|
|
--- |
|
|
license: cc-by-sa-4.0 |
|
|
language: |
|
|
- tig |
|
|
tags: |
|
|
- tigre |
|
|
- language-model |
|
|
- kenlm |
|
|
- n-gram |
|
|
--- |
|
|
|
|
|
# Tigre 3-gram Language Model (KenLM) |
|
|
|
|
|
### Overview |
|
|
|
|
|
This repository provides a **3-gram Language Model (LM)** for the **Tigre** language, trained using the **KenLM** toolkit. This model is a foundational resource for various downstream NLP and speech applications, including: |
|
|
|
|
|
- Rescoring hypotheses in Automatic Speech Recognition (ASR). |
|
|
- Improving text generation and fluency in Machine Translation (MT). |
|
|
- Performing basic text filtering and quality control. |
|
|
The model is provided in the highly optimized binary (`.arpa`) format, making it suitable for efficient use in production environments. |
|
|
|
|
|
## Model Statistics |
|
|
This language model was trained using KenLM on the **Tigre Monolingual Text Dataset (Tigre-Data 1.0)**. |
|
|
|
|
|
| Statistic | Value | |
|
|
| :----------------------------------- | :-------- | |
|
|
| **Model Order** | 3-gram | |
|
|
| **Vocabulary Size (Unique 1-grams)** | 316,548 | |
|
|
| **Total Unique N-grams (1-to-3)** | 1,285,462 | |
|
|
| **Example Perplexity** (on 'α€α΅') | 147.12 | |
|
|
|
|
|
_Note: The total raw training tokens used for this model can be found in the Tigre Monolingual Text Dataset card (approximately 14.7 million tokens)._ |
|
|
|
|
|
## Training Data Source |
|
|
This model was trained exclusively on the **BeitTigreAI/tigre-data-monolingual-text** dataset. |
|
|
More detailed information about the training data, including its domain, bias, preprocessing steps, and source statistics, can be found in the dataset's documentation: |
|
|
[Tigre Monolingual Text Dataset README](https://huggingface.co/datasets/BeitTigreAI/tigre-data-monolingual-text/blob/main/README.md) |
|
|
|
|
|
--- |
|
|
|
|
|
## Files and Structure |
|
|
|
|
|
The repository contains the following files: |
|
|
tigre-data-kenLM/ |
|
|
βββ README.md |
|
|
βββ hf_readme.ipynb |
|
|
βββ tigre-data-kenLM.arpa |
|
|
|
|
|
## How to Use the Model |
|
|
You can load and query the model using the Python bindings for **KenLM** (`kenlm`). |
|
|
|
|
|
### Installation |
|
|
To use the model in Python, install the KenLM bindings: |
|
|
```bash |
|
|
!pip install kenlm |
|
|
|
|
|
## Example Usage (Perplexity and Score) |
|
|
The following Python code demonstrates how to load the model and query it for log probability and perplexity: |
|
|
|
|
|
```python |
|
|
import kenlm |
|
|
from huggingface_hub import hf_hub_download |
|
|
# 1. Download the ARPA model file from the Hugging Face Hub |
|
|
arpa_path = hf_hub_download( |
|
|
repo_id="BeitTigreAI/tigre-data-kenLM", |
|
|
filename="tigre-data-kenLM.arpa", |
|
|
repo_type="model" |
|
|
) |
|
|
# 2. Load the KenLM model |
|
|
lm = kenlm.Model(arpa_path) |
|
|
# Example single sentence to score |
|
|
test_sentence = "αααα αααα ααα₯α" # Or use one of the lines from your list |
|
|
# A. Calculate Log10 Probability of the entire sentence |
|
|
log_prob = lm.score(test_sentence) |
|
|
print(f"Sentence: '{test_sentence}'") |
|
|
print(f"Log10 Probability: {log_prob:.4f}") |
|
|
# B. Calculate Perplexity of the entire sentence |
|
|
perplexity = lm.perplexity(test_sentence) |
|
|
print(f"Perplexity: {perplexity:.2f}") |
|
|
|
|
|
``` |
|
|
|
|
|
## Licensing and Citation |
|
|
The Tigre 3-gram Language Model is licensed under CC-BY-SA-4.0. |
|
|
|
|
|
## Citation |
|
|
If you use this resource in your work, please cite the repository by referencing its Hugging Face entry: |
|
|
|
|
|
## Recommended Citation Format: |
|
|
|
|
|
## Repository Name: Tigre 3-gram Language Model (KenLM) |
|
|
|
|
|
## Organization: BeitTigreAI |
|
|
URL: https://huggingface.co/datasets/BeitTigreAI/tigre-data-kenLM |
|
|
|
