phonsobon/khmer-word-prediction

Khmer Word Prediction Model

Caption

A Khmer language model for next-word prediction and text generation, designed to support applications such as autocomplete, intelligent typing assistants, and Khmer NLP research.

---

Model Description

This model is trained to predict the next word in a Khmer sentence using deep learning techniques.

• Model file: khmer_lm_best.pt
• Framework: PyTorch
• Architecture: (update this: LSTM / GRU / Transformer)
• Task: Next-word prediction (Language Modeling)

---

Dataset

• Source: Synthetic + real Khmer text
• Training samples: ~350,000 sentences
• Validation samples: ~47,000 sentences

---

How to Use (Test Code)

1. Install dependencies

pip install torch
import torch

# Load model
model = torch.load("khmer_lm_best.pt", map_location="cpu")
model.eval()
def predict_next_word(model, input_text):
    # TODO: replace with your tokenizer logic
    tokens = input_text.split()

    # Dummy example (you must adapt to your model)
    input_ids = [0] * len(tokens)  # replace with real encoding

    input_tensor = torch.tensor([input_ids])

    with torch.no_grad():
        output = model(input_tensor)

    predicted_id = output.argmax(dim=-1)[0, -1].item()

    # TODO: decode predicted_id to word
    return predicted_id


# Example
text = "ខ្ញុំចង់ទៅ"
print("Input:", text)
print("Predicted next word:", predict_next_word(model, text))