Inheriting from BartForConditionalGeneration into a new class - weight not initializing

Beginners
5

I am starting (before fine tuning) with:

class Exp(nn.Module):
    def __init__(self,config):
        super(Exp,self).__init__()
        self.bart = BartForConditionalGeneration.from_pretrained(config)
    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        decoder_input_ids=None,
        decoder_attention_mask=None,
        head_mask=None,
        decoder_head_mask=None,
        encoder_outputs=None,
        past_key_values=None,
        inputs_embeds=None,
        decoder_inputs_embeds=None,
        labels=None,
        use_cache=None,
        output_attentions=None,
        output_hidden_states=None,
        return_dict=None,
    ):
        output = self.bart(input_ids=input_ids,attention_mask=attention_mask,
                     labels=labels,decoder_input_ids=decoder_input_ids,encoder_outputs=encoder_outputs)
        return output

Loading the pre-trained model before fine-tuning :
model = BartExp(‘facebook/bart-base’)

Since the model is being trained by multiple GPUs , I transform model = torch.nn.DataParallel(model)
and then my saving method is

    model_to_save = model.module if hasattr(
        model, 'module') else model  # Only save the model it-self
    model_to_save.bart.save_pretrained(args.output_dir)
    tokenizer.save_vocabulary(args.output_dir)

So, when loading the trained model , I am doing the same procedure :
model = BartForQaSimplification(args.output_dir)
where output_dir is the path to my pytorch_model.bin file. I am loading the fine-tuned model, not initiating the pre-trained model again , right ?