Help！RuntimeError: a leaf Variable that requires grad is being used in an in-place operation.

[doqizuo]

creating data loader...
creating model and diffusion...
training...
Traceback (most recent call last):
File "scripts/segmentation_train.py", line 118, in
main()
File "scripts/segmentation_train.py", line 70, in main
TrainLoop(
File "D:\MedSegDiff-master.\guided_diffusion\train_util.py", line 83, in init
self._load_and_sync_parameters()
File "D:\MedSegDiff-master.\guided_diffusion\train_util.py", line 139, in _load_and_sync_parameters
dist_util.sync_params(self.model.parameters())
File "D:\MedSegDiff-master.\guided_diffusion\dist_util.py", line 111, in sync_params
dist.broadcast(p, 0)
File "D:\Anaconda\envs\sg\lib\site-packages\torch\distributed\distributed_c10d.py", line 1195, in broadcast
work.wait()
RuntimeError: a leaf Variable that requires grad is being used in an in-place operation.

Plz！Tell me where the problem lies? and why? Hope someone nice can help me~ Thanks!!!

[Eins152]

Try add p = p + 0 in the sync_params function within dist_util.py as follows:
def sync_params(params):
"""
Synchronize a sequence of tensors across ranks from rank 0.
"""
for p in params:
with th.no_grad():
p = p + 0
dist.broadcast(p, 0)

[Eins152]

#84

[Issues-translate-bot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

#84

[doqizuo]

Try add p = p + 0 in the sync_params function within dist_util.py as follows: def sync_params(params): """ Synchronize a sequence of tensors across ranks from rank 0. """ for p in params: with th.no_grad(): p = p + 0 dist.broadcast(p, 0)

THANKSSSSSS!!