Speed up nnUnet inference#110

Merged

Hendrik-code merged 22 commits into

mainfrom

development_robert

Jun 10, 2026

Commits on May 27, 2026

Commits on May 28, 2026

update memory requirements
robert-graf
committed

Commits on Jun 4, 2026

perf(nnunet): enable cuDNN autotune + TF32 for sliding-window inference

Hendrik-code
and
claude
committed
perf(nnunet): use torch.inference_mode() instead of no_grad() for inference

Hendrik-code
and
claude
committed
perf(nnunet): stop clearing CUDA cache on the per-fold happy path

Hendrik-code
and
claude
committed
fix(nnunet): repair fold ensembling broken by loaded_networks cache

Hendrik-code
and
claude
committed
perf(nnunet): add opt-in persistent model cache (cache_model)

Hendrik-code
and
claude
committed
perf(nnunet): optionally batch sliding-window tiles (tile_batch_size)

Hendrik-code
and
claude
committed
bench: add nnU-Net inference timing harness

Hendrik-code
and
claude
committed

Commits on Jun 8, 2026

Commits on Jun 9, 2026

fix bug made by claude
robert-graf
committed