File size: 1,155 Bytes
11ec98e
b2bb218
 
 
11ec98e
b2bb218
11ec98e
 
ee370ff
b2bb218
 
 
 
c79e04b
b2bb218
 
 
 
 
 
c79e04b
b2bb218
c79e04b
b2bb218
c79e04b
b2bb218
c79e04b
b2bb218
c79e04b
b2bb218
c79e04b
b2bb218
c79e04b
 
 
 
 
 
 
 
 
b2bb218
 
 
c79e04b
 
 
 
 
 
 
b2bb218
 
 
 
c79e04b
 
 
b2bb218
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
---
language: 
- es
thumbnail: 
tags:
- summarization
- mt5
- spanish
license: apache-2.0
datasets:
- larazonpublico
- es
metrics:
- rouge
---

# mt5-small-spanish-summarization

## Model description

This is a mt5-small model finetuned for generating headlines from the body of the news in Spanish. 

## Training data

The model was trained with 58425 news extracted from the La Raz�n (31477) and P�blico (26948) newspapers. These news belong to the following categories: "Espa�a", "Cultura", "Econom�a", "Igualdad" and "Pol�tica".

## Training procedure

It was trained with Google Colab's GPU Tesla P100-PCIE-16GB for 2 epochs.

### Hyperparameters

{evaluation_strategy = "epoch",
learning_rate = 2e-4,
per_device_train_batch_size = 6,
per_device_eval_batch_size = 6,
weight_decay = 0.01,
save_total_limi t= 3,
num_train_epochs = 2,
predict_with_generate = True,
fp16 = False}


## Eval results
| metric | score |
| --- | ----- |
| rouge1 | 44.03 |
| rouge2 | 28.2900 |
| rougeL | 40.54 |
| rougeLsum | 40.5587 |


### BibTeX entry and citation info

```bibtex
@inproceedings{ mt5lrpjosmunpen,
  year={2020},
  author = {Jos� Manuel Mu�iz Pe�a},
}
```