Utente commited on
Commit
64a6538
2 Parent(s): c9376d1 5b97e74

Merge branch 'main' of https://huggingface.co/aiknowyou/mt5-base-it-paraphraser

Browse files
Files changed (1) hide show
  1. README.md +81 -0
README.md CHANGED
@@ -1,3 +1,84 @@
1
  ---
 
 
2
  license: cc-by-nc-sa-4.0
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language: it
3
+ datasets: tapaco
4
  license: cc-by-nc-sa-4.0
5
+ tags:
6
+ - mt5
7
+ - paraphrase-generation
8
+ - paraphrasing
9
+
10
  ---
11
+
12
+ # MT5-base fine-tuned on Tapaco dataset for Paraphrasing
13
+
14
+ MT5-base Italian paraphraser fine-tuned on [TaPaCo](https://huggingface.co/datasets/tapaco) dataset.
15
+
16
+ ## Details of MT5
17
+
18
+ The **MT5** model was presented in [mT5: A massively multilingual pre-trained text-to-text transformer](https://arxiv.org/abs/2010.11934) by *Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, Colin Raffel* in 2020. Here the abstract:
19
+
20
+ The recent "Text-to-Text Transfer Transformer" (T5) leveraged a unified text-to-text format and scale to attain state-of-the-art results on a wide variety of English-language NLP tasks. In this paper, we introduce mT5, a multilingual variant of T5 that was pre-trained on a new Common Crawl-based dataset covering 101 languages. We detail the design and modified training of mT5 and demonstrate its state-of-the-art performance on many multilingual benchmarks. We also describe a simple technique to prevent "accidental translation" in the zero-shot setting, where a generative model chooses to (partially) translate its prediction into the wrong language. All of the code and model checkpoints used in this work are publicly available.
21
+
22
+ ## Model fine-tuning
23
+
24
+ The training script is a slightly modified version of this [Colab notebook](https://colab.research.google.com/drive/1DGeF190gJ3DjRFQiwhFuZalp427iqJNQ) after having prepared an adapted italian version of mt5 model by following this other [Colab notebook](https://gist.github.com/avidale/44cd35bfcdaf8bedf51d97c468cc8001)
25
+
26
+ ## Model in Action
27
+
28
+ ```python
29
+ from transformers import T5ForConditionalGeneration, T5Tokenizer
30
+ import torch
31
+
32
+ raw_model = 'aiknowyou/mt5-base-it-paraphraser'
33
+
34
+ # Model and Tokenizer definition #
35
+ model = T5ForConditionalGeneration.from_pretrained(raw_model)
36
+ tokenizer = T5Tokenizer.from_pretrained(raw_model)
37
+
38
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
39
+ max_size = 10000
40
+
41
+ def get_paraphrased_sentences(model, tokenizer, sentence, num_return_sequences=5, num_beams=5):
42
+ # tokenize the text to be form of a list of token IDs
43
+ inputs = tokenizer([sentence], truncation=True, padding="longest", return_tensors="pt").to(model.device)
44
+ # generate the paraphrased sentences
45
+ outputs = model.generate(
46
+ **inputs,
47
+ num_beams=num_beams,
48
+ num_return_sequences=num_return_sequences,
49
+ max_length=max_size
50
+ )
51
+ # decode the generated sentences using the tokenizer to get them back to text
52
+ return tokenizer.batch_decode(outputs, skip_special_tokens=True)
53
+
54
+ # sentence = "Vorrei chiedervi la procedura per recuperare la chiave di accesso al mio profilo personale. L'ho persa e vorrei recuperarla."
55
+ # get_paraphrased_sentences(model, tokenizer, sentence, num_beams=100, num_return_sequences=5)
56
+
57
+ ```
58
+
59
+ ## Output
60
+ ```
61
+ Original Question ::
62
+ "Vorrei chiedervi la procedura per recuperare la chiave di accesso al mio profilo personale. L'ho persa e vorrei recuperarla."
63
+
64
+ Paraphrased Questions ::
65
+ 'Vorrei chiedervi la procedura per recuperare la chiave di accesso al mio profilo personale.',
66
+ 'Vorrei chiederti la procedura per recuperare la chiave di accesso al mio profilo personale.',
67
+ 'Vorrei chiedervi la procedura per restituire la chiave di accesso al mio profilo personale.',
68
+ 'Vorrei chiedere la procedura per recuperare la chiave di accesso al mio profilo personale.',
69
+ 'Vorrei chiedervi la procedura per recuperare la chiave di accesso al mio profilo personale. L'ho persa e vorrei recuperarla.'
70
+ ```
71
+
72
+ ## Contribution
73
+
74
+ Thanks to [@tradicio](https://huggingface.co/tradicio) for adding this model.
75
+
76
+ ## License
77
+
78
+ This work is licensed under a
79
+ [Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License][cc-by-nc-sa].
80
+
81
+ [![CC BY-NC-SA 4.0][cc-by-nc-sa-image]][cc-by-nc-sa]
82
+
83
+ [cc-by-nc-sa]: http://creativecommons.org/licenses/by-nc-sa/4.0/
84
+ [cc-by-nc-sa-image]: https://licensebuttons.net/l/by-nc-sa/4.0/88x31.png