Performance Prediction on WMT En-De 2014 (val)
0.22MAELLM-Distill-PP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLM-Distill-PPTeacher Backbone=GPT-42023.10 | 0.22 | 0.64 | |
| LLM-PPBackbone=GPT-4, Ablation=+ Fourth instruction2023.10 | 0.25 | 0.63 | |
| LLM-PPBackbone=GPT-4, Ablation=+ First instruction2023.10 | 0.26 | 0.6 | |
| LLM-PPBackbone=GPT-4, Ablation=+ Role + Hyp.2023.10 | 0.27 | 0.53 | |
| LLM-PPBackbone=GPT-4, Ablation=+ Second instruction2023.10 | 0.27 | 0.6 | |
| LLM-PPBackbone=GPT-42023.10 | 0.28 | 0.65 | |
| LLM-PPBackbone=GPT-4, Ablation=+ Fifth instruction2023.10 | 0.28 | 0.65 | |
| LLM-PPBackbone=GPT-4, Ablation=Demonstrations only2023.10 | 0.31 | 0.52 | |
| LLM-PPBackbone=GPT-4, Ablation=+ Third instruction2023.10 | 0.31 | 0.5 | |
| LLM-Distill-PPTeacher Backbone=ChatGPT2023.10 | 0.32 | 0.6 | |
| LLM-PPBackbone=ChatGPT2023.10 | 0.42 | 0.52 | |
| LLM-PPBackbone=Mistral2023.10 | 0.73 | 0.22 | |
| Neuron-wise MoSType=Baseline2023.10 | 0.87 | 0.79 | |
| Layer-wise MoSType=Baseline2023.10 | 0.97 | 0.56 | |
| Supernet (Sandwich)Type=Baseline2023.10 | 1.05 | 0.81 | |
| HATType=Baseline2023.10 | 1.14 | 0.71 |