Update README.md
Browse files
README.md
CHANGED
|
@@ -13,21 +13,30 @@ pipeline_tag: sentence-similarity
|
|
| 13 |
|
| 14 |
# KoE5
|
| 15 |
|
| 16 |
-
|
|
|
|
|
|
|
| 17 |
|
| 18 |
-
|
| 19 |
|
|
|
|
| 20 |
|
| 21 |
-
|
| 22 |
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
|
|
|
|
|
|
| 25 |
First install the Sentence Transformers library:
|
| 26 |
|
| 27 |
```bash
|
| 28 |
pip install -U sentence-transformers
|
| 29 |
```
|
| 30 |
-
|
| 31 |
Then you can load this model and run inference.
|
| 32 |
```python
|
| 33 |
from sentence_transformers import SentenceTransformer
|
|
@@ -53,6 +62,33 @@ print(similarities)
|
|
| 53 |
# [0.3897, 0.3740, 1.0000]])
|
| 54 |
```
|
| 55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
## FAQ
|
| 57 |
|
| 58 |
**1. Do I need to add the prefix "query: " and "passage: " to input texts?**
|
|
@@ -69,7 +105,16 @@ Here are some rules of thumb:
|
|
| 69 |
## Citation
|
| 70 |
|
| 71 |
If you find our paper or models helpful, please consider cite as follows:
|
| 72 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
```
|
| 74 |
@article{wang2024multilingual,
|
| 75 |
title={Multilingual E5 Text Embeddings: A Technical Report},
|
|
|
|
| 13 |
|
| 14 |
# KoE5
|
| 15 |
|
| 16 |
+
Introducing KoE5, a model with advanced retrieval abilities.
|
| 17 |
+
It has shown remarkable performance in Korean text retrieval, speficially overwhelming most multilingual embedding models.
|
| 18 |
+
To our knowledge, It is one of the best publicly opened Korean retrieval models.
|
| 19 |
|
| 20 |
+
For details, visit the [KoE5 repository](https://github.com/nlpai-lab/KoE5)
|
| 21 |
|
| 22 |
+
### Model Description
|
| 23 |
|
| 24 |
+
This is the model card of a π€ transformers model that has been pushed on the Hub.
|
| 25 |
|
| 26 |
+
- **Developed by:** [NLP&AI Lab](http://nlp.korea.ac.kr/)
|
| 27 |
+
- **Language(s) (NLP):** Korean, English
|
| 28 |
+
- **License:** MIT
|
| 29 |
+
- **Finetuned from model:** [intfloat/multilingual-e5-large](https://huggingface.co/intfloat/multilingual-e5-large)
|
| 30 |
+
- **Finetuned dataset:** [ko-triplet-v1.0](nlpai-lab/ko-triplet-v1.0)
|
| 31 |
|
| 32 |
+
## Example code
|
| 33 |
+
### Install Dependencies
|
| 34 |
First install the Sentence Transformers library:
|
| 35 |
|
| 36 |
```bash
|
| 37 |
pip install -U sentence-transformers
|
| 38 |
```
|
| 39 |
+
### Python code
|
| 40 |
Then you can load this model and run inference.
|
| 41 |
```python
|
| 42 |
from sentence_transformers import SentenceTransformer
|
|
|
|
| 62 |
# [0.3897, 0.3740, 1.0000]])
|
| 63 |
```
|
| 64 |
|
| 65 |
+
## Training Details
|
| 66 |
+
|
| 67 |
+
### Training Data
|
| 68 |
+
|
| 69 |
+
- [ko-triplet-v1.0](nlpai-lab/ko-triplet-v1.0)
|
| 70 |
+
- Korean query-document-hard_negative data pair (open data)
|
| 71 |
+
- About 700000+ examples used totally
|
| 72 |
+
|
| 73 |
+
### Training Procedure
|
| 74 |
+
|
| 75 |
+
- **loss:** Used **[CachedMultipleNegativesRankingLoss](https://sbert.net/docs/package_reference/sentence_transformer/losses.html#cachedmultiplenegativesrankingloss)** by sentence-transformers
|
| 76 |
+
- **batch size:** 512
|
| 77 |
+
- **learning rate:** 1e-05
|
| 78 |
+
- **epochs:** 1
|
| 79 |
+
|
| 80 |
+
## Evaluation
|
| 81 |
+
### Metrics
|
| 82 |
+
- NDCG@1, F1@1, NDCG@3, F1@3
|
| 83 |
+
### Benchmark Datasets
|
| 84 |
+
- Ko-strategyQA
|
| 85 |
+
- AutoRAG-benchmark
|
| 86 |
+
- PublicHealthQA
|
| 87 |
+
|
| 88 |
+
## Results
|
| 89 |
+
|
| 90 |
+
|
| 91 |
+
|
| 92 |
## FAQ
|
| 93 |
|
| 94 |
**1. Do I need to add the prefix "query: " and "passage: " to input texts?**
|
|
|
|
| 105 |
## Citation
|
| 106 |
|
| 107 |
If you find our paper or models helpful, please consider cite as follows:
|
| 108 |
+
```text
|
| 109 |
+
@misc{KoE5,
|
| 110 |
+
author = {NLP & AI Lab and Human-Inspired AI research},
|
| 111 |
+
title = {KoE5: A New Dataset and Model for Improving Korean Embedding Performance},
|
| 112 |
+
year = {2024},
|
| 113 |
+
publisher = {Youngjoon Jang, Junyoung Son, Taemin Lee},
|
| 114 |
+
journal = {GitHub repository},
|
| 115 |
+
howpublished = {\url{https://github.com/nlpai-lab/KoE5}},
|
| 116 |
+
}
|
| 117 |
+
```
|
| 118 |
```
|
| 119 |
@article{wang2024multilingual,
|
| 120 |
title={Multilingual E5 Text Embeddings: A Technical Report},
|