SuperposPrompt_Thesis/09_Cluster/train_single.py

import numpy as np
import torch

import os
import sys
sys.path.insert(1, os.path.join(sys.path[0], '..'))

from _utils import silent_logs, sp_decode
from _datasets import AutoLoad
from _trainer import auto_train
from _mydelta import auto_mutate
from _models import auto_model
from _config import Config


DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

def run_experminent(config, task_name):
    silent_logs()
    np.random.seed(config.random_seed)
    torch.manual_seed(config.random_seed)
    # ______________________LOAD MODEL_____________________________
    model, tokenizer = auto_model(config.model_name, AutoLoad.get_task_output(task_name))
    
    # ______________________MUTATE MODEL_____________________________
    n_prefix_token = 0
    if config.peft_params is not None:
        n_prefix_token = config.peft_params.n_tokens
        delta_module = auto_mutate(
            model=model,
            tokenizer=tokenizer,
            peft_params=config.peft_params.to_dict(),
            remove_dropout=config.remove_dropout
        )
    
    # ______________________LOAD DATA_____________________________
    autoload = AutoLoad(tokenizer, n_prefix_token=n_prefix_token)

    # ______________________TRAIN_____________________________
    dataset = autoload.get_and_map(task_name)
    auto_train(model, tokenizer, dataset, config, device=DEVICE)
    
if __name__ == '__main__':
    config_json = sp_decode(sys.argv[1])
    config = Config(config_json, '')
    task_name = sp_decode(sys.argv[2])
    run_experminent(config, task_name)