Changelog

v0.10.0-alpha.2 (2025-11-15)

Bump candle requirement to 0.9.2-alpha.1: this adds support for CUDA 13.0 via cudarc 0.17.1+
Use published alpha versions from crates.io instead of git revisions, allowing this version to be published to crates.io

Bump candle requirtement to 0.9.0: this is considered a breaking change due to the reliance of this library on candle-core and candle-nn

Bump candle requirtement to 0.8.0: this is considered a breaking change due to the reliance of this library on candle-core and candle-nn

Bump candle requirtement to 0.5.0: this is considered a breaking change due to the reliance of this library on candle-core and candle-nn
Internal changes for LBFGS line search

Bump candle requirtement to 0.4.0: this is considered a breaking change due to the reliance of this library on candle-core and candle-nn
Explicit reliance on the candle crates hosted on crates.io : as cargo does not support git dependecies in published crates, this library now points only to the crates.io releases (previously cargo would default to the crates.io instead of git repo anyway: if the git repo is specifically desired this can be obtained by patching the Cargo.toml to point at the candle repo)
Remove intel-mkl feature: features in this library are mainly used for running examples: any code that uses this library should instead use the features directly from the candle crates

Improved Documentation
Add ability to set more optimiser parameters (see issue huggingface/candle#1448 regarding LR schedulers in candle)
All params are now Clone, PartialEq and PartialOrd