sage-v1.1.0
Summary
The model corrects spelling and punctuation errors and typos by bringing all the words in the text to the norm of the Russian language. Corrector had been trained based on the model FRED-T5-1.7B. An extensive dataset with “artificial” errors was taken as a training corpus: the corpus was assembled on the basis of the Russian-language Wikipedia and transcripts of Russian-language videos, then typos and spelling errors were automatically introduced into it using the library SAGE.
Public references
- SAGE library announcement, DataFest 2023
- Paper about synthetic error generation methods, Dialogue 2023
- SAGE EACL 2024 paper
Examples
Input | Output |
---|---|
И не чсно прохожим в этот день непогожйи почему я веселый такйо | И не ясно прохожим в этот день непогожий, почему я веселый такой. |
Каждй день воттак делой, и спена балеть нибудет. А вотак каждый день ниделай | Каждый день вот так делай и спина болеть не будет. А вот так каждый день не делай. |
Основая цель мероприятия практическая отработка навыков по оказанию помощи гражданам, попавшим в ДТП а также повышение и совершенствование уровня профессиональной подготовки сотрудников МЧС при проведении аварийно-спасательных работ по ликвидации последствий дорожно-транспортных проишествий сокращение временных показателей реагирования. | Основная цель мероприятия — практическая отработка навыков по оказанию помощи гражданам, попавшим в ДТП, а также повышение и совершенствование уровня профессиональной подготовки сотрудников МЧС при проведении аварийно-спасательных работ по ликвидации последствий дорожно-транспортных происшествий, сокращение временных показателей реагирования |
Metrics
Quality
Below are automatic metrics for determining the correctness of the spell checkers. We compare our solution with both open automatic spell checkers and the ChatGPT family of models on all four available datasets:
- RUSpellRU: texts collected from (LiveJournal), with manually corrected typos and errors;
- MultidomainGold: examples from 7 text sources, including the open web, news, social media, reviews, subtitles, policy documents and literary works;
- MedSpellChecker: texts with errors from medical anamnesis;
- GitHubTypoCorpusRu: spelling errors and typos in commits from GitHub;
RUSpellRU
Model | Pr. (spell) | Rec. (spell) | F1 (spell) | Pr. (punc) | Rec. (punc) | F1 (punc) | Pr. (case) | Rec. (case) | F1 (case) |
---|---|---|---|---|---|---|---|---|---|
sage-v1.1.0 | 90.3 | 86.3 | 88.2 | 90.3 | 86.6 | 88.4 | 95.2 | 95.9 | 95.6 |
sage-fredt5-large | 57.3 | 68.0 | 62.2 | 86.7 | 46.1 | 60.2 | 92.1 | 67.8 | 78.1 |
sage-fredt5-large (ft) | 88.4 | 80.9 | 84.5 | 88.2 | 85.3 | 86.8 | 95.5 | 94.0 | 94.7 |
gpt-3.5-turbo | 33.6 | 58.5 | 42.7 | 85.9 | 64.6 | 73.7 | 84.9 | 73.9 | 79.0 |
gpt-4 | 54.9 | 76.7 | 64.0 | 84.0 | 82.3 | 83.2 | 91.5 | 90.2 | 90.9 |
MultidomainGold
Model | Pr. (spell) | Rec. (spell) | F1 (spell) | Pr. (punc) | Rec. (punc) | F1 (punc) | Pr. (case) | Rec. (case) | F1 (case) |
---|---|---|---|---|---|---|---|---|---|
sage-v1.1.0 | 81.6 | 77.7 | 79.6 | 70.2 | 67.5 | 68.8 | 80.5 | 80.5 | 80.5 |
sage-fredt5-large | 43.4 | 49.7 | 46.3 | 21.8 | 21.3 | 21.6 | 58.8 | 23.9 | 34.0 |
sage-fredt5-large (ft) | 80.3 | 75.1 | 77.6 | 69.0 | 66.5 | 67.7 | 78.6 | 80.0 | 79.3 |
gpt-3.5-turbo | 18.8 | 48.1 | 27.1 | 42.0 | 31.8 | 36.2 | 47.1 | 51.3 | 49.1 |
gpt-4 | 25.4 | 68.0 | 37.0 | 57.8 | 54.3 | 56.0 | 54.0 | 67.5 | 60.0 |
MedSpellChecker
Model | Pr. (spell) | Rec. (spell) | F1 (spell) | Pr. (punc) | Rec. (punc) | F1 (punc) | Pr. (case) | Rec. (case) | F1 (case) |
---|---|---|---|---|---|---|---|---|---|
sage-v1.1.0 | 71.3 | 73.5 | 72.4 | 75.1 | 69.2 | 72.0 | 80.9 | 72.8 | 76.6 |
sage-fredt5-large | 35.2 | 54.5 | 42.8 | 19.2 | 13.2 | 15.7 | 48.7 | 36.8 | 41.9 |
sage-fredt5-large (ft) | 72.5 | 72.2 | 72.3 | 74.6 | 66.4 | 70.3 | 79.3 | 85.1 | 82.1 |
gpt-3.5-turbo | 14.7 | 45.9 | 22.3 | 69.9 | 52.3 | 59.8 | 26.4 | 41.8 | 32.3 |
gpt-4 | 37.8 | 72.3 | 49.6 | 81.4 | 64.3 | 71.9 | 73.0 | 62.1 | 67.1 |
GitHubTypoCorpusRu
Model | Pr. (spell) | Rec. (spell) | F1 (spell) | Pr. (punc) | Rec. (punc) | F1 (punc) | Pr. (case) | Rec. (case) | F1 (case) |
---|---|---|---|---|---|---|---|---|---|
sage-v1.1.0 | 70.8 | 56.3 | 62.7 | 48.9 | 35.8 | 41.4 | 32.9 | 45.3 | 38.1 |
sage-fredt5-large | 46.0 | 46.6 | 46.3 | 22.7 | 18.3 | 20.2 | 12.0 | 13.2 | 12.6 |
sage-fredt5-large (ft) | 67.5 | 53.2 | 59.5 | 48.5 | 38.0 | 42.6 | 37.3 | 50.0 | 42.7 |
gpt-3.5-turbo | 23.7 | 38.7 | 29.4 | 37.6 | 23.3 | 28.7 | 19.6 | 35.9 | 25.3 |
gpt-4 | 27.0 | 52.8 | 35.7 | 45.9 | 32.6 | 38.2 | 25.7 | 36.8 | 30.2 |
How to use
import re
import torch
from transformers import AutoTokenizer, T5ForConditionalGeneration
tokenizer = AutoTokenizer.from_pretrained("ai-forever/FRED-T5-1.7B")
model = T5ForConditionalGeneration.from_pretrained("ai-forever/sage-v1.1.0")
model.to('cuda')
tokenizer_config = {
'max_length': None,
'padding': 'longest',
'truncation': False,
"return_tensors": "pt",
}
def inference(sentence):
text = "<LM>" + sentence
with torch.inference_mode():
encodings = tokenizer(text, **tokenizer_config)
for k, v in encodings.items():
encodings[k] = v.to('cuda:0')
res = model.generate(
**encodings,
use_cache=True,
max_length = encodings['input_ids'].size(1) * 1.5
)
res = res.cpu().tolist()
res = tokenizer.batch_decode(res, skip_special_tokens=True)
return res
text = 'Првет какдила'
text = re.sub(r'\n+', '\n', text)
print(inference(text))
# ['Привет, как дела?']
Resources
- SAGE library, GitHub
- sage-fredt5-large, HuggingFace
- sage-fredt5-distilled-95m, HuggingFace
- sage-m2m100-1.2B, HuggingFace
- sage-mt5-large, HuggingFace
Specifications
- File size: 7 Gb;
- Framework: pytorch
- Version: v1.1.0
- Developer: SberDevices, AGI NLP
Contacts
- Downloads last month
- 57
Evaluation results
- F1 (spell) on RUSpellRU (spell&punct)self-reported88.200
- F1 (punct) on RUSpellRU (spell&punct)self-reported88.400
- F1 (case) on RUSpellRU (spell&punct)self-reported95.600
- F1 (spell) on RUSpellRU (spell&punct)self-reported79.600
- F1 (punct) on RUSpellRU (spell&punct)self-reported68.800
- F1 (case) on RUSpellRU (spell&punct)self-reported80.500
- F1 (spell) on RUSpellRU (spell&punct)self-reported72.400
- F1 (punct) on RUSpellRU (spell&punct)self-reported72.000
- F1 (case) on RUSpellRU (spell&punct)self-reported76.600
- F1 (spell) on RUSpellRU (spell&punct)self-reported62.700