Instruction Following on Alpaca GPT4
5.33MT-Bench ScoreFed full FT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Fed full FTBackbone=Llama 3-8B2025.11 | 5.33 | 16 | |
| FedBCDBackbone=Llama 3-8B2025.11 | 5.14 | 30.2 | |
| ParaBlockBackbone=Llama 3-8B2025.11 | 5.14 | 21.1 | |
| FedITBackbone=Llama 3-8B2025.11 | 5.11 | 34.5 | |
| FFA-LoRABackbone=Llama 3-8B2025.11 | 5.08 | 30.2 | |
| FLoRABackbone=Llama 3-8B2025.11 | 4.95 | 65.4 | |
| FedCyBGDBackbone=Llama 3-8B2025.11 | 4.87 | 59.9 | |
| BaseBackbone=Llama 3-8B2025.11 | 4.72 | — | |
| ParaBlockBackbone=Llama 3.2-3B2025.11 | 4.4 | 11.9 | |
| Fed full FTBackbone=Llama 3.2-3B2025.11 | 4.36 | 12.5 | |
| FedBCDBackbone=Llama 3.2-3B2025.11 | 4.33 | 19.1 | |
| FedITBackbone=Llama 3.2-3B2025.11 | 4.31 | 23.8 | |
| FedCyBGDBackbone=Llama 3.2-3B2025.11 | 4.29 | 57.8 | |
| FFA-LoRABackbone=Llama 3.2-3B2025.11 | 4.25 | 23.3 | |
| FLoRABackbone=Llama 3.2-3B2025.11 | 4.23 | 34 | |
| BaseBackbone=Llama 3.2-3B2025.11 | 4.18 | — |