How to modify batch data to reflect changes in original data in dataloader/pytorch?

I recently created a dataset class and am having trouble modifying the data in the batch so that it is reflected in future batches and original data

I have the following dataset class

class KrakenDataSet(Dataset):
    """Creates a tensor Xt as defined by equation 18 in paper to feed into ANN

    Parameters
    ----------
    Dataset : _type_
        Base class from torch utils class that allows to iterate over dataset
    """

    def __init__(
        self,
        portfolio: Portfolio,
        pvm: torch.tensor,
        window_size: int = 50,
        step_size: int = 1,
        device="mps",
    ):
        self.portfolio = portfolio
        self.window_size = window_size
        self.step_size = step_size
        self.close_pr = torch.tensor(
            self.portfolio.get_close_price().values[:, 1:], dtype=torch.float32
        ).to(device)
        self.high_pr = torch.tensor(
            self.portfolio.get_high_price().values[:, 1:], dtype=torch.float32
        ).to(device)
        self.low_pr = torch.tensor(
            self.portfolio.get_low_price().values[:, 1:], dtype=torch.float32
        ).to(device)
        self.pvm = pvm

    def __len__(self):
        return self.portfolio.nobs

    def __getitem__(self, idx):
        msecurities = self.portfolio.m_noncash_securities
        start = idx * self.step_size
        end = start + self.window_size
        xt = torch.zeros(3, msecurities, self.window_size)
        xt[0] = (self.close_pr[start:end:,] / self.close_pr[end - 1,]).T
        xt[1] = (self.high_pr[start:end:,] / self.close_pr[end - 1,]).T
        xt[2] = (self.low_pr[start:end:,] / self.close_pr[end - 1,]).T
        return xt, self.pvm[end - 2]

The purpose of this class is to take closing prices, high and low from last window_size and return the current xt. I have a pvm variable which holds all the weights.

When i do this:

pvm = (torch.ones(nobs, msecurities) / msecurities).to(device)
kraken_ds = KrakenDataSet(port, pvm, window_size=WINDOW_SIZE, device=device)

for xt, wt in kraken_ds:
    wt[0] = 5

im able to view the changes in pvm[48] as thats the index i want to make the change in

But the moment i put this into a batch, the same modification doeesn't propagate through the original pvm

kraken_dl = DataLoader(
    kraken_ds,
    batch_size=BATCH_SIZE,
    shuffle=False,
    drop_last=False,
    # generator=torch.Generator(device="mps"),
)

for xt, wt, in kraken_dl:
    wt[0][0] = 5

I searched along and im not sure if dataloader creates a deep copy of the dataset and hence im unable to modify the original dataset when i make modifications to the batch dataset? Can anyone suggest a workaround? Thanks

Changes made inside the batch of dataloader modifies the original dataset