kanadelsy/generalized-warmup-assets

Generalized warm-up: reproduction assets

Assets for Generalized warm-up for confidence calibration of deep neural network classifiers.

Code: https://github.com/LeeShuaiyu/generalized-warmup-release

Contents

• 388 checkpoints in 50 experiment-specific TAR archives.
• gw-results.zip: archived predictions, source measurements and fitted parameters.
• model_packages.json: archive sizes, SHA256 checksums and checkpoint membership.

Use the code repository's download planner to select an experiment. Each model archive contains one experiment. The code repository also records the result archive and per-member checksums.

Dataset sources and acquisition steps are listed in the code repository's data/README.md. Numerical sources are described in docs/NUMERICAL_SOURCES.md.

License

Contributor-owned model weights use MIT. Original numerical result data use CC BY 4.0. See LICENSE for scope, attribution and license texts. Provider materials and pretrained components retain their applicable source terms.