# bert-tiny-stage2-sbert

SentenceTransformer checkpoint fine-tuned for Vietnamese legal retrieval.

## Evaluation

- Dataset: `another-symato/VMTEB-Zalo-legel-retrieval-wseg`
- Split: `train`
- Truncate dims: `[384, 256, 128, 64]`

|   dim |   Accuracy@1 |   Accuracy@3 |   Accuracy@5 |   Accuracy@10 |   NDCG@3 |   NDCG@5 |   NDCG@10 |    MRR@3 |    MRR@5 |   MRR@10 |   MAP@100 |
|-------|--------------|--------------|--------------|---------------|----------|----------|-----------|----------|----------|----------|-----------|
|   384 |     0.203443 |     0.340767 |     0.418232 |      0.524648 | 0.282964 | 0.314989 |  0.349523 | 0.263041 | 0.280882 | 0.295201 |  0.305581 |
|   256 |     0.193662 |     0.339202 |     0.409233 |      0.511737 | 0.277548 | 0.306474 |  0.339769 | 0.256325 | 0.272424 | 0.286249 |  0.296624 |
|   128 |     0.175665 |     0.321205 |     0.391236 |      0.482394 | 0.259807 | 0.288579 |  0.318217 | 0.238654 | 0.254577 | 0.266897 |  0.277913 |
|    64 |     0.14241  |     0.2723   |     0.351721 |      0.442879 | 0.216473 | 0.249186 |  0.278849 | 0.197314 | 0.215467 | 0.227817 |  0.238123 |