Instruction Following on AlpacaEval Leaderboard (Mixtral 8x7B vs Mixtral 8x22B reference)
44.28LC Win RateTRLM-FoBa (Rev)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TRLM-FoBa (Rev)Inference Style=Response -> Query2024.12 | 44.28 | 46.67 | 45.71 | 1.5 | 368 | 437 | 0 | |
| TRLM-BaInference Style=Response -> Query2024.12 | 44.13 | 46.98 | 47.39 | 1.52 | 381 | 423 | 1 | |
| SelfInference Style=Query -> Response2024.12 | 43.56 | 41.88 | 42.11 | 1.52 | 339 | 466 | 0 | |
| TRLM-FoBa (Forw)Inference Style=Response -> Query2024.12 | 42.88 | 47.11 | 46.58 | 1.52 | 375 | 430 | 0 | |
| TRLM-FoInference Style=Response -> Query2024.12 | 42.07 | 47.54 | 47.08 | 1.51 | 379 | 426 | 0 | |
| Forward BaselineInference Style=Query -> Response2024.12 | 40.11 | 43.85 | 42.92 | 1.52 | 345 | 459 | 1 |