Preference Alignment on UltraFeedback (test)
74.18AccuracyFedPDPO
Evaluation results
| Method | Links | |
|---|---|---|
| FedPDPOBackbone=TinyLlama-1.1B, Federated Strategy=FedPDPO, Optimization Algorithm=DPO, Number of Clients=32026.03 | 74.18 | |
| FedPer(+DPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedPer, Optimization Algorithm=DPO, Number of Clients=32026.03 | 73.12 | |
| FedRep(+DPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedRep, Optimization Algorithm=DPO, Number of Clients=32026.03 | 72.67 | |
| Per-FedAvg(+DPO)Backbone=TinyLlama-1.1B, Federated Strategy=Per-FedAvg, Optimization Algorithm=DPO, Number of Clients=32026.03 | 72.41 | |
| FedAMP(+DPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedAMP, Optimization Algorithm=DPO, Number of Clients=32026.03 | 71.56 | |
| FedPer(+PPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedPer, Optimization Algorithm=PPO, Number of Clients=32026.03 | 71.38 | |
| FedAvg(+DPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedAvg, Optimization Algorithm=DPO, Number of Clients=32026.03 | 71.28 | |
| FedRep(+PPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedRep, Optimization Algorithm=PPO, Number of Clients=32026.03 | 70.29 | |
| Per-FedAvg(+PPO)Backbone=TinyLlama-1.1B, Federated Strategy=Per-FedAvg, Optimization Algorithm=PPO, Number of Clients=32026.03 | 70.15 | |
| FedAvg(+PPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedAvg, Optimization Algorithm=PPO, Number of Clients=32026.03 | 69.54 | |
| FedAMP(+PPO)Backbone=TinyLlama-1.1B, Federated Strategy=FedAMP, Optimization Algorithm=PPO, Number of Clients=32026.03 | 69.12 |