RichardErkhov
commited on
Commit
•
ed0f024
1
Parent(s):
96c9a48
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,97 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Quantization made by Richard Erkhov.
|
2 |
+
|
3 |
+
[Github](https://github.com/RichardErkhov)
|
4 |
+
|
5 |
+
[Discord](https://discord.gg/pvy7H8DZMG)
|
6 |
+
|
7 |
+
[Request more models](https://github.com/RichardErkhov/quant_request)
|
8 |
+
|
9 |
+
|
10 |
+
transformer-turkish-summarization - bnb 4bits
|
11 |
+
- Model creator: https://huggingface.co/mukayese/
|
12 |
+
- Original model: https://huggingface.co/mukayese/transformer-turkish-summarization/
|
13 |
+
|
14 |
+
|
15 |
+
|
16 |
+
|
17 |
+
Original model description:
|
18 |
+
---
|
19 |
+
datasets:
|
20 |
+
- mlsum
|
21 |
+
metrics:
|
22 |
+
- rouge
|
23 |
+
model-index:
|
24 |
+
- name: mukayese/transformer-turkish-summarization
|
25 |
+
results:
|
26 |
+
- task:
|
27 |
+
name: Summarization
|
28 |
+
type: summarization
|
29 |
+
dataset:
|
30 |
+
name: mlsum tu
|
31 |
+
type: mlsum
|
32 |
+
args: tu
|
33 |
+
metrics:
|
34 |
+
- name: Rouge1
|
35 |
+
type: rouge
|
36 |
+
value: 43.2049
|
37 |
+
license: mit
|
38 |
+
language:
|
39 |
+
- tr
|
40 |
+
pipeline_tag: summarization
|
41 |
+
---
|
42 |
+
|
43 |
+
# [Mukayese: Turkish NLP Strikes Back](https://arxiv.org/abs/2203.01215)
|
44 |
+
|
45 |
+
## Summarization: mukayese/transformer-turkish-summarization
|
46 |
+
|
47 |
+
_This model is uncased_, it was initialized from scratch and trained only the mlsum/tu dataset with no pre-training.
|
48 |
+
|
49 |
+
It achieves the following results on the evaluation set:
|
50 |
+
|
51 |
+
- Rouge1: 43.2049
|
52 |
+
- Rouge2: 30.7082
|
53 |
+
- Rougel: 38.1981
|
54 |
+
- Rougelsum: 39.9453
|
55 |
+
|
56 |
+
Check [this](https://arxiv.org/abs/2203.01215) paper for more details on the model and the dataset.
|
57 |
+
|
58 |
+
## Training procedure
|
59 |
+
|
60 |
+
### Training hyperparameters
|
61 |
+
|
62 |
+
The following hyperparameters were used during training:
|
63 |
+
- learning_rate: 0.0001
|
64 |
+
- train_batch_size: 4
|
65 |
+
- eval_batch_size: 8
|
66 |
+
- seed: 42
|
67 |
+
- distributed_type: multi-GPU
|
68 |
+
- num_devices: 8
|
69 |
+
- gradient_accumulation_steps: 2
|
70 |
+
- total_train_batch_size: 64
|
71 |
+
- total_eval_batch_size: 64
|
72 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
73 |
+
- lr_scheduler_type: linear
|
74 |
+
- num_epochs: 15.0
|
75 |
+
- mixed_precision_training: Native AMP
|
76 |
+
- label_smoothing_factor: 0.1
|
77 |
+
|
78 |
+
### Framework versions
|
79 |
+
|
80 |
+
- Transformers 4.11.3
|
81 |
+
- Pytorch 1.8.2+cu111
|
82 |
+
- Datasets 1.14.0
|
83 |
+
- Tokenizers 0.10.3
|
84 |
+
|
85 |
+
### Citation
|
86 |
+
|
87 |
+
```
|
88 |
+
@misc{safaya-etal-2022-mukayese,
|
89 |
+
title={Mukayese: Turkish NLP Strikes Back},
|
90 |
+
author={Ali Safaya and Emirhan Kurtuluş and Arda Göktoğan and Deniz Yuret},
|
91 |
+
year={2022},
|
92 |
+
eprint={2203.01215},
|
93 |
+
archivePrefix={arXiv},
|
94 |
+
primaryClass={cs.CL}
|
95 |
+
}
|
96 |
+
```
|
97 |
+
|