torchexplorer
TorchExplorer facilitates detailed analysis of neural networks by offering an interactive inspection of the inputs, outputs, parameters, and gradients of each nn.Module during training. Compatible with weights and biases or operable as a standalone solution, it enables the visualization of intermediate tensors, examination of model structures, and tracking of network dynamics, such as gradient and parameter distributions. Easily installable on Linux and Mac, it assists in debugging issues like vanishing gradients or uneven module input distribution, providing a clear graphical interface for enhanced network understanding.
