The problem of multitasking parallel processing

[Pter61]

While I was training in 2 GPU(V100 16G), this error occurred in the middle of the second epoch
...
File: ".../python3.6/site-packages/torch/autograd/inti/.py", line93, in backward allow_unreachable=True) # allow_unreachable flag
RuntimeRrror: element 0 of tensors does not require grad and does no have grad_fn

However, I can continue to train in a single GPU.
Could you please help me deal with this problem?

[alice-cool]

Could you help me the #include<torch.extension.h> didn't have the file?

[jackroos]

@Pter61 What config do you run? And what is the full command you ran? And please provide your pytorch version and cuda version also.

[alice-cool]

I just follow the readme'' Environment to run conda create -n...&nbsp; conda install...&nbsp; git clone(apex)but it throws exception pop install Cython -r requiremen t ./scripts/init.sh （It has errors, which didn't know #include<torch.extension.h&gt;） Thanks for you attention

…

---Original--- From: "Weijie Su"<[email protected]&gt; Date: Wed, Dec 30, 2020 11:53 AM To: "jackroos/VL-BERT"<[email protected]&gt;; Cc: "Comment"<[email protected]&gt;;"Donglearner"<[email protected]&gt;; Subject: Re: [jackroos/VL-BERT] The problem of multitasking parallel processing (#68) @Pter61 What config do you run? And what is the full command you ran? — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.

[alice-cool]

I have solved the problem. Thanks