Natural Language Processing on FLAN 8-Task Subset
71Closed-book QAFFA-LoRA
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FFA-LoRABackbone=Qwen3-4B-Instruct-2507, Training Rounds=6002026.02 | 71 | 70.54 | 77.18 | 79.89 | 85.65 | 86.33 | 73.57 | 77.06 | 77.65 | |
| POTARABackbone=Qwen3-4B-Instruct-2507, Training Rounds=300 (FedIT) + 300 (Local)2026.02 | 70.61 | 70.61 | 88.26 | 81.56 | 88.8 | 86.78 | 75.18 | 76.1 | 79.74 | |
| FedDPABackbone=Qwen3-4B-Instruct-2507, Training Rounds=6002026.02 | 70.03 | 70.57 | 81.89 | 81.44 | 86.93 | 87.55 | 74.01 | 76.45 | 78.61 | |
| FedITBackbone=Qwen3-4B-Instruct-2507, Training Rounds=6002026.02 | 69.96 | 71.37 | 83.19 | 80.44 | 85.37 | 86.33 | 73.78 | 76.68 | 78.39 | |
| FedSABackbone=Qwen3-4B-Instruct-2507, Training Rounds=6002026.02 | 69.54 | 69.54 | 87.78 | 81.89 | 84.05 | 86 | 72.72 | 76.47 | 78.5 | |
| LocalBackbone=Qwen3-4B-Instruct-2507, Training Rounds=3002026.02 | 69.53 | 69.38 | 88.01 | 80 | 87.62 | 85.78 | 73.87 | 76.16 | 78.79 | |
| FedALTBackbone=Qwen3-4B-Instruct-2507, Training Rounds=6002026.02 | 68.07 | 72.5 | 84.84 | 82.44 | 85.17 | 86.78 | 73.43 | 77.4 | 78.83 |