/home/admin/.conda/envs/llama_etuning/bin/deepspeed  
num_gpus 8                                     src/train_bash.py                                     
deepspeed ds_config_7b.json                                     
stage sft                                     
model_name_or_path ../.././model/Mistral-7B-v0.1/                                     
do_train                                     
dataset slimorca_all                                     
template alpaca                                     
finetuning_type lora                                     
output_dir  ../.././model/Ads_Mistral7B-slimorca_all-Lqv-r4b128/                                     
overwrite_cache                                     
per_device_train_batch_size 4                                     
gradient_accumulation_steps 4                                     
lr_scheduler_type cosine                                     
logging_steps 10                                     
learning_rate 2e-05                                     
weight_decay 0                                     
num_train_epochs 1                                     
plot_loss                                     
bf16                                     
save_strategy epoch                                     
save_steps 10000                                     
save_total_limit 1                                                                          
warmup_steps 100                                     
ddp_find_unused_parameters False                                     
cutoff_len 4096                                     
group_by_length False                                     
preprocessing_num_workers 256                            
lora_rank 4                            
lora_alpha 16                            
lora_dropout 0.05                            
lora_target q_proj,v_proj