MQL4GRec

This is the code for the ICLR 2025 paper:
Multimodal Quantitative Language for Generative Recommendation

Setup

pytorch==2.1.0
transformers <= 4.45.0

• We found that different versions of transformers significantly impact convergence speed and performance under default parameters.
  For newer versions, parameter adjustments are required.

We tested different transformers versions with accelerate==0.28.0:

• v4.47.0, 4.48.0, 4.50.0

• v4.38.2, 4.39.0, 4.40.0, 4.45.0

• v4.46.0 shows high training loss
  

v4.43.0, 4.44.0 are incompatible with accelerate==0.28.0 - version change required.

Quick Start

Data Processing

cd data_process

1. Download images
2. Process data to ensure each item corresponds to one image and one text description
3. Generate text embeddings
4. Generate image embeddings

Preprocessed data, pretrained checkpoints, and training logs:
Google Drive Folder

Training the Quantitative Translator

cd index
bash script/run.sh          # Run training  
bash script/gen_code_dis.sh # Generate code  

Pre-training

bash script/pretrain.sh

Fine-tuning

bash finetune.sh

Notes

• Adjust file paths according to your local directory structure

Contributing

PRs and issues are welcome!

License

N/A