Continual Knowledge Learning on LAMA CKL
45Top-1 AccuracyTAALM(QLORA) + Mix-review
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TAALM(QLORA) + Mix-reviewBase Model=Llama2-7B, Epoch=32024.07 | 45 | 90.12 | 1.3512 | |
| TAALM(QLORA) + RecAdamBase Model=Llama2-7B, Epoch=72024.07 | 43.6 | 89.82 | 1.3342 | |
| TAALM(QLORA)Base Model=Llama2-7B, Epoch=42024.07 | 42.9 | 89.83 | 1.3273 | |
| TAALM + K-AdapterBase Model=Llama2-7B, Epoch=212024.07 | 33.2 | 97.47 | 1.3067 | |
| TAALM + K-Adapter + Mix-review + RecAdamBase Model=Llama2-7B, Epoch=252024.07 | 31.4 | 96.77 | 1.2817 | |
| TAALM + K-Adapter + RecAdamBase Model=Llama2-7B, Epoch=132024.07 | 26.4 | 97.3 | 1.237 |