L-electron-Rare/mlx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Commit Graph

  • 257c422a8c Find system-installed cuDNN on Windows (#3052) Cheng 2026-01-25 12:24:22 +09:00
  • 1935ab4452 Faster two pass sdpa (#3023) Awni Hannun 2026-01-24 14:16:33 -08:00
  • 617fd9cbbd Use C++20 (#3050) Cheng 2026-01-24 08:48:41 +09:00
  • 8e93b7448c Fix some MSVC compilation errors (#3048) Cheng 2026-01-24 07:56:56 +09:00
  • fd27829efa Build and test python package on Windows CI (#3049) Cheng 2026-01-24 07:22:36 +09:00
  • dc81c1503a Add missing <algorithm> include to buffer_cache.h (#3053) Anri Lombard 2026-01-23 21:52:36 +02:00
  • 9bac6f8584 Allow take on empty array when it makes sense (#3046) Awni Hannun 2026-01-23 07:25:46 -08:00
  • 1650c4905a Link with prebuilt OpenBLAS and fix shared libs build on Windows (#3036) Cheng 2026-01-23 11:17:26 +09:00
  • becc769012 CUDA gather mv (#3039) Angelos Katharopoulos 2026-01-22 17:20:48 -08:00
  • 687508dd98 win: symbol exports and minor fixes (#3024) Daniel Hiltgen 2026-01-22 17:16:22 -08:00
  • c46c3833ee Use cuda::std for math ops (#3041) Cheng 2026-01-23 08:38:26 +09:00
  • faea3e6d34 Turn nccl_stub into a normal target (#3037) Cheng 2026-01-23 08:12:31 +09:00
  • d98776e190 Columnwise quantize (#2989) Anastasiia Filippova 2026-01-22 15:08:56 +01:00
  • b2f86214bb Remove xmlrunner from macOS CI (#3032) Cheng 2026-01-22 08:06:28 +09:00
  • f28f9f0155 build 26.0 release in actions (#3035) Awni Hannun 2026-01-21 14:04:14 -08:00
  • 0d698bc9a5 Handle data smaller than BUFFER_SIZE in jaccl recv (#3033) rltakashige 2026-01-21 21:44:41 +00:00
  • 1d56dfdf59 Use higher precision for linspace with double (#3029) Awni Hannun 2026-01-21 06:20:50 -08:00
  • 9a277a277a PR 3007 Fix Seg Fault (#3008) Dan Anderson 2026-01-21 00:39:15 -05:00
  • 8017d438a9 [CUDA] Faster grouped mm (#3011) Cheng 2026-01-21 09:30:12 +09:00
  • 634b148dd4 Optimize erf function with expm1f in Metal backend (#3025) Robert 2026-01-20 18:57:12 -05:00
  • bfd62a50f4 Windows CI (#3021) Cheng 2026-01-21 08:06:32 +09:00
  • 83bb7891db Fix negative dim indexing (#2994) Dan Anderson 2026-01-20 09:24:33 -05:00
  • 65b42c8476 Do not give workflow boolean inputs default values (#3014) Cheng 2026-01-20 15:27:14 +09:00
  • 0b25c9c06c Do not clear disk space in setup-linux (#3013) Cheng 2026-01-20 07:22:19 +09:00
  • 46d0fdc5ec Type Enhancement for Func Transforms and Bug Fix (#3003) XXXXRT666 2026-01-19 22:19:57 +00:00
  • d96a2bdf57 Fix python package install path in stubgen (#3009) Cheng 2026-01-19 09:34:02 +09:00
  • 9052f678b3 Update CCCL to v3.1.3 (#3012) Cheng 2026-01-19 07:50:09 +09:00
  • ca14d3d835 Fix sharding of quantized models with non-power-of-2 bits (#3006) Tarjei Mandt 2026-01-18 10:21:56 -05:00
  • d2bef3c6bb fix distributed all_to_sharded bias shard axis from -2 to -1 (#2987) gufengc 2026-01-17 22:51:42 +08:00
  • 12f766efe9 Fix edge checking in attention mask attn-mask-fix Jagrit Digani 2026-01-15 14:10:24 -08:00
  • 097abca156 Fix edge checking in attention mask Jagrit Digani 2026-01-15 14:05:54 -08:00
  • 3fe7794f22 Reverts changing the MLX_IBV_DEVICES to MLX_JACCL_DEVICES (#2999) Angelos Katharopoulos 2026-01-14 15:44:17 -08:00
  • 47430159fc Fix fence (#2998) Awni Hannun 2026-01-14 11:59:09 -08:00
  • 2469fc2939 patch bump for next release (#2991) Awni Hannun 2026-01-14 08:46:09 -08:00
  • ac26a4cc0d Allow some non 2D inputs in qqmm (#2981) v0.30.3 Awni Hannun 2026-01-13 15:48:30 -08:00
  • 099dcc0f4c Expose to/from fp8 in Python and don't auto-convert fp8 when loading from safetensors (#2985) Awni Hannun 2026-01-13 15:48:21 -08:00
  • 8654b8281d Don't try to use NAX at run-time if kernels aren't there (#2982) Awni Hannun 2026-01-13 15:47:45 -08:00
  • 4160ec10f7 Fix RandomBits::is_equivalent to include width (#2978) MillaFleurs 2026-01-13 15:42:37 -05:00
  • a8197795f5 replace MLX_IBV_COORDINATOR with MLX_JACCL_COORDINATOR (#2986) Evan Quiney 2026-01-13 19:26:25 +00:00
  • 7b1c46982a fix doc (#2988) CCYeh 2026-01-12 22:33:26 +01:00
  • edab937248 Add asarray to __array_namespace__ (#2966) Anri Lombard 2026-01-12 16:16:27 +02:00
  • 46ee0e9068 Fix grid_dim_x calculations (#2980) CCYeh 2026-01-12 15:16:05 +01:00
  • 43341e8d53 Swizzle scales (#2979) Anastasiia Filippova 2026-01-11 00:32:54 +01:00
  • 07de3da0b9 Use uv remove-conda Angelos Katharopoulos 2026-01-08 14:01:53 -08:00
  • 281afc8ac3 Remove conda Angelos Katharopoulos 2026-01-08 13:43:36 -08:00
  • 1596839256 fix array allocator with user buffer and deleter (#2971) Ronan Collobert 2026-01-07 10:08:22 -08:00
  • 503731727d QQ linear (#2931) Anastasiia Filippova 2026-01-05 20:20:54 +01:00
  • 1680b6fe38 fix numpy dtype bug (#2960) Awni Hannun 2026-01-05 11:20:40 -08:00
  • 1df6c2a009 Fix doc issues in mlx.nn.init.he_normal and mlx.nn.hard_tanh (#2968) 1ndig0 2026-01-05 23:23:41 +08:00
  • 8de9ceb7d6 BUG FIX - Addition of missing parameter in random::uniform (#2963) hwiesmann 2026-01-01 01:02:50 +01:00
  • d9b950eb2f refactor: use time.perf_counter for consistent and accurate benchmarking (#2943) Satyam singh 2025-12-28 19:46:13 +05:30
  • 26dfe4f651 Fetch nanobind with cmake (#2949) Cheng 2025-12-24 10:23:45 +09:00
  • 1d21d0e696 [CUDA] Implement gather_mm_rhs (#2902) Cheng 2025-12-24 09:42:56 +09:00
  • 1eef1d155c Metal/CPU nvfp4 and mxfp8 (#2946) Awni Hannun 2025-12-22 20:45:19 -08:00
  • 9cfda1a86e Fixes in mlx.distributed_config (#2947) Angelos Katharopoulos 2025-12-22 17:38:52 -08:00
  • af2fca5b74 Fix float64 size in data_types.rst (#2948) Patrick Devine 2025-12-22 16:24:07 -08:00
  • 5205de563e ci: add macOS 26 target (#2937) Mike Drob 2025-12-22 14:01:58 -06:00
  • b01fc7eac7 Fix stubgen (#2942) Cheng 2025-12-22 09:42:20 +09:00
  • c0fea26ed2 Fix for non row-contig scales (#2941) Awni Hannun 2025-12-21 06:12:41 -08:00
  • e6de81c963 refactor: use perf_counter for accurate benchmarking (#2940) Satyam singh 2025-12-21 19:37:00 +05:30
  • 7652f1c152 Make CUDA CI run faster (#2939) Cheng 2025-12-21 07:38:48 +09:00
  • d9f4d8d508 Fix pid in local launch (#2936) Angelos Katharopoulos 2025-12-19 13:09:15 -08:00
  • fc19a08caa Set install rpath of python bindings with cmake (#2934) Cheng 2025-12-19 16:43:00 +09:00
  • 49f774904b Fix nightly build (#2933) Cheng 2025-12-19 16:42:53 +09:00
  • b2e2b19bf7 Set rpath with cmake for CUDA build (#2932) Cheng 2025-12-19 12:53:38 +09:00
  • ab4dce4e18 Allow dry run for PyPI release workflow (#2928) Cheng 2025-12-19 09:07:50 +09:00
  • c96bd7d239 Move allocate_workspace to cuda/utils.h (#2923) Cheng 2025-12-19 09:07:22 +09:00
  • 4b88f859b6 Fix CUDA pypi release (#2929) Awni Hannun 2025-12-18 13:43:43 -08:00
  • 32cd28a10e patch bump (#2927) Awni Hannun 2025-12-18 12:15:59 -08:00
  • ff26b00cb1 new[CI]: add linux sanitizer tests (#2860) Melissa Kilby 2025-12-18 12:15:26 -08:00
  • 7ddeb70057 fix cuda release part 2 (#2926) Awni Hannun 2025-12-17 22:14:21 -08:00
  • 1fc313db9d Metal logging (#2904) CCYeh 2025-12-18 05:48:07 +01:00
  • f06a45f967 Fix cuda release (#2925) Awni Hannun 2025-12-17 20:20:12 -08:00
  • 116fda628e Faster copy for col contig to row contig (#2917) Awni Hannun 2025-12-17 19:21:05 -08:00
  • ca731f48b8 Bump the patch version (#2922) Angelos Katharopoulos 2025-12-17 18:06:40 -08:00
  • c215b6f88c Fix warnings for the NAX build (#2921) v0.30.1 Angelos Katharopoulos 2025-12-17 15:58:59 -08:00
  • 3cc9f506bd Add JIT support for NAX kernels (#2916) Jagrit Digani 2025-12-17 13:40:40 -08:00
  • 9194ec20a8 Thunderbolt RDMA communications backend (#2808) Angelos Katharopoulos 2025-12-17 11:27:54 -08:00
  • 4cf5b29fc5 qqmm (#2789) Anastasiia Filippova 2025-12-16 18:28:28 +01:00
  • 6b330eb2d5 DOC : Add compile state example (#2910) Satyam singh 2025-12-16 20:02:58 +05:30
  • f9004103ca Use CUDA runtime headers from local python package (#2906) Cheng 2025-12-16 08:36:32 +09:00
  • c2764d1073 Bump actions/download-artifact from 6 to 7 (#2912) dependabot[bot] 2025-12-15 06:10:16 -08:00
  • 093a62d2ed Bump actions/upload-artifact from 5 to 6 (#2911) dependabot[bot] 2025-12-15 06:09:55 -08:00
  • 1b591ec736 No VJP for mask or sinks in attention (#2909) Awni Hannun 2025-12-13 19:48:39 -08:00
  • 47d2505ea9 Fix attention for large sizes (#2903) Awni Hannun 2025-12-13 06:54:30 -08:00
  • bedefed784 Fix ccache getting disabled (#2905) Cheng 2025-12-13 13:00:51 +09:00
  • ccaaa7d6df fix: possible heap-buffer-overflow in RandomBits::eval_cpu (#2877) Melissa Kilby 2025-12-12 02:11:18 -08:00
  • f3e5ca5414 [CUDA] Add host nodes to subgraph types for graph update (#2901) Awni Hannun 2025-12-11 19:13:44 -08:00
  • 81dfe5f137 Fix grad in place updates (#2899) Awni Hannun 2025-12-11 14:44:58 -08:00
  • 012fb220a1 fp quantize (#2892) Anastasiia Filippova 2025-12-11 15:11:25 +01:00
  • e1fee0074b Update nanobind pin to most recent version (#2896) Nathan Goldbaum 2025-12-11 07:07:36 -07:00
  • 3c8ce9b00e Fix input buffer donation in compile (#2897) CCYeh 2025-12-11 15:07:03 +01:00
  • 937ce79660 do not use simd neon intrinsics on x86 (#2893) David Koski 2025-12-10 12:23:28 -08:00
  • 208f5441a7 bump minimum required Python version (#2891) Nathan Goldbaum 2025-12-09 17:54:38 -07:00
  • b862d842e1 Allow events in sub graph to be updatable (#2886) Awni Hannun 2025-12-09 12:34:37 -08:00
  • f7a400951a Fix docs: replace mx.random.randn with mx.random.normal (#2890) Satyam singh 2025-12-10 01:16:30 +05:30
  • 27232db1ba [CUDA] Enable more graphs to be updatable (#2883) Awni Hannun 2025-12-08 06:18:01 -08:00
  • a4b3bc969b Try not to fail when there should be memory available (#2869) Awni Hannun 2025-12-07 06:11:00 -08:00
  • 667c0f3bb9 [Metal] No copy array init (#2875) Awni Hannun 2025-12-05 13:36:45 -08:00
  • 6245824d42 Make allocator::malloc throw on allocation failure (#2874) Cheng 2025-12-05 17:44:38 +09:00