EfficientMPT: Efficient Message-Passing Transformer for Error Correcting Codes

This repository provides the official implementation of "Efficient Message-Passing Transformer for Error Correcting Codes" (ICLR 2026).

Paper link: https://openreview.net/forum?id=Xk8cwnwu2e

Installation

• Pytorch

Running the code

Codes for training EfficientMPT on GPU 0, 6 decoder layers, dimension 128 on (121, 60) LDPC code and (31,16) BCH code

python Main.py --gpu=0 --N_dec=6 --d_model=128 --code_type=LDPC --code_n=121--code_k=60
python Main.py --gpu=0 --N_dec=6 --d_model=128 --code_type=BCH --code_n=31--code_k=16 --standardize

Code arguments

--epochs                   number of epoch
--batch_size               batch size
--code_type                code type
--code_k                   codeword dimension k
--code_n                   codeword length n
--N_dec                    number of decoder layers N
--d_model                  embedding vector dimension d

Citation

@inproceedings{
Park2026Efficientmpt,
title={Efficient Message-Passing Transformer for Error Correcting Codes},
author={Seong-Joon Park and Taewoo Park and Hee-Youl Kwak and Sang-Hyo Kim and Yongjune Kim and Jong-Seon No},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=gFvRRCnQvX}
}

License

Codes are available only for non-commercial research purposes.

Acknowledgement

This repository is based on CrossMPT.