RuntimeError: Can't call numpy() on Variable that requires grad. Use var.detach().numpy() instead.

[urushi3]

An error occurred when I tried to execute train_birds.sh. I did not change the file or model.
Please tell me what kind of correction should be made.

environment
windows 10
Anaconda 3.6
Python 3.6
Pytorch 0.3.1
Tensorflow 1.4.1

---

(py36) C:\Users\c0116\Anaconda3\envs\py36\HDGan-master\train\train_gan>sh train_birds.sh device=0
mkdir: created directory '../../Models/HDGAN_256_birds'
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
WARNING:root:Setting up a new session...
....\HDGan\proj_utils\torch_utils.py:23: UserWarning: volatile was removed and now has no effect. Use with torch.no_grad(): instead.
x.volatile = volatile
Traceback (most recent call last):
File "train_worker.py", line 113, in

---

train_gans((dataset_train, dataset_test), model_root, model_name, netG, netD, args)
File "....\HDGan\HDGan.py", line 242, in train_gans
d_plot_dict[key].plot(to_numpy(img_loss).mean())
Launch python -m visdom.server -port 43426 to monitor File "....\HDGan\proj_utils\torch_utils.py", line 41, in to_numpy
return x.cpu().numpy()

---

RuntimeError: Can't call numpy() on Variable that requires grad. Use var.detach().numpy() instead.
Namespace(KL_COE=4, batch_size=10, cuda=True, d_lr=0.0002, dataset='birds', device_id=0, display_freq=200, epoch_decay=100, finest_size=256, g_lr=0.0002, gpus='0', init_256generator_from='', load_from_epoch=0, maxepoch=600, model_name='HDGAN_256', ncritic=1, noise_dim=100, num_emb=4, num_resblock=1, reuse_weights=False, save_freq=5, test_sample_num=4, verbose_per_iter=50, visdom_port=8097)

Init HDGAN Generator
side output at [64, 128, 256]
Init HDGAN Discriminator
Add adversarial loss at scale [64, 128, 256]
Parallel models in [0] GPUS
Init basic data loader train
8855 samples (batch_size = 10)
[64, 128, 256] output resolutions
4 embeddings used
Init basic data loader test
2933 samples (batch_size = 10)
[64, 128, 256] output resolutions
1 embeddings used
Start training ...

[ypxie]

can you try var.detach().numpy()? maybe it's torch version issue.