Multitask Language Understanding on MMLU (Accuracy and AVERAGE MEAN)
50.12AccuracyFull-data Fine-tuning
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Full-data Fine-tuningBase Model=LLAMA-2-7B, Coreset size=100%, Number of seeds=32025.10 | 50.12 | 48.99 | |
| TRIMBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 49.33 | 48.56 | |
| LESSBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 49.23 | 48.27 | |
| TAGCOSBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 48.12 | 46.6 | |
| S2LBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 46.7 | 45.56 | |
| CLDBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 46.13 | 42.61 | |
| BM25Base Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 46.12 | 45.41 | |
| RandomBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 45.84 | 45.14 | |
| DSIRBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 45.73 | 42.02 | |
| Pretrained (no Fine-tuning)Base Model=LLAMA-2-7B, Number of seeds=32025.10 | 45.6 | 43.43 | |
| RDSBase Model=LLAMA-2-7B, Coreset size=5%, Number of seeds=32025.10 | 45.27 | 42.17 |