AntixK/PyTorch-Model-Compare

Implements Centered Kernel Alignment (CKA), a representation similarity metric based on the Hilbert-Schmidt Independence Criterion, with a minibatch-scalable version for comparing deep architectures. Works with any PyTorch `nn.Module` across diverse model families—CNNs, Vision Transformers, and models from torchHub/timm/HuggingFace—enabling ablation studies, architecture analysis, and dataset drift detection through per-layer feature similarity heatmaps.

379 stars. No commits in the last 6 months.