rel-3.3.0

What's Changed

Major Changes 🛠

• Allow building in free-threaded python by @ricardoV94 in #2286

New Features 🎉

• Add mlx dispatch for Clip by @jessegrabowski in #2324
• Add MLX dispatches for gammaln and psi by @jessegrabowski in #2325
• Rewrite MLX erfc and erfcx on a rational kernel by @jessegrabowski in #2326
• Add an AS241 rational kernel for MLX erfcinv by @jessegrabowski in #2331
• Add MLX i0 and i1, and retire the erf family's float64 exp ceiling by @jessegrabowski in #2332
• Add MLX dispatch for gamma by @jessegrabowski in #2333
• Add MLX dispatch for ive at real order by @jessegrabowski in #2340
• Take the gammaln and psi reflections through the tangent by @jessegrabowski in #2341

Bugfixes 🐛

• Fix linspace retstep dtype handling by @Sanjays2402 in #2304
• Fix SparseDenseMultiply gradient when the dense input is a scalar by @Sanjays2402 in #2319
• Fix AdvancedIncSubtensor runtime-broadcast check crashing on a scalar update by @velochy in #2339
• Make minimun/maximum gradient more robust to numerical precision by @ricardoV94 in #2342

Maintenance 🔧

• Stabilize sigmoid(x) / (1 - sigmoid(x)) -> exp(x) by @ricardoV94 in #2322
• Refactor and improve Pad by @jessegrabowski in #2315
• Use target-sized indices in Numba vectorized codegen by @anutosh491 in #2334

New Contributors

• @Sanjays2402 made their first contribution in #2304
• @anutosh491 made their first contribution in #2334

Full Changelog: rel-3.2.4...rel-3.2.5