search

syoamakase / ASR

Speech Recognition Toolkit
8 stars 0 forks source link

RNN warning? #2

Open syoamakase opened 4 years ago

syoamakase commented 4 years ago

Environment Python==3.7.4 torch==1.5.0.dev20200204

"User warning" displays per iteration. I consider that this is a PyTorch problem.

/opt/conda/conda-bld/pytorch_1580803991946/work/aten/src/ATen/native/cudnn/RNN.cpp:1232: UserWarning: RNN module weights are not part of single contiguous chunk of memory. This means they need to be compacted at every call, possibly greatly increasing memory usage. To compact weights again call flatten_parameters().

Although it's not a problem for the performance, it's hard to trace the training progress.

A simple solution is that we add warnings.simplefilter('ignore'). However, the program also ignores another warning.

syoamakase commented 4 years ago

And when I used multi-gpu mode in torch==1.5.0, the performance was very very BAD You should use torch<=1.3.0.