Model Description
This Memory Decoder model is trained on the Law domain and can be adapted to enhance any model in the Llama3, Llama3.1, and Llama3.2 families.
[!IMPORTANT] These Llama models are initialized from Qwen models with the embedding layer adapted to fit the Llama tokenizer. This enables efficient cross-model family knowledge transfer.
Paper: Memory Decoder: A Pretrained, Plug-and-Play Memory for Large Language Models
GitHub: https://github.com/LUMIA-Group/MemoryDecoder
Training & Evaluation Data
Law Domain Dataset: AsyLex
Test Split: MemoryDecoder-domain-data
Performance Results
Llama3 Family
| Model | Base Model | Base + MemDec |
|---|---|---|
| Llama3-8B | 5.96 | 4.46 |
| Llama3-70B | 4.90 | 4.07 |
Llama3.1 Family
| Model | Base Model | Base + MemDec |
|---|---|---|
| Llama3.1-8B | 5.88 | 4.42 |
| Llama3.1-70B | 4.89 | 4.06 |
Llama3.2 Family
| Model | Base Model | Base + MemDec |
|---|---|---|
| Llama3.2-1B | 8.23 | 5.11 |
| Llama3.2-3B | 6.83 | 4.76 |
Perplexity scores on Law domain test set. Lower is better.
Citation
@article{cao2025memory,
title={Memory decoder: A pretrained, plug-and-play memory for large language models},
author={Cao, Jiaqi and Wang, Jiarui and Wei, Rubin and Guo, Qipeng and Chen, Kai and Zhou, Bowen and Lin, Zhouhan},
journal={arXiv preprint arXiv:2508.09874},
year={2025}
}
Contact
For questions and support: maximus.cao@outlook.com