Expand description
HDF5 (.h5 / .hdf5) and Keras v3 (.keras) checkpoint loading.
HDF5 files are walked recursively: every numeric dataset becomes a tensor
named by its group path with / separators replaced by . (e.g.
/model_weights/dense/kernel:0 → model_weights.dense.kernel:0).
F32 and F16 datasets are preserved exactly; F64, integer, and bool datasets
are cast to F32 with a note; non-numeric datasets (strings, compounds) are
skipped with a note.
.keras archives are zip files wrapping model.weights.h5 — the embedded
weights file is extracted to a temp location and parsed the same way.