NPROC_PER_NODE=8 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 swift sft \
    --model_type qwen2-audio-7b \
    --sft_type full \
    --output_dir ./nisqa_abtest \
    --num_train_epochs 2 \
    --eval_steps  100000 \
    --save_steps  100000 \
    --batch_size 4 \
    --eval_batch_size 4 \
    --logging_steps 5 \
    --learning_rate 1e-5 \
    --deepspeed default-zero2 \
    --dtype fp16 \
    --dataset ../data/nisqa/NISQA_Corpus/NISQA_TRAIN_SIM/train_nisqa_abtest_llama_10k.json  \
    --val_dataset ../data/nisqa/NISQA_Corpus/NISQA_TRAIN_SIM/test_nisqa_abtest_2.5k.json  \
    --val_dataset_sample 500 \



#    NPROC_PER_NODE=8 CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 swift sft \
#    --model_type qwen2-audio-7b \
#    --sft_type ia3 \
#    --ia3_feedforward_modules all \
#    --sft_type full \
#    --output_dir ./nisqa_abtest \
#    --num_train_epochs 2 \
#    --eval_steps  25 \
#    --save_steps  25 \
#    --batch_size 4 \
#    --eval_batch_size 4 \
#    --logging_steps 5 \
#    --learning_rate 5e-6 \
#    --deepspeed default-zero2 \
#    --dtype fp16 \
#    --dataset ../data/nisqa/NISQA_Corpus/NISQA_TRAIN_SIM/abtest/train_ab900.json   \
#    --val_dataset ../data/nisqa/NISQA_Corpus/NISQA_TRAIN_SIM/abtest/test_ab100.json  \
#    --val_dataset_sample -1 \
#