Instruction Following on Fed-WildChat Real Dataset 1.0 (test)
4.78MT-Bench ScoreFedDQC
Evaluation results
| Method | Links | |
|---|---|---|
| FedDQCBackbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=FedDQC2024.10 | 4.78 | |
| FedAvg+IFDBackbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=Instruction-following Difficulty (IFD)2024.10 | 4.6 | |
| FedAvg+PPLBackbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=Perplexity (PPL)2024.10 | 4.525 | |
| FedAvg (oracle)Backbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=Full Clean Data (Oracle)2024.10 | 4.475 | |
| FedAvg+DataInfBackbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=DataInf2024.10 | 4.443 | |
| FedAvg+NUGGETSBackbone=LLama2-7b, Fine-tuning=LoRA, Data Selection=NUGGETS2024.10 | 4.443 |