run.sh

for data in 'Industrial_and_Scientific' 'Prime_Pantry' 'Musical_Instruments' 'Arts_Crafts_and_Sewing' 'Video_Games'
do

    gpu_num=4

    if [ "${data}" == 'Arts_Crafts_and_Sewing' ]; then
        epochs=10
        lr=1.2e-4
        sd=0.4
        btz=10
        all_steps=56083
        klws=(0.5)
        fas=(0.7)
        fts=(1.0)
        lrs=(1.6e-4)
    elif [ "${data}" == 'Industrial_and_Scientific' ]; then
        epochs=10
        lr=1.3e-4
        sd=0.5
        btz=10
        all_steps=10975
        klws=(0.7)
        fas=(0.5)
        fts=(1.0)
        lrs=(1.5e-4)
    elif [ "${data}" == 'Musical_Instruments' ]; then
        epochs=8
        lr=1.2e-4
        sd=0.4
        btz=10
        all_steps=27518
        klws=(0.7)
        fas=(0.7)
        fts=(1.0)
        lrs=(1.5e-4)
    elif [ "${data}" == 'Prime_Pantry' ]; then
        epochs=10
        lr=1.3e-4
        sd=0.5
        btz=10
        all_steps=14177
        klws=(0.3)
        fas=(0.3)
        fts=(1.0)
        lrs=(1.5e-4)
    elif [ "${data}" == 'Video_Games' ]; then
        epochs=12
        lr=1.2e-4
        sd=0.4
        btz=10
        all_steps=55126
        klws=(0.3)
        fas=(0.5)
        fts=(1.2)
        lrs=(1.5e-4)
    fi
    
    models=("deepseek-ai/DeepSeek-R1-Distill-Llama-8B")

    lf=1
    ld=0.5
    max_item_num=10
    hdf=8
    hd=0.05
    wd=1e-2
    
    for kt in 1.0;do
    for ft in "${fts[@]}";do
    for fa in "${fas[@]}";do
    for klw in "${klws[@]}";do
    for lr in "${lrs[@]}"; do
    for model in "${models[@]}"; do
        processed_model="${model//\//-}"
        warmup_steps=$(awk "BEGIN {print int($all_steps / ($gpu_num * $btz) * $epochs * 0.06 + 0.5)}")

        if [ -d "./output/${data}/${processed_model}_all_klw${klw}_fa${fa}" ]; then
            echo "${data} ${processed_model}_all_klw${klw}_fa${fa} exists"
        else
            CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
            --suffix all_klw${klw}_fa${fa} \
            --kl_loss_weight ${klw} \
            --kl_temperature ${kt} \
            --fusion_temperature ${ft} \
            --fusion_alpha ${fa} \
            --use_gate \
            --use_small_model \
            --skip_valid 0 \
            --weight_decay ${wd} \
            --hidden_dropout ${hd} \
            --hd_frequency ${hdf} \
            --score_dropout ${sd} \
            --adapter_dropout ${ld} \
            --lora_frequency ${lf} \
            --use_lora \
            --mixed_precision bf16 \
            --learning_rate ${lr} \
            --max_item_num ${max_item_num} \
            --patient 1 \
            --model_name_or_path ${model} \
            --fix_backbone \
            --fix_emb \
            --batch_size ${btz} \
            --num_train_epochs ${epochs} \
            --warmup_steps ${warmup_steps} \
            --dataset ${data}
        fi

        # if [ -d "./output/${data}/${processed_model}_only_text" ]; then
        #     echo "${data} ${processed_model}_only_text exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix only_text \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_only_id" ]; then
        #     echo "${data} ${processed_model}_only_id exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix only_id \
        #     --only_id \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_early_fusion" ]; then
        #     echo "${data} ${processed_model}_early_fusion exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix early_fusion \
        #     --early_fusion \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_fusion_before_loss" ]; then
        #     echo "${data} ${processed_model}_fusion_before_loss exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix fusion_before_loss \
        #     --fusion_before_loss \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_random_init" ]; then
        #     echo "${data} ${processed_model}_random_init exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix random_init \
        #     --method_of_preference random \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi
        
        # if [ -d "./output/${data}/${processed_model}_oneLoRA_1" ]; then
        #     echo "${data} ${processed_model}_oneLoRA_1 exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix oneLoRA_1 \
        #     --alternating_learning 0 \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_oneLoRA_2" ]; then
        #     echo "${data} ${processed_model}_oneLoRA_2 exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix oneLoRA_2 \
        #     --lora_r 16 \
        #     --lora_alpha 64 \
        #     --alternating_learning 0 \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_MainId" ]; then
        #     echo "${data} ${processed_model}_MainId exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix MainId \
        #     --fusion_type 'id' \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_both" ]; then
        #     echo "${data} ${processed_model}_both exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix both \
        #     --fusion_type 'both' \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi
        
        # if [ -d "./output/${data}/${processed_model}_two_score" ]; then
        #     echo "${data} ${processed_model}_two_score exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix two_score \
        #     --use_two_score \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_SASRec_${psize}" ]; then
        #     echo "${data} ${processed_model}_SASRec_${psize} exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix SASRec_${psize} \
        #     --method_of_preference SASRec \
        #     --preference_dim ${psize} \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi

        # if [ -d "./output/${data}/${processed_model}_GRU4Rec_${psize}" ]; then
        #     echo "${data} ${processed_model}_GRU4Rec_${psize} exists"
        # else
        #     CUDA_VISIBLE_DEVICES=0,1,2,3 torchrun --nproc_per_node=$gpu_num --master_port=29105 src/main.py \
        #     --suffix GRU4Rec_${psize} \
        #     --method_of_preference GRU4Rec \
        #     --preference_dim ${psize} \
        #     --kl_loss_weight ${klw} \
        #     --kl_temperature ${kt} \
        #     --fusion_temperature ${ft} \
        #     --fusion_alpha ${fa} \
        #     --use_gate \
        #     --use_small_model \
        #     --skip_valid 0 \
        #     --weight_decay ${wd} \
        #     --hidden_dropout ${hd} \
        #     --hd_frequency ${hdf} \
        #     --score_dropout ${sd} \
        #     --adapter_dropout ${ld} \
        #     --lora_frequency ${lf} \
        #     --use_lora \
        #     --mixed_precision bf16 \
        #     --learning_rate ${lr} \
        #     --max_item_num ${max_item_num} \
        #     --patient 1 \
        #     --model_name_or_path ${model} \
        #     --fix_backbone \
        #     --fix_emb \
        #     --batch_size ${btz} \
        #     --num_train_epochs ${epochs} \
        #     --warmup_steps ${warmup_steps} \
        #     --dataset ${data}
        # fi
    done
    done
    done
    done
    done
    done
done

# CUDA_VISIBLE_DEVICES=3 torchrun --nproc_per_node=1 --master_port=29105 src/main.py \
#     --suffix debug \
#     --bo_percent 0.25 \
#     --bo_num_trials 3 \
#     --valid_metric MRR \
#     --debug \
#     --skip_valid 0 \
#     --use_gate \
#     --use_context_mask \
#     --use_small_model \
#     --warmup_steps 10 \
#     --down_size 96 \
#     --model_name_or_path facebook/opt-125m \
#     --fix_backbone \
#     --fix_emb \
#     --adapter_mode 1 \
#     --houlsby_mode 0 \
#     --batch_size 2 \
#     --num_train_epochs 10 \
#     --dataset 'Prime_Pantry'