Question Answering on CQA
83.1AccuracyThree-Stage Fine-Tuning Method
Evaluation Results
| Method | Links | |
|---|---|---|
| Three-Stage Fine-Tuning MethodInstruction Model=Qwen2.5-14B2025.04 | 83.1 | |
| SFTInstruction Model=Qwen2.5-14B2025.04 | 81.4 | |
| OriginalInstruction Model=Qwen2.5-14B2025.04 | 80.9 | |
| Three-Stage Fine-Tuning MethodInstruction Model=Qwen2-7B2025.04 | 78.8 | |
| Three-Stage Fine-Tuning MethodInstruction Model=Llama3-8B2025.04 | 76.4 | |
| SFTInstruction Model=Qwen2-7B2025.04 | 72.1 | |
| OriginalInstruction Model=Llama3-8B2025.04 | 70.8 | |
| OriginalInstruction Model=Qwen2-7B2025.04 | 70.4 | |
| SFTInstruction Model=Llama3-8B2025.04 | 69.3 | |
| CentralizedExperimental Setting=S3, Server Model=LLaMa2-7B2024.11 | 69 | |
| CentralizedExperimental Setting=S2, Server Model=OPT-6.7B2024.11 | 68.6 | |
| Three-Stage Fine-Tuning MethodInstruction Model=Llama3.2-3B2025.04 | 68.5 | |
| FedCoLLMExperimental Setting=S2, Server Model=OPT-6.7B2024.11 | 68.1 | |
| FedCoLLMExperimental Setting=S3, Server Model=LLaMa2-7B2024.11 | 67.1 | |
| SFTInstruction Model=Llama3.2-3B2025.04 | 66.2 | |
| OriginalInstruction Model=Llama3.2-3B2025.04 | 65.1 | |
| MI-based distillationBackbone=T5-base2024.03 | 63.88 | |
| DSSBackbone=T5-base2024.03 | 63.29 | |
| FinetuningBackbone=T5-base2024.03 | 62.19 | |
| Single-taskBackbone=T5-base2024.03 | 61.37 | |
| CentralizedExperimental Setting=S1, Server Model=GPT-2-Large2024.11 | 54.7 | |
| FedCoLLMExperimental Setting=S1, Server Model=GPT-2-Large2024.11 | 53.5 | |
| Zero-ShotExperimental Setting=S2, Server Model=OPT-6.7B2024.11 | 48.7 | |
| Zero-ShotExperimental Setting=S3, Server Model=LLaMa2-7B2024.11 | 39.5 | |
| Zero-ShotExperimental Setting=S1, Server Model=GPT-2-Large2024.11 | 36.3 |