Mathematical Reasoning on MATH 2021 (test)
78.3pass@1FedAvg-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| FedAvg-GRPOModel=Qwen2-1.5B, Local steps (τ)=102026.04 | 78.3 | |
| FedProx-GRPOModel=Qwen2-1.5B, Local steps (τ)=102026.04 | 77.5 | |
| FedAvg-PubSwapModel=Qwen2-1.5B, Local steps (τ)=102026.04 | 77 | |
| FedAvg-PubSwapModel=Qwen2-1.5B, Local steps (τ)=402026.04 | 76.9 | |
| FedAvg-PubSwapModel=Qwen2-1.5B, Local steps (τ)=902026.04 | 76.9 | |
| FedProx-GRPOModel=Qwen2-1.5B, Local steps (τ)=402026.04 | 76.7 | |
| FedAvg-PubSwapModel=Qwen2-1.5B, Local steps (τ)=1202026.04 | 76.6 | |
| FedProx-GRPOModel=Qwen2-1.5B, Local steps (τ)=902026.04 | 76.5 | |
| FedAvg-GRPOModel=Qwen2-1.5B, Local steps (τ)=902026.04 | 76.3 | |
| FedAvg-GRPOModel=Qwen2-1.5B, Local steps (τ)=402026.04 | 76.1 | |
| FedAvg-GRPOModel=Qwen2-1.5B, Local steps (τ)=1202026.04 | 75.9 | |
| FedProx-GRPOModel=Qwen2-1.5B, Local steps (τ)=1202026.04 | 75.6 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=102026.04 | 73.7 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, Local steps (τ)=402026.04 | 73.5 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, Local steps (τ)=902026.04 | 73.2 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, Local steps (τ)=102026.04 | 73.1 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=402026.04 | 73.1 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=102026.04 | 72.8 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=902026.04 | 71.9 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, Local steps (τ)=1202026.04 | 71.9 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=902026.04 | 71.5 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=402026.04 | 71.4 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=1202026.04 | 71.1 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, Local steps (τ)=1202026.04 | 70.6 | |
| Base modelModel=Qwen2.5-Math-1.5B, Local steps (τ)=102026.04 | 58.4 | |
| Base modelModel=Qwen2.5-Math-1.5B, Local steps (τ)=402026.04 | 58.4 | |
| Base modelModel=Qwen2.5-Math-1.5B, Local steps (τ)=902026.04 | 58.4 | |
| Base modelModel=Qwen2.5-Math-1.5B, Local steps (τ)=1202026.04 | 58.4 | |
| Base modelModel=Qwen2-1.5B, Local steps (τ)=102026.04 | 55.2 | |
| Base modelModel=Qwen2-1.5B, Local steps (τ)=402026.04 | 55.2 | |
| Base modelModel=Qwen2-1.5B, Local steps (τ)=902026.04 | 55.2 | |
| Base modelModel=Qwen2-1.5B, Local steps (τ)=1202026.04 | 55.2 |