16 Pull requests merged by 5 people

• LogCumSumExp

  #2069 merged Apr 13, 2025

• Fix release build

  #2072 merged Apr 13, 2025

• Add float mask to sdpa vector

  #2068 merged Apr 12, 2025

• nit: fix exception handling in distributed run

  #2066 merged Apr 11, 2025

• Fix stubgen

  #2065 merged Apr 11, 2025

• fix fft bugs

  #2062 merged Apr 11, 2025

• Fix the test and add custom min/max reductions for uncommon MPI types

  #2060 merged Apr 11, 2025

• Fix MSVC build due to use of M_LN2

  #2058 merged Apr 10, 2025

• Min / max reductions

  #2041 merged Apr 10, 2025

• Do not load the default lib if another is requested

  #2055 merged Apr 9, 2025

• no sdpa in grad

  #2054 merged Apr 9, 2025

• Fix causal mask sdpa vec

  #2053 merged Apr 8, 2025

• Remove Event::Signal()

  #2052 merged Apr 8, 2025

• Only request residency once

  #2051 merged Apr 7, 2025

• add a half simd gemm fallback

  #2046 merged Apr 7, 2025

• Only add to residency set once

  #2049 merged Apr 7, 2025

5 Pull requests opened by 4 people

• Remove static initializers

  #2059 opened Apr 10, 2025

• Search mlx.metallib in macOS framework "Resources" dir

  #2061 opened Apr 10, 2025

• fix: conv_general differences between gpu, cpu

  #2070 opened Apr 12, 2025

• Introduce macros for dispatching dynamic dtypes as static types

  #2073 opened Apr 13, 2025

• Added Short time fourier transform and ISTFT implementations

  #2074 opened Apr 13, 2025

8 Issues closed by 3 people

• MLX default weight init seems incorrect

  #2001 closed Apr 13, 2025

• MultiOptimizer documentation missing

  #2022 closed Apr 11, 2025

• [BUG] Missing type stubs for python 3.11 in version 0.24.2

  #2063 closed Apr 11, 2025

• [BUG] irfft does not produce correct results

  #2034 closed Apr 11, 2025

• Swap support for low ram machines

  #1907 closed Apr 10, 2025

• Tiled cholesky kernel to contribute to your MLX backend

  #2056 closed Apr 9, 2025

• [BUG] float16 and bfloat16 eval on linux build

  #2037 closed Apr 7, 2025

• Request of Implementation of lgamma

  #2030 closed Apr 7, 2025

3 Issues opened by 3 people

• [BUG] thread issues with evaluation

  #2067 opened Apr 11, 2025

• Document the stability and algorithm of sort

  #2057 opened Apr 10, 2025

• Request of implantmentation of lgamma function

  #2050 opened Apr 7, 2025

4 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.

• Missing f8 dtypes

  #1670 commented on Apr 10, 2025 • 0 new comments

• [BUG] conv_general differences between gpu, cpu and linux build

  #2038 commented on Apr 12, 2025 • 0 new comments

• [WIP] CUDA backend

  #1983 commented on Apr 12, 2025 • 0 new comments

• Gather mm new kernel and small refactoring

  #2040 commented on Apr 13, 2025 • 0 new comments