Continual Learning on 8-domain NLP benchmark
11.8Vanilla ForgettingVanilla
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VanillaScale=1B, Architecture=Transformer2026.04 | 11.8 | — | — | |
| VanillaScale=256M, Architecture=HOPE2026.04 | 12.6 | — | — | |
| VanillaScale=350M, Architecture=Transformer2026.04 | 13.1 | — | — | |
| VanillaScale=125M, Architecture=Transformer2026.04 | 14.2 | — | — | |
| VanillaScale=7B, Architecture=Llama2 (LoRA)2026.04 | 21.8 | — | — | |
| OGPScale=125M, Architecture=Transformer2026.04 | — | 10.4 | 26.8 | |
| OGPScale=256M, Architecture=HOPE2026.04 | — | 9.1 | 27.8 | |
| OGPScale=350M, Architecture=Transformer2026.04 | — | 9.3 | 29 | |
| OGPScale=1B, Architecture=Transformer2026.04 | — | 8.1 | 31.4 | |
| OGPScale=7B, Architecture=Llama2 (LoRA)2026.04 | — | 10.8 | 50.5 |