ml-explore/mlx

Issues

• Supported python versions?

  #2629 opened a month ago by jrp2014
  1

• [BUG] [CUDA] Compiled kernel fails with resource issue

  #2628 opened a month ago by awni
  0

• JIT compilation prevents optimization of GEMM with bias

  #2585 opened a month ago by zcbenz
  3

• ⚠️ SentinelOne EDR Quarantines `distributed_run.py` Due to Network & Interface Modifications

  #2625 opened a month ago by joshmiller83
  2

• SDPA error with masked full attention

  #2626 opened a month ago by c0g
  4

• [BUG] make error on Amazon Graviton3: aarch64, Ubuntu 24.04.3 LTS

  #2624 opened a month ago by jli05
  2

• [BUG] MPI backend with working passwordless SSH hosts, "No route to host"

  #2540 opened a month ago by muellerzr
  2

• arm neon not detected: sometimes the cpu flag is "asimd" instead

  #2623 opened a month ago by jli05
  0

• How to profile mlx programs?

  #2597 opened a month ago by ZelinMa557
  4

• Make `zeros_like` work with shapeless compile / export.

  #2599 opened 2 months ago by josharian
  3

• [Question] Best way to compile to a single file executable

  #2615 opened 2 months ago by bitanath
  1

• [BUG] Learning Rate Scheduler Bug: Cosine Decay Starts During Warmup Phase

  #2617 opened 2 months ago by yuemingruoan
  1

• Qwen3 LoRA Training Has Significantly Lower Trainable Parameters Than Expected

  #2616 opened 2 months ago by yuemingruoan
  1

• [BUG] Performance regression when yaml opening many files

  #2583 opened 2 months ago by awni
  2

• [Feature]Qwen3-Next support?

  #2589 opened 2 months ago by bright8192
  3

• [Feature] Possible to have a clear pipeline to coreml conversion?

  #2610 opened 2 months ago by bitanath
  1

• [BUG] shapeless matmul isn't

  #2607 opened 2 months ago by josharian
  1

• [Feature] block GPU access on iOS simulator

  #2605 opened 2 months ago by davidkoski
  2

• should I add relu2 to mlx-python

  #2579 opened 2 months ago by Goekdeniz-Guelmez
  2

• [Documentation] Shapeless only means sizes change but not number of dimensions

  #2596 opened 2 months ago by umbertomig
  2

• [Feature Request] Option to return attention weights from fast.scaled_dot_product_attention

  #2590 opened 2 months ago by magic-YuanTian
  0

• Convert gguf weight to safetensor format

  #2593 opened 2 months ago by JunchengYan
  0

• Performance discrepancy vs torch on cuda

  #2566 opened 2 months ago by arogozhnikov
  9

• Add ROCm Support for AMD GPUs

  #2556 opened 2 months ago by MrDewitt88
  1

• [BUG] Provide CUDA implementation of FFT

  #2561 opened 2 months ago by ekamperi
  8

• [BUG] MLX 0.29 causes pylint failure

  #2575 opened 2 months ago by mdales
  1

• [BUG] Missing `Union, Optional, Tuple, etc.` imports in `__init__.pyi` leads to Unknown type hints

  #2478 opened 2 months ago by XXXXRT666
  3

• Add capability of read+write in input arguments of custom kernels

  #2547 opened 2 months ago by spichardo
  3

• [BUG] Linux: ImportError: libmlx.so: cannot open shared object file: No such file or directory

  #2500 opened 3 months ago by anthonywu
  7

• [BUG] [CUDA] Error in graph capture for argpartition

  #2537 opened 2 months ago by awni
  1

• [BUG] Enable multiprocessing queues with mx.array

  #2457 opened 3 months ago by awni
  1

• [feature request] `mlx.core.load` and `mlx.core.save` should accept `pathlib.Path`

  #2522 opened 2 months ago by neilmehta24
  1

• [BUG] `prepare_tb_ring` gives KeyError: 'domain_uuid_key'

  #2538 opened 2 months ago by muellerzr
  1

• [BUG] RuntimeError: QuantizedMatmul has no CUDA implementation

  #2536 opened 2 months ago by ehartford
  1

• [BUG] Cannot install mlx-coda - InvalidRecordEntry `hash` does not follow the required format

  #2529 opened 3 months ago by markito
  1

• [BUG] Bug in Metal conv2D

  #2504 opened 3 months ago by awni
  0

• [CUDA] mx.eval() fails when evaluating ConvTranspose2d gradients

  #2516 opened 3 months ago by ekamperi
  2

• [BUG] mlx-cuda requirements all met: `RuntimeError: cudaMemAdvise`

  #2519 opened 3 months ago by anthonywu
  1

• Missing supports on ARM CPUs nvidia jetson / Ampere Altra

  #2469 opened 3 months ago by 6thday
  1

• Use cibuildwheel to build wheels for all Python versions, platforms and architectures

  #2497 opened 3 months ago by reneleonhardt
  0

• Feature Request: Add AdaBelief Optimizer

  #2479 opened 3 months ago by jganbar
  0

• [BUG] Segmentation fault with ArrayAt object

  #2503 opened 3 months ago by ghcollin
  0

• Suggestion: Update compile examples to better showcase performance benefits

  #2509 opened 3 months ago by swartz-k
  1

• [BUG] Strides are zero following `eval` of result of `reciprocal`

  #2501 opened 3 months ago by stemann
  6

• [BUG] mlx doesn't compile on a MacBook Pro

  #2498 opened 3 months ago by jrp2014
  1

• [BUG] `mx.compile` raise Error with custom Module

  #2481 opened 3 months ago by XXXXRT666
  1

• [QUESTION] How to Quantize a Mostly Convolutional model to fit in memory? Or how to cast it to bfloat16 for lower memory requirements?

  #2490 opened 3 months ago by bitanath
  4

• [BUG] Strange results when calculating the mean and std of integer array

  #2475 opened 3 months ago by Antares95
  0

• [BUG] CommandEncoder::outputs() returns copy instead of reference

  #2467 opened 3 months ago by ethereon
  0

• Add `#include <algorithm>` to `jit_compiler.cpp` for Linux builds

  #2459 opened 3 months ago by zamderax
  0