A simple interface to load checkpoints from the TopoNets Paper (ICLR 2025). We closely followed the standard training procedures for these models. The only difference being the addition of topoloss to the original loss.
All of the checkpoints are hosted on the MurtyLab HuggingFace 🤗
Start by installing this repo as a module:
pip install git+https://github.com/toponets/toponets.git
Then load the pre-trained checkpoints like this:
import toponets
topo_resnet18 = toponets.resnet18(tau=10.0, checkpoint_path = f"resnet18_tau_{tau}.pt")
topo_resnet50 = toponets.resnet50(tau=30.0, checkpoint_path = f"resnet50_tau_{tau}.pt")
topo_vit_b_32 = toponets.vit_b_32(tau=10.0, checkpoint_path = f"vit_b_32_tau_{tau}.pt")
topo_nanogpt = toponets.nanogpt(tau=50.0, checkpoint_path = f"nanogpt_tau_{tau}.pt")Check out example.py for more examples.