Issues

• QZ with output="real" marks AA upper triangular, so det and solve of it return wrong values

  assumptionsbuglinalg
  pymc-devs/pytensor#2619 · Filter by author jessegrabowski opened on Nov 8, 2026
  1

• No rewrite moves an Alloc out of an Assert, so local_blockwise_alloc misses asserted inputs

  graph rewritingnumbavectorization
  pymc-devs/pytensor#2618 · Filter by author jessegrabowski opened on Nov 8, 2026
  1

• Numba LAPACK kernels make a slow C-to-F copy of C-contiguous "A" inputs

  linalgnumbaperformance
  pymc-devs/pytensor#2617 · Filter by author jessegrabowski opened on Nov 7, 2026
  1

• Numba: fused set and inc at the same indices share one write mode

  bugindexingnumba
  pymc-devs/pytensor#2616 · Filter by author jessegrabowski opened on Nov 7, 2026
  1

• BUG: pt.cov raises AxisError on 1-D input (NumPy accepts it as a single variable)

  pymc-devs/pytensor#2614 · Filter by author parth-patel010 opened on Nov 7, 2026
  1

• Rewrite IfElse with a constant condition to its taken branch

  feature requestgraph rewriting
  pymc-devs/pytensor#2610 · Filter by author jessegrabowski opened on Nov 7, 2026
  1

• BUG: take ignores mode when axis is None

  pymc-devs/pytensor#2598 · Filter by author RaphaelFakhri opened on Nov 3, 2026

• BUG: linalg.norm with a 2-tuple axis reduces the wrong axes (or raises) on batched inputs

  pymc-devs/pytensor#2583 · Filter by author parth-patel010 opened on Oct 30, 2026
  1

• JAX shape handling in inner graphs and static argument detection

  backend compatibilityjaxshape inference
  pymc-devs/pytensor#2577 · Filter by author ricardoV94 opened on Oct 29, 2026
  1

• Numba multivariate_normal raises LinAlgError on a NaN covariance

  linalgnumbarandom variables
  pymc-devs/pytensor#2576 · Filter by author jessegrabowski opened on Oct 29, 2026
  1

• BUG: config.change_flags is not thread-safe: concurrent graph construction can permanently clear config.cxx

  bug
  pymc-devs/pytensor#2575 · Filter by author Saethox opened on Oct 28, 2026
  1

• local_subtensor_merge_slice overflows int8 on PyMC random walks longer than 127

  buggraph rewritingindexingNumPy compatibility
  pymc-devs/pytensor#2567 · Filter by author jessegrabowski opened on Oct 25, 2026
  1

• OpFromGraph stops pickling after a gradient because it caches its pullback

  buggraph objectsOpFromGraph
  pymc-devs/pytensor#2563 · Filter by author jessegrabowski opened on Oct 25, 2026

• BUG: geomspace returns wrong sign for the endpoints when start and stop are negative

  pymc-devs/pytensor#2561 · Filter by author JoeyTan21 opened on Oct 24, 2026

• Sparse Jacobian and Hessian detection via per-Op local sparsity rules

  feature requestgradients
  pymc-devs/pytensor#2544 · Filter by author jessegrabowski opened on Oct 21, 2026
  1

• mkl<=2024.0 CI pin installs a libtorch build that won't import on glibc 2.41+

  GitHub CI/CDtorch
  pymc-devs/pytensor#2539 · Filter by author jessegrabowski opened on Oct 20, 2026

• pytensor-cache CLI: cleanup and improvements

  pymc-devs/pytensor#2536 · Filter by author williambdean opened on Oct 20, 2026

• ENH: consistent reporting and exit codes for pytensor-cache clear and unlock

  pymc-devs/pytensor#2534 · Filter by author williambdean opened on Oct 20, 2026
  3

• ENH: make pytensor-cache list pipeable (one record per line)

  pymc-devs/pytensor#2533 · Filter by author williambdean opened on Oct 20, 2026

• ENH: return cache introspection data instead of printing it

  pymc-devs/pytensor#2532 · Filter by author williambdean opened on Oct 20, 2026

• ENH: pytensor-cache should support --help/-h and standard CLI expectations

  pymc-devs/pytensor#2530 · Filter by author williambdean opened on Oct 20, 2026

• ge(exp(x), 0) is not simplified

  graph rewritingnew feature
  pymc-devs/pytensor#2526 · Filter by author jessegrabowski opened on Oct 19, 2026
  3

• MLX mode never reaches the mx.addmm dispatches because BlasOpt is excluded

  mlxperformance
  pymc-devs/pytensor#2525 · Filter by author jessegrabowski opened on Oct 19, 2026

• PyTorch Max/Min/Prod violate empty-axis ownership; Prod ignores accumulator dtype

  pymc-devs/pytensor#2521 · Filter by author cetagostini opened on Oct 19, 2026
  1

• MLX ndarray typification makes a full NumPy copy for Fortran-contiguous matrices

  pymc-devs/pytensor#2520 · Filter by author cetagostini opened on Oct 19, 2026