Merge branch 'main' of https://huggingface.co/yazge/turkish-colbert-onnx
Browse files
README.md
CHANGED
|
@@ -1,23 +1,28 @@
|
|
| 1 |
---
|
| 2 |
tags:
|
| 3 |
-
- Turkish
|
| 4 |
-
- turkish
|
| 5 |
-
- passage-retrieval
|
| 6 |
license: mit
|
| 7 |
language:
|
| 8 |
-
- tr
|
| 9 |
base_model: ytu-ce-cosmos/turkish-base-bert-uncased
|
| 10 |
---
|
|
|
|
| 11 |
# Turkish-ColBERT
|
|
|
|
| 12 |
This is a Turkish passage retrieval model based on the [ColBERT](https://doi.org/10.48550/arXiv.2112.01488) architecture.
|
| 13 |
|
| 14 |
The [Cosmos Turkish Base BERT](https://huggingface.co/ytu-ce-cosmos/turkish-base-bert-uncased) model was fine-tuned on 500k triplets (query, positive passage, negative passage) from a Turkish-translated version of the [MS MARCO dataset](https://huggingface.co/datasets/parsak/msmarco-tr).
|
| 15 |
|
| 16 |
#### ⚠ Uncased use requires manual lowercase conversion
|
| 17 |
-
|
|
|
|
|
|
|
| 18 |
```python
|
| 19 |
text.replace("I", "ı").lower()
|
| 20 |
```
|
|
|
|
| 21 |
This is due to a [known issue](https://github.com/huggingface/transformers/issues/6680) with the tokenizer.
|
| 22 |
|
| 23 |
## Example Usage
|
|
@@ -51,16 +56,19 @@ print(results[0]['content']) # "marie curie, radyoaktivite üzerine yaptığı
|
|
| 51 |
```
|
| 52 |
|
| 53 |
# Evaluation
|
| 54 |
-
|
| 55 |
-
|
|
| 56 |
-
|
|
| 57 |
-
| [
|
|
|
|
| 58 |
|
| 59 |
# Acknowledgments
|
|
|
|
| 60 |
- Research supported with Cloud TPUs from [Google's TensorFlow Research Cloud](https://sites.research.google/trc/about/) (TFRC). Thanks for providing access to the TFRC ❤️
|
| 61 |
- Thanks to the generous support from the Hugging Face team, it is possible to download models from their S3 storage 🤗
|
| 62 |
|
| 63 |
# Citations
|
|
|
|
| 64 |
```bibtex
|
| 65 |
@article{kesgin2023developing,
|
| 66 |
title={Developing and Evaluating Tiny to Medium-Sized Turkish BERT Models},
|
|
@@ -70,7 +78,9 @@ print(results[0]['content']) # "marie curie, radyoaktivite üzerine yaptığı
|
|
| 70 |
}
|
| 71 |
```
|
| 72 |
|
| 73 |
-
### Contact
|
| 74 |
-
|
|
|
|
| 75 |
https://cosmos.yildiz.edu.tr/ <br>
|
| 76 |
-
cosmos@yildiz.edu.tr <br>
|
|
|
|
|
|
| 1 |
---
|
| 2 |
tags:
|
| 3 |
+
- Turkish
|
| 4 |
+
- turkish
|
| 5 |
+
- passage-retrieval
|
| 6 |
license: mit
|
| 7 |
language:
|
| 8 |
+
- tr
|
| 9 |
base_model: ytu-ce-cosmos/turkish-base-bert-uncased
|
| 10 |
---
|
| 11 |
+
|
| 12 |
# Turkish-ColBERT
|
| 13 |
+
|
| 14 |
This is a Turkish passage retrieval model based on the [ColBERT](https://doi.org/10.48550/arXiv.2112.01488) architecture.
|
| 15 |
|
| 16 |
The [Cosmos Turkish Base BERT](https://huggingface.co/ytu-ce-cosmos/turkish-base-bert-uncased) model was fine-tuned on 500k triplets (query, positive passage, negative passage) from a Turkish-translated version of the [MS MARCO dataset](https://huggingface.co/datasets/parsak/msmarco-tr).
|
| 17 |
|
| 18 |
#### ⚠ Uncased use requires manual lowercase conversion
|
| 19 |
+
|
| 20 |
+
Convert your text to lower case as follows:
|
| 21 |
+
|
| 22 |
```python
|
| 23 |
text.replace("I", "ı").lower()
|
| 24 |
```
|
| 25 |
+
|
| 26 |
This is due to a [known issue](https://github.com/huggingface/transformers/issues/6680) with the tokenizer.
|
| 27 |
|
| 28 |
## Example Usage
|
|
|
|
| 56 |
```
|
| 57 |
|
| 58 |
# Evaluation
|
| 59 |
+
|
| 60 |
+
| Dataset | R@1 | R@5 | R@10 | MRR@10 |
|
| 61 |
+
| ------------------------------------------------------------------------ | ----- | ----- | ----- | ------ |
|
| 62 |
+
| [Scifact-tr](https://huggingface.co/datasets/AbdulkaderSaoud/scifact-tr) | 48.38 | 67.85 | 75.52 | 56.88 |
|
| 63 |
+
| [WikiRAG-TR](https://huggingface.co/datasets/Metin/WikiRAG-TR) | 31.21 | 75.63 | 79.63 | 49.08 |
|
| 64 |
|
| 65 |
# Acknowledgments
|
| 66 |
+
|
| 67 |
- Research supported with Cloud TPUs from [Google's TensorFlow Research Cloud](https://sites.research.google/trc/about/) (TFRC). Thanks for providing access to the TFRC ❤️
|
| 68 |
- Thanks to the generous support from the Hugging Face team, it is possible to download models from their S3 storage 🤗
|
| 69 |
|
| 70 |
# Citations
|
| 71 |
+
|
| 72 |
```bibtex
|
| 73 |
@article{kesgin2023developing,
|
| 74 |
title={Developing and Evaluating Tiny to Medium-Sized Turkish BERT Models},
|
|
|
|
| 78 |
}
|
| 79 |
```
|
| 80 |
|
| 81 |
+
### Contact
|
| 82 |
+
|
| 83 |
+
COSMOS AI Research Group, Yildiz Technical University Computer Engineering Department <br>
|
| 84 |
https://cosmos.yildiz.edu.tr/ <br>
|
| 85 |
+
cosmos@yildiz.edu.tr <br>
|
| 86 |
+
|