Code of Paper "Enhancing LLM-based Recommendation through Semantic-Aligned Collaborative Knowledge"

📊 Evaluation

To reproduce the evaluation results of SeLLa, follow these steps:

📦 Step 1: Download Datasets and Finetuned Model Weights

Download datasets and corresponding finetuned model weights from the links below and place them in eval_data_book/eval_data_movie and final_ckpt_path_book/final_ckpt_path_movie

Datasets link	Finetuned Model weights link
Amazon-Book	checkpoint-656-Book
ML-1M	checkpoint-254-Movie

🚀 Step 2: Run Evaluation Script

Execute the following command to evaluate the model:

ds_config_path=examples/sft/ds_config_zero3.json

bash examples/sft/finetune_stage2_book_step1_cycle_two.sh ${final_ckpt_path_book} ${train_data_dummy} ${ds_config_path} ${eval_data_book} ${output_dir_book}

bash examples/sft/finetune_stage2_movie_step1_cycle_two.sh ${final_ckpt_path_movie} ${train_data_dummy} ${ds_config_path} ${eval_data_movie} ${output_dir_movie}

📝 Expected Output

The script will compute metrics like AUC, UAUC (the values will match those reported in the paper's Table 3). Detailed evaluation results and full logs d will be stored at output_dir_book/output_dir_movie

---

⚡ Training

...