Use fine-tuned model

README.md

@@ -10,10 +10,13 @@ Finetuned version of [FRED-T5 large 820M](https://huggingface.co/ai-forever/FRED
 
 Code [repo](https://github.com/saarus72/text_normalization/tree/dev).
 
-Trained on [ficbook](https://huggingface.co/datasets/IlyaGusev/ficbook), [librusec](https://huggingface.co/datasets/IlyaGusev/librusec) and [pikabu](https://huggingface.co/datasets/IlyaGusev/pikabu) sentences, inverse text normalized using [NeMo Text Processing](https://github.com/NVIDIA/NeMo-text-processing). Haven't trained anything yet but number normalization.
+Trained on [ficbook](https://huggingface.co/datasets/IlyaGusev/ficbook), [librusec](https://huggingface.co/datasets/IlyaGusev/librusec) and [pikabu](https://huggingface.co/datasets/IlyaGusev/pikabu) sentences, inverse text normalized using modified version of [Russian_w2n](https://github.com/Oknolaz/Russian_w2n) package. Haven't trained anything yet but number normalization.
 
 ## Usage
 
+Put numbers inside of squared brackets and follow with T5 tokens `<extra_id_0>`, `<extra_id_1>` and so on.
+One is advised to split numbers on groups by 3 digits always and do not put anything else inside of `[ ]` brackets.
+
 ```python
 import torch
 from transformers import GPT2Tokenizer, T5ForConditionalGeneration 
@@ -22,8 +25,22 @@ from transformers import GPT2Tokenizer, T5ForConditionalGeneration
 device='cuda'
 tokenizer = GPT2Tokenizer.from_pretrained('saarus72/russian_text_normalizer', eos_token='</s>')
 model = T5ForConditionalGeneration.from_pretrained('saarus72/russian_text_normalizer').to(device)
-lm_text = '<SC1>Было у отца [3]<extra_id_0> сына, но не было даже [2-3]<extra_id_1> пиджаков с блёстками за [142 990 руб]<extra_id_2>.'
+lm_text = '<SC1>Было у отца [3]<extra_id_0> сына, но не было даже [2-3]<extra_id_1> пиджаков с блёстками за [142 990]<extra_id_2> руб.'
 input_ids = torch.tensor([tokenizer.encode(lm_text)]).to(device)
 outputs = model.generate(input_ids, eos_token_id=tokenizer.eos_token_id, early_stopping=True)
 print(tokenizer.decode(outputs[0][1:]))
-# <extra_id_0>  три  <extra_id_1>  двух-трех  <extra_id_2>  сто сорок две тысячи девятьсот рублей </s>
+# '<extra_id_0>  три\n <extra_id_1>  двух- трёх\n <extra_id_2>  сто сорок две тысячи девятьсот девяносто\n</s>'
+```
+
+### More examples
+
+| Model input | After replacement | Is correct |
+|---|---|---|
+| `<SC1>Временами я думаю, какое применение найти тем [14 697]<extra_id_0> рублям, что лежат уже больше [33]<extra_id_1> лет?` | `Временами я думаю, какое применение найти тем четырнадцати тысячам шестистам девяносто семи рублям, что лежат уже больше тридцати трёх лет?` | Yes |
+| `<SC1>я купил iphone [12]<extra_id_0> за [142 990]<extra_id_1> руб без [3-x]<extra_id_2> часов полдень и т.д.` | `я купил iphone двенадцатый за сто сорок две тысячи девятьсот девяносто руб без трех часов полдень и т.д.?` | Yes |
+| `<SC1>В школе у меня одни [5]<extra_id_0>.` | `В школе у меня одни пятые.` | **No (`пятёрки` is correct)** |
+| `<SC1>Было у отца [3]<extra_id_0> сына. Старшему было [35]<extra_id_1>, среднему - не меньше [33]<extra_id_2>, а младший на [4]<extra_id_3> младше всех. Бывает.` | `Было у отца три сына. Старшему было тридцать пять, среднему - не меньше тридцати трех, а младший на четыре младше всех. Бывает.` | Yes |
+| `<SC1>я подарю тебе [1 000 000]<extra_id_0> алых роз, я научу тебя смеяться` | `я подарю тебе миллион алых роз, я научу тебя смеяться` | Yes |
+| `<SC1>запомни телефон — [8 9 555 35 35]<extra_id_0>, проще позвонить, чем занимать` | `запомни телефон — восемь девять пятьсот пятьдесят пять тридцать пять тридцать пять, проще позвонить, чем занимать` | Yes |
+| `<SC1>стоимость айфон [32]<extra_id_0> — всего [189 553 600]<extra_id_1> руб.!` | `стоимость айфон тридцать два — всего сто восемьдесят девять миллионов пятьсот пятьдесят три тысячи шестьсот руб.!` | Yes |
+| `<SC1>[91 957 552 336]<extra_id_0> минус [3 474 354 354]<extra_id_1> равно ?` | `Девяносто девять миллиардов пятьдесят семь миллионов пятьсот пятьдесят два миллиона трис��а тридцать шесть минус три миллиарда четыреста семьдесят четыре тысячи триста пятьдесят четыре тысячи триста пятьдесят четыре равно ?` | **No :(** |

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ca16862297381f44bf8ce00d1fbb188f06aac895d5b5af4d2b7fbe7e3be91893
+oid sha256:02ba17fe51850c19c2bbccb9604eea848bd9ef1f9cd67fdb7e3d3de6d64a8bdd
 size 3282183426