0.1.1

This is bug-fix release to 0.1.0 to resolve #80 (Training does not honor custom 25-parameter data.noise_model during training process).

What's Changed

• fix #66/docs: add Windows & Blackwell troubleshooting by @kvmto in #67
• docs: add training recommendations for epochs, shots, and noise upscaling by @ivanbasov in #72
• fix(timing): warmup pass before timing loop to amortise torch.compile JIT by @ivanbasov in #70
• Add publication reference by @ivanbasov in #74
• ci: avoid deadsnakes in GPU workflows by @ivanbasov in #78
• ci: avoid flaky GPU apt cache by @ivanbasov in #79
• [bug fix] Fix 25p noise and noise upscaling by @bmhowe23 in #75

Full Changelog: 0.1.0...0.1.1

Initial Release - AI-based pre-decoders for surface codes

This is an open-source release of the training framework and inference framework used for [1] and [2]. See the README.md and cookbook/predecoder.ipynb for full details on how to explore the current implementation and begin extending with your own noise models. This release can be paired with the fast and accurate models on Hugging Face!

Additionally, this can be paired with CUDA-Q QEC 0.6.0 to run a trained decoder on NVQLink!

[1] https://research.nvidia.com/publication/2026-04_fast-ai-based-pre-decoders-surface-codes
[2] https://arxiv.org/abs/2604.12841