VyDat/transformer-attres-vie-vsl

Transformer AttRes: Vietnamese to Vietnamese Sign Language (VSL)

A Transformer model with Attention Residuals (AttRes) for translating Vietnamese text to Vietnamese Sign Language (VSL) gloss.

Model Details

• Architecture: Transformer with Attention Residuals (AttRes)
• Parameters: 49,912,999
• Layers: 6
• d_model: 512
• dff: 2048
• Heads: 8
• Block size: 2
• Max length: 50

Usage

# Clone the repo or download files
from model import load_model, translate

# Load model
model, src_tok, tgt_tok = load_model(".", device="cuda")

# Translate
result = translate(model, src_tok, tgt_tok, "Xin chào, bạn khỏe không?", device="cuda")
print(result)

Files

• config.json - Model configuration
• pytorch_model.bin - Model weights
• src_tokenizer.pkl - Source (Vietnamese) tokenizer
• tgt_tokenizer.pkl - Target (VSL) tokenizer
• model.py - Model definition (required for loading)

Citation

If you use this model, please cite appropriately.