Fix tests

def get_dummy_batch(ntokens, src_dict, dst_dict, src_len=128, tgt_len=128):

    bsz = int(ntokens / max(src_len, tgt_len))

    bsz = (bsz // 8) * 8

    bsz = math.ceil(bsz / 8) * 8

    assert src_dict.pad() == dst_dict.pad()

    pad_idx = src_dict.pad()

    src_vocab, dst_vocab = len(src_dict), len(dst_dict)

                multiple of 8, which is important on some hardware (e.g., Nvidia

                Tensor Cores).

        """

        if padding_factor > 1:

        if nwords == -1:

            nwords = len(self)

        if padding_factor > 1:

            i = 0

            while nwords % padding_factor != 0:

                if nwords >= len(self):

        for k in params_keys:

            if k not in params_dict:

                params_dict[k] = []

            params_dict[k].append(model_params[k].float())

            p = model_params[k]

            if isinstance(p, torch.HalfTensor):

                p = p.float()

            params_dict[k].append(p)

    averaged_params = collections.OrderedDict()

    # v should be a list of torch Tensor.

    for i in range(vocab_size):

        token = prefix + str(i)

        d.add_symbol(token)

    d.finalize()

    d.finalize(padding_factor=1)  # don't add extra padding symbols

    return d