Expand description
Token-level KL divergence operations (exact and Schulman 2020 estimator).
The impl_kl_ops! macro generates both f64_ops and f32_ops sub-modules,
each exposing the same function set for their respective float type. The
f64 variants are re-exported at module level for ergonomic use as the default.
Re-exports§
pub use f64_ops::*;