-Base2026.024542.7372.5
Qwen2.5-7BBase Model=Qwen2.5-7B, RL Framework=None, VERL Integration=false2025.0945——
Llama-3.2-3B-Instruct + GRPOBase Model=Llama-3.2-3B-Instruct, RL Framework=GRPO, VERL Integration=false2025.0927.5——
Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct, RL Framework=None, VERL Integration=false2025.0925——
Llama-3.2-3B-Instruct + GRPO w/ VERL.Base Model=Llama-3.2-3B-Instruct, RL Framework=GRPO, VERL Integration=true2025.0925——
Llama-3.2-3B-Instruct + PPO w/ VERL.Base Model=Llama-3.2-3B-Instruct, RL Framework=PPO, VERL Integration=true2025.0925——
Llama-3.2-3B-Instruct + PPOBase Model=Llama-3.2-3B-Instruct, RL Framework=PPO, VERL Integration=false2025.0922.5——