update metrics and citations
Browse files
README.md
CHANGED
@@ -65,6 +65,36 @@ SparseEncoder(
|
|
65 |
)
|
66 |
```
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
## Usage
|
69 |
|
70 |
### Direct Usage (Sentence Transformers)
|
@@ -324,6 +354,40 @@ You can finetune this model on your own dataset.
|
|
324 |
}
|
325 |
```
|
326 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
327 |
#### SpladeLoss
|
328 |
```bibtex
|
329 |
@misc{formal2022distillationhardnegativesampling,
|
|
|
65 |
)
|
66 |
```
|
67 |
|
68 |
+
### Metrics
|
69 |
+
```
|
70 |
+
{
|
71 |
+
"NDCG": {
|
72 |
+
"NDCG@2": 0.90484,
|
73 |
+
"NDCG@10": 0.91822,
|
74 |
+
"NDCG@20": 0.9204,
|
75 |
+
"NDCG@100": 0.92605
|
76 |
+
},
|
77 |
+
"MAP": {
|
78 |
+
"MAP@2": 0.90125,
|
79 |
+
"MAP@10": 0.91146,
|
80 |
+
"MAP@20": 0.91216,
|
81 |
+
"MAP@100": 0.91316
|
82 |
+
},
|
83 |
+
"Recall": {
|
84 |
+
"Recall@2": 0.9045,
|
85 |
+
"Recall@10": 0.931,
|
86 |
+
"Recall@20": 0.938,
|
87 |
+
"Recall@100": 0.963
|
88 |
+
},
|
89 |
+
"Precision": {
|
90 |
+
"P@2": 0.9045,
|
91 |
+
"P@10": 0.1862,
|
92 |
+
"P@20": 0.0938,
|
93 |
+
"P@100": 0.01926
|
94 |
+
}
|
95 |
+
}
|
96 |
+
```
|
97 |
+
|
98 |
## Usage
|
99 |
|
100 |
### Direct Usage (Sentence Transformers)
|
|
|
354 |
}
|
355 |
```
|
356 |
|
357 |
+
#### LIMIT
|
358 |
+
@misc{weller2025theoreticallimit,
|
359 |
+
title={On the Theoretical Limitations of Embedding-Based Retrieval},
|
360 |
+
author={Orion Weller and Michael Boratko and Iftekhar Naim and Jinhyuk Lee},
|
361 |
+
year={2025},
|
362 |
+
eprint={2508.21038},
|
363 |
+
archivePrefix={arXiv},
|
364 |
+
primaryClass={cs.IR},
|
365 |
+
url={https://arxiv.org/abs/2508.21038},
|
366 |
+
}
|
367 |
+
|
368 |
+
#### OpenSearch Models
|
369 |
+
@inproceedings{Shen_2025, series={SIGIR ’25},
|
370 |
+
title={Exploring $\ell_0$ parsification for Inference-free Sparse Retrievers},
|
371 |
+
url={http://dx.doi.org/10.1145/3726302.3730192},
|
372 |
+
DOI={10.1145/3726302.3730192},
|
373 |
+
booktitle={Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval},
|
374 |
+
publisher={ACM},
|
375 |
+
author={Shen, Xinjie and Geng, Zhichao and Yang, Yang},
|
376 |
+
year={2025},
|
377 |
+
month=jul, pages={2572–2576},
|
378 |
+
collection={SIGIR ’25}
|
379 |
+
}
|
380 |
+
|
381 |
+
@misc{geng2025competitivesearchrelevanceinferencefree,
|
382 |
+
title={Towards Competitive Search Relevance For Inference-Free Learned Sparse Retrievers},
|
383 |
+
author={Zhichao Geng and Yiwen Wang and Dongyu Ru and Yang Yang},
|
384 |
+
year={2025},
|
385 |
+
eprint={2411.04403},
|
386 |
+
archivePrefix={arXiv},
|
387 |
+
primaryClass={cs.IR},
|
388 |
+
url={https://arxiv.org/abs/2411.04403},
|
389 |
+
}
|
390 |
+
|
391 |
#### SpladeLoss
|
392 |
```bibtex
|
393 |
@misc{formal2022distillationhardnegativesampling,
|