Email Writing on Email Writing Weak User
0.3047Mean Total Edit DistanceDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DPOTest-time User Model=Llama-3.3-70B-Instruct, Online learning phase=true2026.01 | 0.3047 | |
| DPOUser LLM=Qwen-3 32B, Seeds=3, Learning Phase=Online2026.01 | 0.3368 | |
| LateEnsembleTest-time User Model=Llama-3.3-70B-Instruct, Online learning phase=true2026.01 | 0.3428 | |
| LateEnsembleUser LLM=Qwen-3 32B, Seeds=3, Learning Phase=Online2026.01 | 0.3739 | |
| SFTTest-time User Model=Llama-3.3-70B-Instruct, Online learning phase=true2026.01 | 0.3866 | |
| EarlyEnsembleTest-time User Model=Llama-3.3-70B-Instruct, Online learning phase=true2026.01 | 0.4001 | |
| BaseTest-time User Model=Llama-3.3-70B-Instruct, Online learning phase=true2026.01 | 0.4239 | |
| SFTUser LLM=Qwen-3 32B, Seeds=3, Learning Phase=Online2026.01 | 0.4539 | |
| EarlyEnsembleUser LLM=Qwen-3 32B, Seeds=3, Learning Phase=Online2026.01 | 0.4864 | |
| BaseUser LLM=Qwen-3 32B, Seeds=3, Learning Phase=Online2026.01 | 0.4923 |