PyTorch
See the Quick Start Guide on how to install and configure the profiler.
To profile PyTorch, add the following code around a code step, e.g. training batch or a prediction. Only some steps will be profiled; the profiler decides which steps to profile for optimal statistics and low overhead. See profiling API reference for full documentation.
Profile PyTorch using with
context manager:
with profile_step():
# training batch, prediction, etc.
Profile using stop
:
from graphsignal.profilers.pytorch import profile_step
step = profile_step()
# training batch, prediction, etc.
step.stop()
Examples
The PyTorch MNIST example illustrates where and how to add the profile_step
method.
Distributed workloads
Graphsignal provides built-in support for distributed training and inference. It is only necessary to provide the same run ID to all workers. Refer to Distributed Workloads section for more information.