Some questions regarding the reproduction of the sequential recommendation task

[AkaliKong]

Hello, thank you for your outstanding work! I am currently facing some challenges in reproducing the results of the sequential recommendation using Rec-R1 on the All_Beauty dataset as described in your paper. If it is convenient, could you kindly share the specific parameter configurations and the training logs from Wandb?

[linjc16]

Hi! Thanks for your interest in our work. For the configuration of Amazon Beauty, you could check the file scripts/train/train_rec-amazon_review_3b.sh.

export N_GPUS=2 export BASE_MODEL=Qwen/Qwen2.5-3B-Instruct export DATA_DIR=data/amazon_review/inst export ROLLOUT_TP_SIZE=2

The eval code is scripts/eval/amazon_review/inst_gen/rec-r1.sh, where we use the global_step_1000 checkpoint, according to the validation set performance. For the wandb logs, I didn't find the share button. Maybe this is because I was using the team version, maybe I should deal with the permission stuff or something like that. Let me find a way.