In your description on the page of the github project. We should use
- freeze_rec: True # freeze the collaborative rec model
- freeze_proj: True # freeze the mapping function or not
- freeze_lora: False # tuning the LoRA module
to tuning Lora as stage 1 and
- freeze_rec: True # freeze the collaborative rec model
- freeze_proj: False # tuning the mapping function or not
- freeze_lora: True # freeze the LoRA module
to tuning CIE as stage 2. This is for MF.
But in your config files 'collm_pretrain_lgcn_ood_cc.yaml' and 'collm_pretrain_din_ood_cc.yaml', 5 line and 6 line tell me that # stage 1: proj false, lora: false and # stage 2: proj true, lora false.
This is contradictory. Which one should I take as the standard?
In your description on the page of the github project. We should use
to tuning Lora as stage 1 and
to tuning CIE as stage 2. This is for MF.
But in your config files 'collm_pretrain_lgcn_ood_cc.yaml' and 'collm_pretrain_din_ood_cc.yaml', 5 line and 6 line tell me that # stage 1: proj false, lora: false and # stage 2: proj true, lora false.
This is contradictory. Which one should I take as the standard?