Tokenizer Card for Ansh-256k!
The tokenizer model Ansh-256k - is trained on a dataset of Wikipedia articles in 22 Official Indic languages and English. We propose the name Ansh as this
tokenizer is designed to meticulously identify every essential token (Ansh in Sanskrit) of our diverse Indic languages. This model is the advanced
version of the Ansh-160k which was trained on 18 Indic languages and English.
Model Description
India is a vast country that has a multi-lingual culture that covers 22 Official languages and more than 1700 languages and dialects. It has been observed that various languages share words among themselves, sometimes even across language families. To capitalize on this observation, we trained our tokenization model with a vocabulary size of 256,000 (256k) using the dataset of Wikipedia articles and Sangraha dataset in 22 Indic languages and English by applying the Byte-Pair Encoding (BPE) algorithm. When compared among all the popular open-source tokenizers trained on multilingual Indic languages on fertility scores, our model outperformed them in 20 Indic languages.
- Developed by: Lingo Research Group at IIT Gandhinagar
- Language(s) (NLP): Multilingual (22 Indic Languages and English)
- License: Apache 2.0
How to Get Started with the Model ๐จ๐ปโ๐ป
Use the code below to get started with the model.
from transformers import AutoTokenizer
try:
tokenizer = tokenizer = AutoTokenizer.from_pretrained("LingoIITGN/Ansh-256k"))
print("Tokenizer loaded successfully!")
except Exception as e:
print(f"Error loading tokenizer: {e}")
print("Please ensure you have the correct model name and are connected to the internet.")
exit()
input_text = "Hello, world! This is an example of how to use the tokenizer."
#input_text = 'เคฎเฅเคเฅ เคฏเคน presentation เคเคฒ morning เคคเค submit เคเคฐเคจเคพ เคนเฅเฅค '
#input_text = 'What is capital city of India?'
encoded_input = tokenizer.encode(example_text)
print("\nOriginal Text:", example_text)
print("Encoded (Token IDs):", encoded_input)
decoded_output = tokenizer.decode(encoded_input)
print("Decoded Text:", decoded_output)
Evaluation
[More Information Needed]
Results ๐
Tokenizers Results
| Language | IndicBERTv2 | Sarvam-1 | MuRIL | Gemma-3 | Llama-3.1 | XLM-RoBERTa | NLLB | Ansh-160k |
|---|---|---|---|---|---|---|---|---|
| Tamil | 1.966 | 2.853 | 1.904 | 2.766 | 12.170 | 2.726 | 2.925 | 1.937 |
| Kannada | 2.035 | 2.651 | 1.992 | 3.498 | 15.302 | 2.835 | 2.955 | 1.876 |
| Malayalam | 2.202 | 3.246 | 2.199 | 3.571 | 15.215 | 2.999 | 3.329 | 2.073 |
| Maithili | 1.534 | 2.269 | 1.549 | 2.036 | 3.414 | 1.991 | 2.058 | 1.270 |
| Konkani | 2.145 | 2.954 | 2.469 | 2.830 | 4.180 | 2.746 | 2.765 | 1.741 |
| Telugu | 1.803 | 2.429 | 1.859 | 3.050 | 13.002 | 2.391 | 2.691 | 1.713 |
| Odia | 1.601 | 2.419 | 1.497 | 4.639 | 15.629 | 2.222 | 2.284 | 1.397 |
| Bengali | 1.610 | 2.083 | 1.555 | 1.890 | 8.389 | 2.374 | 2.396 | 1.515 |
| Nepali | 1.629 | 2.450 | 1.484 | 2.163 | 3.768 | 1.903 | 2.070 | 1.466 |
| Punjabi | 1.458 | 1.822 | 1.459 | 2.968 | 8.277 | 2.031 | 1.983 | 1.445 |
| Urdu | 1.565 | 9.004 | 1.402 | 1.984 | 3.153 | 1.582 | 1.807 | 1.383 |
| Hindi | 1.456 | 1.784 | 1.450 | 1.719 | 2.997 | 1.716 | 1.790 | 1.364 |
| Gujarati | 1.505 | 2.228 | 1.428 | 2.491 | 9.926 | 2.195 | 2.332 | 1.387 |
| Kashmiri | 2.722 | 9.237 | 2.220 | 3.204 | 4.119 | 3.155 | 2.966 | 1.528 |
| Marathi | 1.529 | 1.906 | 1.493 | 2.026 | 3.964 | 2.032 | 2.173 | 1.494 |
| Sindhi | 1.740 | 8.337 | 1.436 | 2.377 | 3.060 | 1.735 | 1.830 | 1.380 |
| Assamese | 1.677 | 4.474 | 1.655 | 2.815 | 8.506 | 3.006 | 2.303 | 1.562 |
| Sanskrit | 2.821 | 3.916 | 2.294 | 3.586 | 5.036 | 3.268 | 3.390 | 1.950 |
| English | 1.491 | 1.844 | 1.526 | 1.537 | 1.486 | 1.574 | 1.587 | 1.521 |
Model Card Contact โ๏ธ
Lingo Research Group at IIT Gandhinagar, India
Mail at: [email protected]