This repository contains the reference code for the paper Gated Rotary-Enhanced Linear Attention with Rank Modulation for Long-term Sequential Recommendation.
Here are our main environment dependencies for running the repository:
- NVIDIA-SMI 535.183.01
- cuda 12.2
- python 3.11.5
- pytorch 2.4.0
- recbole 1.2.0
- casual-conv1d 1.4.0
- timm 1.0.11
This repository contains the ML-1M dataset. If you want to train our model on other datasets, the LFM-1B, Tmall, and Netflix datasets can be downloaded from Google Drive. ML-32M can also be found at MovieLens and processed by conversion tools.
To reproduce the results reported in our paper, just run it:
python run_RecGRELA.py
You can also check the training log in📁 log.
Our code references RecBole, Mamba4Rec. We appreciate their outstanding works.
