Mathematical Reasoning on DeepMath (Pass@1)
70.5Pass@1FedAvg-PubSwap
Evaluation results
| Method | Links | |
|---|---|---|
| FedAvg-PubSwapModel=Qwen3-4B-Instruct-2507, local steps (τ)=1202026.04 | 70.5 | |
| FedAvg-PubSwapModel=Qwen3-4B-Instruct-2507, local steps (τ)=902026.04 | 70.2 | |
| FedAvg-PubSwapModel=Qwen3-4B-Instruct-2507, local steps (τ)=102026.04 | 70 | |
| FedAvg-GRPOModel=Qwen3-4B-Instruct-2507, local steps (τ)=1202026.04 | 67.4 | |
| FedAvg-GRPOModel=Qwen3-4B-Instruct-2507, local steps (τ)=902026.04 | 67 | |
| FedAvg-GRPOModel=Qwen3-4B-Instruct-2507, local steps (τ)=102026.04 | 66.9 | |
| FedAvg-GRPOModel=Qwen3-4B-Instruct-2507, local steps (τ)=402026.04 | 66.5 | |
| FedAvg-PubSwapModel=Qwen3-4B-Instruct-2507, local steps (τ)=402026.04 | 65.4 | |
| FedAvg-PubSwapModel=Qwen3-1.7B, number of local steps (τ)=1202026.04 | 55.8 | |
| FedAvg-PubSwapModel=Qwen3-1.7B, number of local steps (τ)=902026.04 | 53.3 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, number of local steps (τ)=402026.04 | 53.1 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, number of local steps (τ)=1202026.04 | 53.1 | |
| FedAvg-PubSwapModel=Qwen3-1.7B, number of local steps (τ)=402026.04 | 53 | |
| FedAvg-GRPOModel=Qwen3-1.7B, number of local steps (τ)=402026.04 | 52.7 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, number of local steps (τ)=102026.04 | 52.4 | |
| FedProx-GRPOModel=Qwen3-1.7B, number of local steps (τ)=402026.04 | 52.3 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=402026.04 | 52.3 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=102026.04 | 52.2 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=402026.04 | 52.1 | |
| FedAvg-PubSwapModel=Qwen3-1.7B, number of local steps (τ)=102026.04 | 51.1 | |
| Base modelModel=Qwen3-4B-Instruct-2507, local steps (τ)=102026.04 | 51.1 | |
| Base modelModel=Qwen3-4B-Instruct-2507, local steps (τ)=402026.04 | 51.1 | |
| Base modelModel=Qwen3-4B-Instruct-2507, local steps (τ)=902026.04 | 51.1 | |
| Base modelModel=Qwen3-4B-Instruct-2507, local steps (τ)=1202026.04 | 51.1 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=102026.04 | 51 | |
| FedAvg-PubSwapModel=Qwen2.5-Math-1.5B, number of local steps (τ)=902026.04 | 50.8 | |
| FedAvg-GRPOModel=Qwen3-1.7B, number of local steps (τ)=1202026.04 | 50.7 | |
| FedProx-GRPOModel=Qwen3-1.7B, number of local steps (τ)=102026.04 | 50.6 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=902026.04 | 50.5 | |
| FedAvg-GRPOModel=Qwen3-1.7B, number of local steps (τ)=902026.04 | 50.4 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=1202026.04 | 49.4 | |
| FedAvg-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=1202026.04 | 49.3 | |
| FedAvg-GRPOModel=Qwen3-1.7B, number of local steps (τ)=102026.04 | 48.9 | |
| FedProx-GRPOModel=Qwen3-1.7B, number of local steps (τ)=902026.04 | 48 | |
| FedProx-GRPOModel=Qwen2.5-Math-1.5B, number of local steps (τ)=902026.04 | 48 | |
| FedProx-GRPOModel=Qwen3-1.7B, number of local steps (τ)=1202026.04 | 47.7 | |
| BaseModel=Qwen2.5-Math-1.5B, number of local steps (τ)=102026.04 | 34.9 | |
| BaseModel=Qwen2.5-Math-1.5B, number of local steps (τ)=402026.04 | 34.9 | |
| BaseModel=Qwen2.5-Math-1.5B, number of local steps (τ)=902026.04 | 34.9 | |
| BaseModel=Qwen2.5-Math-1.5B, number of local steps (τ)=1202026.04 | 34.9 | |
| Base modelModel=Qwen3-1.7B, number of local steps (τ)=102026.04 | 14.9 | |
| Base modelModel=Qwen3-1.7B, number of local steps (τ)=402026.04 | 14.9 | |
| Base modelModel=Qwen3-1.7B, number of local steps (τ)=902026.04 | 14.9 | |
| Base modelModel=Qwen3-1.7B, number of local steps (τ)=1202026.04 | 14.9 |