v3.0.2

Version 3.0.2

Version 3.0.2 fixes low-precision reductions and regularization, small-batch
regularization, and NumPy sampler inputs. Public signatures and minimum
supported Python, PyTorch, and NumPy versions are unchanged.

Numerical fixes

• ClassWeightedReducer multiplies and averages float16 and bfloat16 losses
  in float32, then returns the scalar in the input dtype. Stored weights keep
  their precision until multiplication, avoiding premature underflow,
  overflow, and rounding
  (#775).
• CenterInvariantRegularizer widens float16 and bfloat16 inputs before
  computing norms and squared deviations. Its reduced scalar remains
  float32 for those inputs so a later regularization weight can scale a
  large finite penalty. This corrects loss values and gradients affected by
  low-precision rounding or overflow; float32 and float64 behavior and the
  configurable Lp norm are preserved
  (#776).
• RegularFaceRegularizer returns differentiable zero for empty and singleton
  inputs. It continues to validate its inputs and retains the existing
  nearest-neighbor objective for two or more centers
  (#777).

Sampler inputs

Samplers preserve categorical and hashable NumPy label values and accept
views with negative strides or non-native byte order. Invalid label
dimensions and hierarchy columns produce descriptive errors at construction
(#778).

Benchmark tooling

Capacity searches report a batch size only after successful confirmation.
They return no verified capacity when every probe or a confirmation fails,
preserve the confirmation failure status, and correctly refine brackets
whose starting batch is not aligned to the search granularity. Invalid
search settings are rejected before launching a probe
(#779).
Previously published benchmark measurements are unchanged.

Contributors

Thanks to @Afloat16 for all five fixes. The
original contribution commits are preserved in this release.