training config for sbert mlttsk

train_sbert_config.py

class TrainingConfig:
    def __init__(self, **kwargs):
        
        self.model_name = "sbert_tuned"
        self.data_dir = "/content/drive/MyDrive/"
        self.module_path = "/content/bert_module/"
        self.pretrained_ckpt = None
        self.generation = "sbert"

        self.ctx_len = 24
        
       
        self.dim = 768
        self.n_tune = 12
        self.lr = 2e-6 
        self.train_bert = True
        self.tune_embs =True

        self.head_dropout_rate  = 0.1

        self.tagger_loss_weight = 0.1
        self.toxic_loss_weight = 0.1
        self.paraphrase_loss_weight = 0.2
        self.sample_dropout = 0.1
        
        self.use_par_head = True
        self.use_ner_head = True
        self.use_toxic_head = True
      
        self.eval_on_start = False
        
        self.batch_size = 24
        self.eval_batch_size = 256
        self.n_epochs = 256
        self.n_tune_epochs = 25
        self.epoch_steps = 512
        self.valid_steps = 32
        self.metric_steps = 16

        self.n_toxic_tags = 2
        self.optimizer = 'adam'
        self.n_tags = 13

        self.type_loss = 'softmax'

        for k, v in kwargs.items():
            setattr(self, k, v)

        self.print_config()

    def to_dict(self):
        return self.__dict__
    
    def to_string(self):
        conf_dict = self.to_dict()
        ser_string = ""
        for k in conf_dict:
            ser_string += f"_{k}_{conf_dict[k]}"
        return ser_string

    def print_config(self):
        log_this(50 * '-')
        log_this(f"Using config:")
        mlen = max([len(v) for v in self.to_dict().keys()])
        for k, v in self.to_dict().items():
            log_this(f"{k}{(mlen - len(k)) * ' '} = {v}")
        log_this(50 * '-')