def train()

in T5DST/T5.py [0:0]

• 33 lines of code
• 4 McCabe index (conditional complexity)

def train(args, *more):
    args = vars(args)
    args["model_name"] = args["model_checkpoint"]+args["model_name"]+"_except_domain_"+args["except_domain"]+ "_slotlang_" +str(args["slot_lang"]) + "_lr_" +str(args["lr"]) + "_epoch_" + str(args["n_epochs"]) + "_seed_" + str(args["seed"])
    # train!
    seed_everything(args["seed"])


    if "t5" in args["model_name"]:
        model = T5ForConditionalGeneration.from_pretrained(args["model_checkpoint"])
        tokenizer = T5Tokenizer.from_pretrained(args["model_checkpoint"], bos_token="[bos]", eos_token="[eos]", sep_token="[sep]")
        model.resize_token_embeddings(new_num_tokens=len(tokenizer))
    elif "bart" in args["model_name"]:
        model = BartForConditionalGeneration.from_pretrained(args["model_checkpoint"])
        tokenizer = BartTokenizer.from_pretrained(args["model_checkpoint"], bos_token="[bos]", eos_token="[eos]", sep_token="[sep]")
        model.resize_token_embeddings(new_num_tokens=len(tokenizer))

    task = DST_Seq2Seq(args, tokenizer, model)

    train_loader, val_loader, test_loader, ALL_SLOTS, fewshot_loader_dev, fewshot_loader_test = prepare_data(args, task.tokenizer)

    #save model path
    save_path = os.path.join(args["saving_dir"],args["model_name"])
    if not os.path.exists(save_path):
        os.makedirs(save_path)

    trainer = Trainer(
                    default_root_dir=save_path,
                    accumulate_grad_batches=args["gradient_accumulation_steps"],
                    gradient_clip_val=args["max_norm"],
                    max_epochs=args["n_epochs"],
                    callbacks=[pl.callbacks.EarlyStopping(monitor='val_loss',min_delta=0.00, patience=5,verbose=False, mode='min')],
                    gpus=args["GPU"],
                    deterministic=True,
                    num_nodes=1,
                    #precision=16,
                    accelerator="ddp"
                    )

    trainer.fit(task, train_loader, val_loader)

    task.model.save_pretrained(save_path)
    task.tokenizer.save_pretrained(save_path)

    print("test start...")
    #evaluate model
    _ = evaluate_model(args, task.tokenizer, task.model, test_loader, save_path, ALL_SLOTS)