ScaffoldAR

Checkpoint of ScaffoldAR, a fully generative hierarchical diffusion model for polymer conformations (code: https://github.com/XingqiaoLin/poly, folder scaffoldar/).

scaffoldar.pt is the complete model in one file:

Key	Content
scale_diffusion	chain-scale diffusion (mean segment length)
path_diffusion	scale-conditioned global path diffusion
local_diffusion	local torus / SO(3) diffusion of each repeat unit
sampler	default sampler settings (selected on the validation benchmark)

Each model entry holds its weights (model) and training arguments (args); training logs are in *.metrics.jsonl.

Download into the code repository's scaffoldar/checkpoints/ and generate:

pip install -U huggingface_hub
python -c "from huggingface_hub import hf_hub_download; hf_hub_download('Xingqiao-2002/ScaffoldAR', 'scaffoldar.pt', local_dir='checkpoints')"
GPUS="0 1 2 3" bash scripts/generate_test2030.sh outputs/test2030