torchexplorer
torchexplorer copied to clipboard
spfrommer
Interactively inspect module inputs, outputs, parameters, and gradients.
The line ```python assert metadata['module'] == current_module ``` should throw a better error message.
Gathering data from https://pytorch.org/tutorials/recipes/recipes/profiler_recipe.html ...it would be fantastic if there was a library with a one-line API comparable to what Tensorboard previously offered with TensorFlow, for color-coded graph visualization of...
The fonts are different somehow. Should get compensated for in the graphviz call.
For networks with many residual connections the graphviz layout can contain a lot of unnecessary whitespace.
Not sure what's the best way to implement. There's a few difficult things to handle: * The binning: would probably have to do a super-fine binning (100s of bins), then...
User should be reminded to switch to a newer interface.
Specifically the floating point numbers can probably be truncated to improve performance. Also check for other potential performance boosts.
Can be seen on the AttachModule demo, the x axis says 0.0e+0 instead of 1
Add dotted / dashed lines for median and 20th / 80th percentiles. Will probably involve removing the current zero line and making it a tick on the side.