CoderRec

This is the implementation of paper "Cross-Scale Collaboration between LLMs and Lightweight Sequential Recommenders with Domain-Specific Latent Reasoning"

Install

Create and activate the conda environment

conda create -n coderrec Python=3.10
conda activate coderrec

Install PyTorch (CUDA 12.1)

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121

Install other dependencies

pip install -r requirements.txt

Usage

Training

• Train the RQVAE.

  CUDA_VISIBLE_DEVICES=0 python train_rqvae.py --config ./configs/rqvae/beauty.json

• Train the decoder.

  CUDA_VISIBLE_DEVICES=0 python train_decoder.py --config ./configs/decoder/beauty.json --pretrained_rqvae_path <PATH_OF_RQVAE_CHECKPOINT>

• Train the overall pipeline with latent reasoning.

  • This step requires substantially greater computational resources than the preceding two steps. However, even if these resources cannot be met, satisfactory performance can still be achieved by relying solely on the first two steps.

  CUDA_VISIBLE_DEVICES=0,1,2,3 accelerate launch --num_processes 4 train_pipeline.py --config ./configs/pipeline/beauty.json --n_passes 2 --enable_reasoning 1 --pretrained_rqvae_path <PATH_OF_RQVAE_CHECKPOINT> --pretrained_decoder_path <PATH_OF_DECODER_CHECKPOINT>

Inference

• You can download the model weights here, then run

  bash ./scripts/inference_beauty.sh
  

Acknowledgements

This repository is developed based on the excellent implementation of RQ-VAE-Recommender.