General Reasoning and Coding on TriviaQA Non-Factoid
31.9GSM8K ScoreREMIX (K-Pile / -)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| REMIX (K-Pile / -)Backbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=K-Pile / -2024.11 | 31.9 | 82.9 | 93.5 | 90.7 | 90.1 | 77.8 | |
| REMIX (Random / -)Backbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=Random / -2024.11 | 27.1 | 84.9 | 71.2 | 87.3 | 70.8 | 68.3 | |
| REMIX (Random / K-Pile)Backbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=Random / K-Pile2024.11 | 12.5 | 81.8 | 71.2 | 74.6 | 70 | 62 | |
| No MixingBackbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=No Mixing2024.11 | 9.4 | 87.6 | 54.4 | 70.4 | 67.6 | 57.9 | |
| REMIX (- / Random)Backbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=- / Random2024.11 | 4.1 | 81 | 84 | 62.2 | 71.6 | 60.6 | |
| REMIX (- / K-Pile)Backbone=Llama-3-8B, Mixing Strategy (DM1 / DM2)=- / K-Pile2024.11 | 2.8 | 79.1 | 75.9 | 53.7 | 69.8 | 56.3 |