Wav2Vec not getting fine tuned by writing a loop using dataloader

Models
1 
        with torch.set_grad_enabled(True):

            model.train()
            for batch in dataloader:

                input = batch["input_val"].to(device)
                labels = batch["label"].to(device)

                optimizer.zero_grad()

                out = model(input)
                
                loss = loss_fn(out["logits"], labels)

                loss_epoch += loss.detach().cpu().item()
                num_steps += input.shape[0]

                loss.backward()
                
                optimizer.step()

here model is

model = AutoModelForAudioClassification.from_pretrained(
    "facebook/wav2vec2-conformer-rope-large", num_labels=num_labels, label2id=label2id, id2label=id2label
).to(device)

I got the exact same results as I got with an untrained model