4 Releases published by 1 person

• rel-2.30.0

  published Mar 27, 2025

• rel-2.30.1

  published Mar 30, 2025

• rel-2.30.2

  published Apr 1, 2025

• rel-2.30.3

  published Apr 9, 2025

22 Pull requests merged by 10 people

• Add Ops for LU Factorization

  #1218 merged Apr 19, 2025

• Fix typo in warning message

  #1364 merged Apr 12, 2025

• Update pyproject license schema according to deprecation warnings

  #1362 merged Apr 9, 2025

• Fix bug in infer_shape of Blockwise(Subtensor)

  #1353 merged Apr 9, 2025

• Allow specializing shape of predefined tensor types

  #1355 merged Apr 9, 2025

• Inplace Composite and ScalarLoop Ops with multiple outputs

  #1322 merged Apr 8, 2025

• correct comparison in env yaml

  #1349 merged Apr 8, 2025

• Propagate static output shapes in Split and avoid copy in C-impl

  #1343 merged Apr 8, 2025

• Support numba compiled sort and argsort functions

  #1309 merged Apr 4, 2025

• More stable fix for JAX Multinomial

  #1339 merged Apr 1, 2025

• Don't run local uint constant indices in C/Python backends

  #1335 merged Apr 1, 2025

• Fix indexing in convolve1d with mode="same"

  #1337 merged Apr 1, 2025

• Fix nan for valid parameters in jax implementation of Multinomial

  #1328 merged Mar 30, 2025

• Implement batched convolve1d

  #1318 merged Mar 27, 2025

• Implement tridiagonal solve in numba backend

  #1311 merged Mar 27, 2025

• Implement gradient for QR decomposition

  #1303 merged Mar 27, 2025

• Implement faster Multinomial JAX dispatch

  #1316 merged Mar 25, 2025

• Fix get_vector_length incorrectly returning for shared variable without static shape

  #1295 merged Mar 24, 2025

• Provide static output shape for constant arange

  #1302 merged Mar 24, 2025

• Include Op in message when raising NotImplementedError from grad

  #1314 merged Mar 24, 2025

• Simplify dispatch of JAX random variables by handling rng split automatically

  #1315 merged Mar 24, 2025

• Add scipy-like kn helper

  #1313 merged Mar 22, 2025

14 Pull requests opened by 6 people

• Remove `value` getter method from `Constant`

  #1317 opened Mar 23, 2025

• Use `shell=False` for `Popen` on Windows

  #1324 opened Mar 27, 2025

• Cache numba stuff

  #1326 opened Mar 28, 2025

• Simplify Numba implementation of Alloc

  #1329 opened Mar 28, 2025

• Make exceptions less verbose by default

  #1330 opened Mar 30, 2025

• Bump pypa/cibuildwheel from 2.23.0 to 2.23.2

  #1331 opened Mar 31, 2025

• Reuse output of Join in C backend

  #1340 opened Apr 1, 2025

• Hardcode common Op parametrizations to allow numba caching

  #1341 opened Apr 2, 2025

• Speedup AdvancedSubtensor1 and AdvancedIncSubtensor1 in C backend

  #1346 opened Apr 7, 2025

• Reduce pytensor function call overhead

  #1351 opened Apr 8, 2025

• Don't run the PyPi job all the time

  #1363 opened Apr 9, 2025

• MLX backend POC

  #1365 opened Apr 11, 2025

• Extend verify_grad to complex gradient

  #1367 opened Apr 14, 2025

• Avoid PyTensor function overhead in OpFromGraph

  #1375 opened Apr 19, 2025

12 Issues closed by 4 people

• Implement specialized inplace rewriter for Elemwise Composite Ops

  #138 closed Apr 8, 2025

• BUG: clang linker fails

  #1347 closed Apr 7, 2025

• BUG: `convolve1d` `mode="same"` indexing error

  #1336 closed Apr 1, 2025

• Nans in JAX multinomial dispatch

  #1327 closed Mar 30, 2025

• Implement gradient for QR decomposition

  #1099 closed Mar 27, 2025

• Implement faster Multinomial JAX dispatch

  #868 closed Mar 25, 2025

• `get_vector_length` incorrectly returns for shared variable without static shape

  #1239 closed Mar 24, 2025

• Cholesky returns truncated integer outputs for integer inputs

  #1071 closed Mar 24, 2025

• `pt.arange(int)` has `.type.shape` set to `None`

  #1301 closed Mar 24, 2025

• Include Op in message when raising NotImplementedError from grad

  #711 closed Mar 24, 2025

• Simplify dispatch of JAX random variables by handling rng split automatically

  #1204 closed Mar 24, 2025

• Add helper for `special.kn` that just calls `special.kv` under the hood

  #1308 closed Mar 22, 2025

27 Issues opened by 5 people

• Solve to Solve LU optimization

  #1374 opened Apr 19, 2025

• Rename to power for numpy

  #1373 opened Apr 18, 2025

• Rename to `where` to match numpy api

  #1372 opened Apr 18, 2025

• Rename to `subtract` to make numpy API

  #1371 opened Apr 18, 2025

• Rename to `multiply` to match numpy

  #1370 opened Apr 18, 2025

• Use `And` over `AND`

  #1369 opened Apr 18, 2025

• Is `get_target_language` needed

  #1368 opened Apr 18, 2025

• Implement gradient support for complex types

  #1366 opened Apr 14, 2025

• Use numpy C-API for nonzero

  #1361 opened Apr 9, 2025

• Use numpy C-API arange implementation

  #1360 opened Apr 9, 2025

• Use numpy C-API for C implementations of Sort/Argsort

  #1359 opened Apr 9, 2025

• Use numpy Correlate2 for Convolve1d implementation

  #1358 opened Apr 9, 2025

• Remove BatchedDot and provide C implementation for batched matmul that uses numpy C-API

  #1357 opened Apr 9, 2025

• Use numpy C-API einsum for unoptimized Einsum

  #1356 opened Apr 9, 2025

• Extend tracks functionality for Composite Ops

  #1354 opened Apr 9, 2025

• Use PyCapsule in CThunks instead of numpy compatibility layer

  #1352 opened Apr 8, 2025

• Implement all Ops in `mlx`

  #1350 opened Apr 8, 2025

• Disable runtime broadcasting in indexing operations

  #1348 opened Apr 8, 2025

• User facing split does not match numpy

  #1344 opened Apr 3, 2025

• BUG: Compatibility Issue for pytensor on macOS 15.4

  #1342 opened Apr 2, 2025

• Sparse `MulSD` Op gradient raises when dense is scalar

  #1338 opened Apr 1, 2025

• Do not allow passing None for arbitrary arguments in Function

  #1332 opened Mar 31, 2025

• Numpy 2.x C implementation for AdvancedIncSubtensor1

  #1325 opened Mar 28, 2025

• Add Ops for specialized dot products with structured matrices

  #1323 opened Mar 26, 2025

• Add correlate helper function

  #1321 opened Mar 25, 2025

• Consider using fft for convolution to reduce cost of multiple convolutions with shared inputs

  #1320 opened Mar 24, 2025

• Convolve1d rewrites

  #1319 opened Mar 24, 2025

12 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Implement several subtensor lift rewrites

  #1158 commented on Apr 6, 2025 • 11 new comments

• Refactor `infer_shape` method of Ops to find output shapes using `gufunc_signature`

  #1294 commented on Mar 24, 2025 • 3 new comments

• Add option to print inner graphs in debugprint function

  #1293 commented on Mar 24, 2025 • 1 new comment

• Use asv instead of pytest-benchmark

  #1122 commented on Mar 24, 2025 • 0 new comments

• Name check on `pt.tensor` is too strict

  #1102 commented on Mar 31, 2025 • 0 new comments

• ENH: Implement COLA library rewrites for linear algebra functions

  #573 commented on Apr 10, 2025 • 0 new comments

• Speed up test times :rocket:

  #1124 commented on Apr 21, 2025 • 0 new comments

• Restore fast convolution Ops, rewrites, and docs

  #548 commented on Mar 24, 2025 • 0 new comments

• Run whole test suite on numba backend

  #811 commented on Apr 2, 2025 • 0 new comments

• [pre-commit.ci] pre-commit autoupdate

  #1095 commented on Apr 21, 2025 • 0 new comments

• Modify np.tri Op to use _iota instead

  #1276 commented on Mar 23, 2025 • 0 new comments

• Simplify subtensor shape inference

  #1299 commented on Mar 25, 2025 • 0 new comments