main.py 文件源码

def batchify(data, bsz):
    # Work out how cleanly we can divide the dataset into bsz parts.
    nbatch = data.size(0) // bsz
    # Trim off any extra elements that wouldn't cleanly fit (remainders).
    data = data.narrow(0, 0, nbatch * bsz)
    # Evenly divide the data across the bsz batches.
    data = data.view(bsz, -1).t().contiguous()
    if args.cuda:
        data = data.cuda()
    return data