PyTensor Files

What's Changed

Major Changes 🛠

• Drop support for Python 3.11 by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2211
• Rewrite ShapeFeature to not hold live variables by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2056
• Remove symbolic axis from join/split by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2144
• Deprecate implicit shared-variable capture in OpFromGraph by @alessandrogentili001 in https://github.com/pymc-devs/pytensor/pull/2047
• Make inner graph Ops immutable by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2243

New Features 🎉

• Add asumption for unique indices by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2225

Bugfixes 🐛

• Fix until scan gradient step count with recurring outputs by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2217
• Fix assumption bug with (1,1) eye by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2224
• Fix ScanMerge cycle by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2226
• Fix numba blockwise layout inference by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2234
• Fix MLX variadic Elemwise crash on broadcast-shaped inputs by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2235
• Make exp/log rewrites dtype-safe under cast_policy="numpy+floatX" by @maresb in https://github.com/pymc-devs/pytensor/pull/2247
• Fix MLX IncSubtensor with slice index by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2240
• Keep _make_scalar_loop updates dtype-stable by @maresb in https://github.com/pymc-devs/pytensor/pull/2249
• Accumulate duplicate indices in MLX AdvancedIncSubtensor by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2242
• Numba scan: fix invalid cross alias and inplace by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2263

Documentation 📖

• Expand documentation on broadcasting limitations by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2084
• Tiny transformer LLM example by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2163

Maintenance 🔧

• Numba: Cache cython-special functions by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2212
• Add MLX GEMV dispatch by @drbenvincent in https://github.com/pymc-devs/pytensor/pull/2214
• Add MLX GER dispatch by @drbenvincent in https://github.com/pymc-devs/pytensor/pull/2216
• Numba: Speedup RV eval by casting float parameters to float64 by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2223
• Do not recompile identical OFG in same graph by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2209
• Exclude general fusion from JAX and MLX modes by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2236
• Fix local_log_div float64 precision loss for integer-literal operands by @maresb in https://github.com/pymc-devs/pytensor/pull/2245
• Fix MLX Shape_i output type by @gaoflow in https://github.com/pymc-devs/pytensor/pull/2230
• Refactor BLAS code by @jessegrabowski in https://github.com/pymc-devs/pytensor/pull/2049
• Generalize local_sum_prod_alloc by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2198
• Fold variadic Add/Mul in JAX and PyTorch backends instead of stacking by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2238
• Add gradient (pullback) to Min Op by @hermesdt in https://github.com/pymc-devs/pytensor/pull/2253
• Fix MLX Metal codegen crash on NaN scalar constants by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2239
• Enforce AdvancedIncSubtensor1 runtime-broadcast check on MLX backend by @cetagostini in https://github.com/pymc-devs/pytensor/pull/2241
• Numba scan: Allow inplacing on discarded mit-mot by @ricardoV94 in https://github.com/pymc-devs/pytensor/pull/2266

New Contributors

• @drbenvincent made their first contribution in https://github.com/pymc-devs/pytensor/pull/2214
• @gaoflow made their first contribution in https://github.com/pymc-devs/pytensor/pull/2230
• @hermesdt made their first contribution in https://github.com/pymc-devs/pytensor/pull/2253

Full Changelog: https://github.com/pymc-devs/pytensor/compare/rel-3.0.5...rel-3.1.0