Dense Retrieval on NFCorpus
49.1NDCG@10OPERA Static Pruning (SP)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OPERA Static Pruning (SP)Method Strategy=Static Pruning, Backbone Model=bge-large-en-v1.52026.03 | 49.1 | — | 26.7 | |
| OPERA Dynamic Pruning (DP)Method Strategy=Dynamic Pruning, Backbone Model=bge-large-en-v1.52026.03 | 48 | — | 30.4 | |
| InfoBatch (IB)Method Strategy=InfoBatch, Backbone Model=bge-large-en-v1.52026.03 | 47 | — | 30.4 | |
| Fine-tuned (FT)Method Strategy=Fine-tuned, Backbone Model=bge-large-en-v1.52026.03 | 46.6 | — | 30 | |
| Pretrained (PT)Method Strategy=Pretrained, Backbone Model=bge-large-en-v1.52026.03 | 45.1 | — | 23.2 | |
| GPLBase Encoder=Tas-B, Adaptation Method=GPL2026.01 | — | 8.1 | — | |
| GPL + BPRBase Encoder=Tas-B, Adaptation Method=GPL + BPR2026.01 | — | -6.2 | — | |
| GPL + JPQBase Encoder=Tas-B, Adaptation Method=GPL + JPQ2026.01 | — | 1.5 | — | |
| PCABase Encoder=Tas-B, Adaptation Method=PCA2026.01 | — | 1.9 | — |