|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- papluca/language-identification |
|
language: |
|
- en |
|
- de |
|
- fr |
|
- es |
|
metrics: |
|
- precision |
|
- recall |
|
- f1 |
|
- accuracy |
|
pipeline_tag: text-classification |
|
--- |
|
# German, English, French and Spanish Language Detector |
|
|
|
The GEFS-language-detector model outperformed by achieving an impressive F1 score close to 100%. This result significantly exceeds typical benchmarks and underscores the model's accuracy and reliability in identifying languages. |
|
This is a fined tuned model by using the dataset of papluca [Language Identification](https://huggingface.co/datasets/papluca/language-identification#additional-information) and the base model [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) . |
|
|
|
|
|
## Predicted output: |
|
|
|
Model will return the language detection in the language codes like: |
|
``` |
|
- de as German |
|
- en as English |
|
- fr as French |
|
- es as Spanish |
|
``` |
|
|
|
## Supported languages |
|
Currently this model support 4 languages but in future more languages will be added. |
|
|
|
Following languages supported by the model: |
|
- German (de) |
|
- English (en) |
|
- French (fr) |
|
- Spanish (es) |
|
|
|
# Use a pipeline as a high-level helper |
|
|
|
```python |
|
from transformers import pipeline |
|
|
|
text=["Mir gefällt die Art und Weise, Sprachen zu erkennen", |
|
"I like the way to detect languages", |
|
"Me gusta la forma de detectar idiomas", |
|
"J'aime la façon de détecter les langues"] |
|
pipe = pipeline("text-classification", model="ImranzamanML/GEFS-language-detector") |
|
lang_detect=pipe(text, top_k=1) |
|
print("The detected language is", lang_detect) |
|
``` |
|
|
|
# Load model directly |
|
|
|
```python |
|
from transformers import AutoTokenizer, AutoModelForSequenceClassification |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("ImranzamanML/GEFS-language-detector") |
|
model = AutoModelForSequenceClassification.from_pretrained("ImranzamanML/GEFS-language-detector") |
|
|
|
``` |
|
|
|
## Model Training |
|
|
|
Epoch Training Loss Validation Loss |
|
1 0.002600 0.000148 |
|
2 0.001000 0.000015 |
|
3 0.000000 0.000011 |
|
4 0.001800 0.000009 |
|
5 0.002700 0.000016 |
|
6 0.001600 0.000012 |
|
7 0.001300 0.000009 |
|
8 0.001200 0.000008 |
|
9 0.000900 0.000007 |
|
10 0.000900 0.000007 |
|
|
|
|
|
## Testing Results |
|
``` |
|
Language Precision Recall F1 Accuracy |
|
de 0.9997 0.9998 0.9998 0.9999 |
|
en 1.0000 1.0000 1.0000 1.0000 |
|
fr 0.9995 0.9996 0.9996 0.9996 |
|
es 0.9994 0.9996 0.9995 0.9996 |
|
``` |
|
|
|
|
|
|
|
## About Author |
|
|
|
**Name**: Muhammad Imran Zaman |
|
**Company**: [Theum AG](https://theum.com/en/index.htm?t=) |
|
**Role**: Lead Machine Learning Engineer |
|
|
|
**Professional Links**: |
|
- Kaggle: [Profile](https://www.kaggle.com/muhammadimran112233) |
|
- LinkedIn: [Profile](linkedin.com/in/muhammad-imran-zaman) |
|
- Google Scholar: [Profile](https://scholar.google.com/citations?user=ulVFpy8AAAAJ&hl=en) |
|
- YouTube: [Channel](https://www.youtube.com/@consolioo) |
|
- GitHub: [Channel](https://github.com/Imran-ml) |
|
|
