[ADD] Added Configure setting

[BaekTree]

No need to change train.py confiturations in hard coding.

• just change config.json as you want.

Added model configures as well

• if you use huggingface module, change the value.
• else define your own model and adjust this configure.

Moved the documentation of each TrainerAargument element to config_parser.py

config e.g.

  {
     "TrainingArguments" : {
         "output_dir":"./results",          
         "save_total_limit":5,              
         "save_steps":500,                 
         "num_train_epochs":20,              
         "learning_rate":5e-5,               
         "per_device_train_batch_size":16,  
         "per_device_eval_batch_size":16,   
         "warmup_steps":500,                
         "weight_decay":0.01,               
         "logging_dir":"./logs",            
         "logging_steps":100,              
         "evaluation_strategy":"steps", 
         "eval_steps" : 500,            
         "load_best_model_at_end" : false 
     },
     "model" :{
       "huggingface": "klue/bert-base"
     }
   } 

resolve #11

[nayohan]

코드잘봤습니다! PR해도 좋을거 같아요! 다른분들도 같이 보고가면 좋을거 같아요~

[Jjackson-dev]

하이퍼 파라미터 관리하기 쉽게 만들어주셨네요 좋습니다!