Pytorch 1.5 optimizer causing inplace operations in .step() (possible bug)

In reference to this issue. The PyTorch release notes mention this change but I'm not sure how to apply their guidelines and there could just be a bug.

This code will display the error.

import torch
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt


# Hyper Parameters
BATCH_SIZE = 64
LR_G = 0.0001
LR_D = 0.0001 
N_IDEAS = 5  
ART_COMPONENTS = 15 
PAINT_POINTS = np.vstack([np.linspace(-1, 1, ART_COMPONENTS) for _ in range(BATCH_SIZE)])


def artist_works():  # painting from the famous artist (real target)
    r = 0.02 * np.random.randn(1, ART_COMPONENTS)
    paintings = np.sin(PAINT_POINTS * np.pi) + r
    paintings = torch.from_numpy(paintings).float()
    return paintings


G = nn.Sequential(  # Generator
    nn.Linear(N_IDEAS, 128),  # random ideas (could from normal distribution)
    nn.ReLU(),
    nn.Linear(128, ART_COMPONENTS),  # making a painting from these random ideas
)

D = nn.Sequential(  # Discriminator
    nn.Linear(ART_COMPONENTS, 128),  # receive art work either from the famous artist or a newbie like G
    nn.ReLU(),
    nn.Linear(128, 1),
    nn.Sigmoid(),  # tell the probability that the art work is made by artist
)

opt_D = torch.optim.Adam(D.parameters(), lr=LR_D)
opt_G = torch.optim.Adam(G.parameters(), lr=LR_G)


for step in range(10000):
    artist_paintings = artist_works()  # real painting from artist
    G_ideas = torch.randn(BATCH_SIZE, N_IDEAS)  # random ideas
    G_paintings = G(G_ideas)  # fake painting from G (random ideas)

    prob_artist0 = D(artist_paintings)  # D try to increase this prob
    prob_artist1 = D(G_paintings)  # D try to reduce this prob

    D_loss = - torch.mean(torch.log(prob_artist0) + torch.log(1. - prob_artist1))
    G_loss = torch.mean(torch.log(1. - prob_artist1))

    opt_D.zero_grad()
    D_loss.backward(retain_graph=True)  # reusing computational graph
    opt_D.step()

    opt_G.zero_grad()
    G_loss.backward()
    opt_G.step()

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [128, 1]], which is output 0 of TBackward, is at version 4; expected version 3 instead.

The only [128,1] tensor parameter would be the linear layer in D(). The error comes from somewhere within the opt_D.step() call and then calling a backward() pass on the retained graph.

The release notes have this example to clone the parameter:

def model(input, target, param):
    return (input * param ** 2 - target).norm()

param = torch.randn(2, requires_grad=True)
input = torch.randn(2)
target = torch.randn(2)
sgd = optim.SGD([param], lr=0.001)
loss = model(input, target, param.clone())
loss.backward(retain_graph=True)
sgd.step()
loss.backward()
param.grad

Which solves the problem here, but not in the initial example because it's a bit more nuanced as the parameter is coming from the Linear layer of the discriminator model.