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

Commit Graph

  • f226eeec9e Fix nn.GRU skipping bhn bias when hidden is None (#3252) mm65x 2026-03-16 20:28:14 +00:00
  • 505fc9850d Fix comparison op JVP returning bool tangents instead of input dtype (#3253) mm65x 2026-03-16 17:57:28 +00:00
  • ea91bd02cf update requirements for Macbook Neo (#3257) Thomas Schranz 2026-03-16 12:33:09 +01:00
  • 1d44d913e6 docs: fix PyTorch to MLX conversion example (#3265) Lik Xun Yuan (Lx) 2026-03-16 19:20:12 +08:00
  • 0bdbfdb838 [CUDA] Implement MaskedScatter (#3151) Long Yixing 2026-03-15 09:33:55 +08:00
  • 5d1700493a [CUDA] Add FFT support (#3243) Lucas Newman 2026-03-14 05:02:19 -07:00
  • b0564a9112 Fix crashes in multi-threaded process teardown (#3167) Valentin Roussellet 2026-03-12 21:45:06 -07:00
  • 7adfc83c7d win: re-enable and fix cuDNN performance (#3242) Daniel Hiltgen 2026-03-12 17:41:59 -07:00
  • 0358c602c7 Bump (#3244) Angelos Katharopoulos 2026-03-11 23:57:22 -07:00
  • ce45c52505 [CUDA] Use qmv kernel for fp quantizations (#3239) v0.31.1 Cheng 2026-03-12 07:25:17 +09:00
  • 0879a6acba Add initial tuning for M5 pro and max (#3211) Jagrit Digani 2026-03-11 14:05:43 -07:00
  • a9573f92f6 [CUDA] Implement SegmentedMM (#3238) Long Yixing 2026-03-12 04:31:43 +08:00
  • 1c2d7041ab Remove quantized_utils.cuh (#3237) Cheng 2026-03-11 19:45:51 +09:00
  • fd6d304b3a win: fix cuda build (#3204) Daniel Hiltgen 2026-03-10 20:58:04 -07:00
  • e1e1399e1b Hybrid sharding (#3194) Anastasiia Filippova 2026-03-10 11:47:25 +01:00
  • 9d03a1b0d9 [CUDA] Support 3/5/6-bit quants in QMV (#3236) Cheng 2026-03-10 19:09:48 +09:00
  • 8d022bcb86 Remove custom fp4/fp8 classes (#3212) Cheng 2026-03-10 16:08:01 +09:00
  • d2702a4fc1 Fix non-strict module update with extra weights (#3214) Michelle DiMarco 2026-03-09 22:03:50 -07:00
  • 6ac5280db4 Fix assigning bool to float16/bfloat16 (#3229) Dan Anderson 2026-03-10 00:50:05 -04:00
  • 572e0a4ac3 Validate dims in rope (#3230) Dan Anderson 2026-03-10 00:48:45 -04:00
  • 9bbd375eec Fix return value in einsum_path for simple contractions (#3232) Dan Anderson 2026-03-10 00:48:26 -04:00
  • a25399cbd4 Validate num_splits in split (#3234) Dan Anderson 2026-03-10 00:47:08 -04:00
  • 5a347b2ec8 [CUDA] Faster compilation and batch support in QMV (#3213) Cheng 2026-03-10 13:45:10 +09:00
  • db487f3649 PR #3220 LayerNorm VJP returns zeros_like(weight) instead of zeros_like(bias placeholder) (#3231) Dan Anderson 2026-03-09 20:06:12 -04:00
  • 8f5ff2ea41 PR#3226 Fix (#3227) Dan Anderson 2026-03-09 18:06:33 -04:00
  • d06c3c8936 Improve mlx.distributed_config (#3199) Angelos Katharopoulos 2026-03-09 13:17:51 -07:00
  • be872ebdef [CUDA] implement Hadamard transform (#3179) Long Yixing 2026-03-05 16:34:19 +08:00
  • 3b3590bf5f [CUDA] Use fp16 accumulation for 4-bit quant in GEMV (#3197) Cheng 2026-03-05 07:58:23 +09:00
  • 3c565437a5 [CUDA] Quantized GEMV (#3180) Cheng 2026-03-04 08:59:31 +09:00
  • 9eef9f1774 Bump actions/upload-artifact from 6 to 7 (#3188) dependabot[bot] 2026-03-04 08:26:24 +09:00
  • e320a2adc2 Bump actions/download-artifact from 7 to 8 (#3189) dependabot[bot] 2026-03-04 08:24:12 +09:00
  • 8cd377b7db Add the bartlett function (#3155) willem adnet 2026-03-03 20:40:54 +01:00
  • f145ece976 Fix/missing libs in docs (#3190) Christophe Prat 2026-03-03 18:53:18 +01:00
  • 1ce7118303 Fix ref leak in mx.save/load with file like object (#3187) AN Long 2026-03-03 09:55:29 +09:00
  • 72e04f7fb7 [CUDA] Fsdp (easy) (#3130) Anastasiia Filippova 2026-03-01 23:29:09 +01:00
  • 6482d13dd3 Skip Hopper-only kernels in CI (#3184) Cheng 2026-02-28 16:22:08 +09:00
  • 1e3736b19d Bump the patch version (#3185) Angelos Katharopoulos 2026-02-27 23:21:37 -08:00
  • 365d6f29b4 Bump the minor version (#3183) v0.31.0 Angelos Katharopoulos 2026-02-27 14:02:03 -08:00
  • d7a553c536 Enable passing in a GPU architecture string via env var (#3176) Angelos Katharopoulos 2026-02-27 11:37:53 -08:00
  • c8536f5248 Fix compile_fuse broadcast split aliasing bug (#3166) Robert Johansson 2026-02-27 03:17:28 +01:00
  • 0c8107ce8a [CUDA] Heuristics for Hopper QMM (#3173) Cheng 2026-02-27 09:22:10 +09:00
  • 5c4abd2f06 JACCL refactor and small update (#3174) Angelos Katharopoulos 2026-02-26 13:56:19 -08:00
  • 4e00919e5c [CUDA][NCCL] group split (#3172) Anastasiia Filippova 2026-02-26 09:26:20 +01:00
  • 6ec0192270 Enable setting thread block cluster for Hopper and later (#3168) Cheng 2026-02-26 16:15:16 +09:00
  • a8ba5ac3e0 Implement mlx.core.blackman (#3136) willem adnet 2026-02-25 22:42:40 +01:00
  • 6304c285d3 [CUDA] FPxINT quantized matmul for Hopper (#3160) Cheng 2026-02-25 09:10:18 +09:00
  • cb198268d5 [Metal] Fix event leak (#3159) Awni Hannun 2026-02-23 19:50:48 -08:00
  • 1d8d693d08 [Metal] Add implicit matmul pathway for mx.conv3d (#3147) Gleb Sterkin 2026-02-24 03:52:50 +02:00
  • d4c81062ad [Metal] Fix 32-bit integer overflow in conv3d unfold kernel (#3143) Kellen Sun 2026-02-19 13:01:22 -05:00
  • f2f2d16451 Export: preserve Dtype state values in export callback arguments (#3145) Alex Skryl 2026-02-19 10:07:28 -06:00
  • daf18e76ca Fix fence synchronization accross command buffers (#3144) Awni Hannun 2026-02-18 20:21:26 -08:00
  • 06305022ab Tensor scale nvfp4 (#3022) Anastasiia Filippova 2026-02-18 11:19:26 +01:00
  • 360639c2df Add the hamming window function (#3135) willem adnet 2026-02-17 09:56:05 +01:00
  • 3bbe87e6dc Add hanning window function (#3124) willem adnet 2026-02-16 18:44:49 +01:00
  • e226af720e Propagate quantization mode in quantized layers (#3133) vskiwi 2026-02-16 05:33:13 +03:00
  • 43f4a74826 Manage stream placement in import function (#3127) Awni Hannun 2026-02-15 06:17:06 -08:00
  • c184262d29 Fix donation in sdpa vector (#3121) Angelos Katharopoulos 2026-02-12 10:46:21 -08:00
  • 72e94c81e1 [CUDA] Attention sinks in cuDNN SDPA (#3118) Cheng 2026-02-11 16:46:39 +09:00
  • 4c86c1e55a Fix precision in Metal fused attention (#3119) Awni Hannun 2026-02-10 14:18:29 -08:00
  • be52cf660b register pressure (#3116) Anastasiia Filippova 2026-02-10 23:17:28 +01:00
  • 54bb3eea42 [CUDA] Use cuDNN SDPA for decoding when using fixed-size KV cache (#3113) Cheng 2026-02-10 09:15:45 +09:00
  • 5e018de4e5 Quantize module to QQLinear (#3106) Anastasiia Filippova 2026-02-09 23:35:17 +01:00
  • 9cd4b9be91 [CUDA] Set current device before allocating memory (#3110) Cheng 2026-02-08 19:04:57 +09:00
  • 566bc16b7c Cleanup test_fast_sdpa.py (#3112) Cheng 2026-02-08 19:04:24 +09:00
  • 8fe1d09207 Fix residency set with user provided buffer (#3108) Awni Hannun 2026-02-06 16:38:36 -08:00
  • ef3fbc60a3 is_available() should check the device index too (#3107) Ronan Collobert 2026-02-06 13:02:04 -08:00
  • 69fd3fa9b1 Patch bump (#3102) Angelos Katharopoulos 2026-02-06 09:15:22 -08:00
  • 185b06d9ef Patch for multi device CUDA (#3100) v0.30.6 Awni Hannun 2026-02-05 17:33:51 -08:00
  • 90e38f7b93 Fix qmv_impl for small N (#3096) Manuel Candales 2026-02-05 20:33:36 -05:00
  • ceea571490 JACCL update (#3094) Angelos Katharopoulos 2026-02-05 15:16:07 -08:00
  • 99ca62c4d3 Fix 2pass sdpa on < M2 (#3099) Awni Hannun 2026-02-05 08:51:29 -08:00
  • 206cf07e5b Fix non simd f16 build (#3097) Awni Hannun 2026-02-05 07:04:02 -08:00
  • f47729c0d8 Disable managed memory on WSL when concurrentManagedAccess is not supported (#3095) Jesse Gross 2026-02-04 17:58:49 -08:00
  • b9b672250e patch (#3093) Awni Hannun 2026-02-03 07:24:30 -08:00
  • adcbb91a9e Fix for NAX overflow. (#3092) v0.30.5 Awni Hannun 2026-02-02 18:54:01 -08:00
  • b56782be52 [Metal] Tune splitk gemm dispatch conditions and partition sizes (#3087) Awni Hannun 2026-02-02 08:45:09 -08:00
  • 8ef539522c Fix failing python tests on Windows (#3076) Cheng 2026-01-30 17:50:18 +09:00
  • 212077f163 Fallback to pinned host memory when managed memory is not supported (#3075) Cheng 2026-01-30 13:18:41 +09:00
  • cc6e4eebad Fix nax condition for iphone (#3083) Awni Hannun 2026-01-29 13:30:37 -08:00
  • fcbdd05022 More useful error for large indices (#3079) Awni Hannun 2026-01-29 13:02:39 -08:00
  • 590b4f1c16 Fix ALiBi slopes for non-power-of-2 num_heads (#3071) atharva 2026-01-29 20:53:11 +05:30
  • 0c6a895ed7 Use lower-right causal mask alignment consistently (#2967) Anri Lombard 2026-01-29 03:15:14 +02:00
  • c86a9bced1 [Docs] Simple example of using MLX distributed (#2973) stef 2026-01-28 20:14:56 -05:00
  • 12e386f308 Tune CUDA gaph sizes on B200 and H100 (#3077) Awni Hannun 2026-01-28 17:14:44 -08:00
  • 2ac18eddb9 [CUDA] Fallback Event impl when there is no hardware cpu/gpu coherency (#3070) Cheng 2026-01-28 10:43:22 +09:00
  • b537b3685f patch (#3074) Awni Hannun 2026-01-27 17:42:55 -08:00
  • 2f324cc3b2 remove thrust (#3067) v0.30.4 Awni Hannun 2026-01-27 08:54:07 -08:00
  • 4912cc47c2 Fp qmv (#2984) Awni Hannun 2026-01-27 06:33:06 -08:00
  • ce4d0a62ef Do not require ConcurrentManagedAccess when not used (#3062) Cheng 2026-01-27 11:19:20 +09:00
  • 73136472e0 Delay load CUDA libs and resolve DLL paths at runtime (#3061) Cheng 2026-01-27 11:01:58 +09:00
  • fed0fe3c73 Better support consumer CUDA GPUs (#3056) Jesse Gross 2026-01-26 16:45:02 -08:00
  • 343ddf0d73 Fix long cache file path on Windows (#3065) Cheng 2026-01-27 08:53:26 +09:00
  • b70fc33ada Improve CPU discovery (#3068) Daniel Hiltgen 2026-01-26 15:01:43 -08:00
  • 7ed2b6b935 Add NAX Split-K GEMM for large-K matmuls to improve performance (#3018) Merlin78 2026-01-26 11:23:20 -08:00
  • a828e769be GPU discovery (#3055) Daniel Hiltgen 2026-01-26 09:54:13 -08:00
  • b6aa03e5b8 Update pre-commit hooks and versions for clang-format, black, and isort (#3059) Nripesh Niketan 2026-01-26 14:57:04 +00:00
  • 5bd99dd5ec Fix flaky macOS test (#3063) Awni Hannun 2026-01-25 16:40:57 -08:00
  • 9e2d2a5957 [CUDA] Fast sorting (#3060) Awni Hannun 2026-01-25 15:10:18 -08:00
  • 3ac892b008 Hide symbols by default for mac/linux (#3057) Cheng 2026-01-25 14:30:41 +09:00
  • 0bb50d99c0 Fix some NVCC warnings when building CUDA backend with MSVC (#3038) Cheng 2026-01-25 12:25:01 +09:00