How to accelerate the speed of creating the huge matrix in python?

Question

I am trying to create a huge matrix(60M X 3.1M),and I have written two script, one is try to use the gpu to speed up, the other is use the cpu to compute. I put the script on the SLURM and it has been running for almost a week, and it seems runs forever.I am a very beginner of python and data analyst, so I wonder if there is any chance to accelerate the process? Here is my code:

cpu version:

f = open('/etc/d_pair.txt', 'r') # the file contain 30M pair of diagnosis (two diseases in one pair)
f2 = open('/etc/ehr.txt', 'r')  # the ehr file, contain 60M patients and theirs diagosis
nf = open('/result/C_matrix.txt', 'w', newline='')
fline = f.readlines()
f2line =f2.readlines()  # a is list of all disease pair
All = []
for Id in f2line:  # each one of 60M patients
    b = Id.split(' ')  # turn each id and diagnosis into list
    nf.write(str(b[0]))
    nf.write(' ')
    for name in fline:  # a is list of all disease pair
        name = name.split(',')
        if name[0] in b:  # if the patient have both 2 d, give the value '1' to him/her
            if name[1] in b:
                nf.write('1')
            else:  # if the patient have 1 or 0 disease in those pair ,give the value 0 to him/her
                nf.write('0')
        else:
            nf.write('0')
        nf.write(' ')
    nf.write('\n')

f, f2, nf.close()

the gpu version:

import torch
import pandas as pd
torch.cuda.set_device(0)
f = open('/etc/d_pair.txt', 'r')
f2 = open('/etc/ehr.txt', 'r')
fline = f.readlines()
f2line =f2.readlines()  # a is list of all d pair
All = []
for Id in f2line:  # each one of 600M
    TIMES = []
    b = Id.split(' ')  # turn each id and diagnosis into list
    TIMES.append(int(b[0]))
    for name in fline:  # a is list of all d pair
        name = name.split(',')
        if name[0] in b:
            if name[1] in b:
                TIMES.append(1)
            else:
                TIMES.append(0)
        else:
            TIMES.append(0)
    All.append(TIMES)
c = torch.as_tensor(All).cuda()  # turn the 2-dimension list into tensor.
pc = pd.DataFrame(c).astype("int")
pc.to_csv('/result/matrix.csv', index=False, header=False)

I have read some similar questions, but some of their matrixs are not as huge as mine, some of them are aim to compute the matrix(dot and etc), so I am not sure if pytable or scipy .sparse.csc.matrix is suitable for my circumstance.

Thanks in advance!