Commit Graph

  • 1865299a30 Metal shaders for memory efficient self attention on large sequences (#964) Brian Keene 2024-06-03 12:16:19 -04:00
  • 3576b547c5 Doc error for default for scale in SinusoidalPositionalEncoding (#1174) Dominik Schlösser 2024-06-02 22:42:45 +02:00
  • 079882495d version bump (#1172) v0.14.1 Awni Hannun 2024-05-31 12:29:12 -07:00
  • ab977109db feat: Added dlpack device (#1165) K Venkat Ramnan 2024-05-31 12:29:01 -07:00
  • fd1c08137b stable cumprod grad at 0 (#1167) Awni Hannun 2024-05-31 12:28:42 -07:00
  • 76b6cece46 Fix multi-block sort stride management (#1169) Jagrit Digani 2024-05-31 11:10:54 -07:00
  • 9f0df51f8d Fix matvec vector stride bug (#1168) Jagrit Digani 2024-05-29 12:18:28 -07:00
  • e7a2a3dcd1 Fix a couple bugs (#1161) Awni Hannun 2024-05-28 15:18:18 -07:00
  • a87ef5bfc1 fix broadcast bug in bitwise ops (#1157) Awni Hannun 2024-05-24 11:44:40 -07:00
  • 9f9cb7a2ef version bump (#1154) v0.14.0 Awni Hannun 2024-05-23 18:08:08 -07:00
  • 7e26fd8032 Option to JIT steel gemm / conv (#1139) Awni Hannun 2024-05-23 18:07:34 -07:00
  • eab2685c67 Float mask update (#1152) Jagrit Digani 2024-05-23 17:20:44 -07:00
  • 50dfb664db Comms (#1097) Angelos Katharopoulos 2024-05-23 17:04:02 -07:00
  • 0189ab6ab6 More jitting (#1132) Awni Hannun 2024-05-23 16:23:44 -07:00
  • 9401507336 Add groups to 2-D convolutions (#1129) Rifur13 2024-05-22 23:01:44 -04:00
  • eb8321d863 list based indexing (#1150) Awni Hannun 2024-05-22 15:52:05 -07:00
  • 79ef49b2c2 add mx.trace (#1143) (#1147) Abe Leininger 2024-05-22 18:50:27 -04:00
  • e110ca11e2 Fix offset bug for device buffers (#1151) Awni Hannun 2024-05-22 15:50:05 -07:00
  • 226748b3e7 JIT compile option for binary minimization (#1091) Awni Hannun 2024-05-22 12:57:13 -07:00
  • d568c7ee36 Rename block sparse (#1149) Awni Hannun 2024-05-22 07:48:34 -07:00
  • e6fecbb3e1 Some fixes in docs (#1141) Awni Hannun 2024-05-20 11:51:47 -07:00
  • da83f899bb Improve qvm speed (#1140) Angelos Katharopoulos 2024-05-20 09:20:44 -07:00
  • 7e5674d8be Treate 'minimum' differently in cosine decay (#1138) jlwitthuhn 2024-05-20 08:00:48 -07:00
  • 0a558577bf Update README.md (#1136) Shixian Sheng 2024-05-20 09:16:40 -04:00
  • fb71a82ada Fix copy bug with many dims (#1137) Awni Hannun 2024-05-17 21:10:03 -07:00
  • 23406c9e9e Choose the right MLX bf16 for extensions (#1135) Awni Hannun 2024-05-17 15:09:28 -07:00
  • b3ec792380 Implemented Cholesky on CPU (#1119) Luca Arnaboldi 2024-05-17 21:31:59 +02:00
  • 6a9b584f3d patch bump (#1131) v0.13.1 Awni Hannun 2024-05-16 20:51:33 -07:00
  • 81dd33af66 allow conversion to dlpack (#1120) Awni Hannun 2024-05-16 16:11:37 -07:00
  • 8b76571896 Fix extensions (#1126) Awni Hannun 2024-05-16 15:36:25 -07:00
  • e78a6518fa Block sparse qmm (#1124) Angelos Katharopoulos 2024-05-16 15:24:14 -07:00
  • 1873ffda01 Detect metal version and propagate correctly for JIT (#1109) Awni Hannun 2024-05-15 17:42:09 -07:00
  • c417e42116 [Fix] minor typo in default argument for argpartition's "axis" parameter (#1125) Jacket 2024-05-15 17:25:25 -05:00
  • 358e1fd6ab Fused GEMM (#1123) Jagrit Digani 2024-05-15 10:30:41 -07:00
  • 631dfbe673 fix scatter index bug (#1122) Awni Hannun 2024-05-14 15:04:58 -07:00
  • 56a4eaed72 Pass missing stream arg in array.flatten (#1111) Cheng 2024-05-14 21:50:16 +08:00
  • bf925d9dc7 Move args in conv_general (#1118) Cheng 2024-05-14 21:50:09 +08:00
  • 1a7ed5dcb6 Fill vector with constructor instead of fill_n (#1113) Cheng 2024-05-14 21:28:55 +08:00
  • 5be5daa6ef Use compiled function in Sigmoid module (#1116) Cheng 2024-05-14 21:25:57 +08:00
  • 60cb11764e Use correct module type in quantized.py (#1115) Cheng 2024-05-14 21:25:42 +08:00
  • cbd5445ea7 The tile op does not accept None as reps (#1117) Cheng 2024-05-14 21:25:25 +08:00
  • 2c7e9b5158 Add missing docs for some ops (#1110) Cheng 2024-05-14 21:09:05 +08:00
  • 2263e4b279 Experiment with medium machines for CI (#1000) Mike Drob 2024-05-13 21:40:19 -05:00
  • 863039da4c Allow scatter type exception to be caught by checking in op (#1077) Awni Hannun 2024-05-13 17:43:53 -07:00
  • 7178ac0111 No CPU option for binary minimization (#1105) Awni Hannun 2024-05-13 16:08:11 -07:00
  • e7f9710499 Fix typo in a variable name in example code. (#1104) Ravindra R. Jaju 2024-05-13 18:34:23 +05:30
  • ff4223904d Conv3d (#993) Max-Heinrich Laves 2024-05-11 15:15:02 +02:00
  • a9f80d60f6 improve error messaging in eval (#1101) Awni Hannun 2024-05-10 10:04:07 -07:00
  • 2e158cf6d0 Add conjugate operator (#1100) Alex Barron 2024-05-10 07:22:20 -07:00
  • 8bd6bfa4b5 version (#1099) v0.13.0 Awni Hannun 2024-05-09 17:52:39 -07:00
  • 8b1906abd0 Add compiler flags to disable safetensors and gguf (#1098) Awni Hannun 2024-05-09 17:39:44 -07:00
  • 06375e6605 Split encoders in non-concurrent context with a max ops per encoder (#1085) Awni Hannun 2024-05-09 16:21:02 -07:00
  • b21242faf1 Allow unary ops to accept array like (#1093) Awni Hannun 2024-05-09 09:36:02 -07:00
  • cc05a281c4 Added ArcTan2 operation (#1079) Rahul Yedida 2024-05-08 11:35:15 -04:00
  • fe96ceee66 Update block offset adjustment to be in size_t (#1087) Jagrit Digani 2024-05-08 08:10:23 -07:00
  • 9814a2ae12 fix conversion to array (#1070) Awni Hannun 2024-05-06 16:02:49 -07:00
  • 6992498e7a add keyword positonal (#1081) Shubham 2024-05-06 10:18:49 -04:00
  • 21623156a3 Reset peak memory (#1074) Awni Hannun 2024-05-03 17:12:51 -07:00
  • 79c859e2e0 feat: implement clip_grad_norm (#1043) Nripesh Niketan 2024-05-03 20:07:02 +04:00
  • b00ac960b4 change initial memory limits and add memory size to device info (#1064) Awni Hannun 2024-05-03 06:50:15 -07:00
  • 02a9fc7bfa Patch bump (#1067) v0.12.2 Awni Hannun 2024-05-02 16:37:31 -07:00
  • f390957685 Block sparse mm (#1058) Jagrit Digani 2024-05-02 14:03:58 -07:00
  • 17f57df797 Improvements in the quantizer and dequantization kernel (#1061) Angelos Katharopoulos 2024-05-01 18:19:11 -07:00
  • 7f7b9662ea Fix leak for multi-output primitives which are never detached (#1059) Awni Hannun 2024-05-01 07:31:45 -07:00
  • 19bef39f5c Add a mx.metal.device_info (#1060) Awni Hannun 2024-04-30 15:47:27 -07:00
  • a30e7ed2da feat: metal formatting and pre-commit bump (#1038) Nripesh Niketan 2024-04-30 18:18:09 +04:00
  • 8db7161c94 Bug fix in quantize (#1054) Angelos Katharopoulos 2024-04-29 20:55:04 -07:00
  • 09f1777896 fix slice update indexing (#1053) Awni Hannun 2024-04-29 12:17:40 -07:00
  • 490c0c4fdc [Fix] expand axes for dimension with integer indices in mlx_slice_update (#1035) v0.12.1 Jacket 2024-04-29 09:57:28 -05:00
  • c4a471c99d Add groups to Conv1d (#948) Rifur13 2024-04-27 09:24:57 -04:00
  • 86f495985b Add bitwise ops (#1037) Awni Hannun 2024-04-26 22:03:42 -07:00
  • 67d1894759 fix order device -> scheduler (#1039) Awni Hannun 2024-04-26 13:46:41 -07:00
  • 5bfe89bdb1 Cpp docs (#1036) Awni Hannun 2024-04-26 12:56:05 -07:00
  • 82463e9938 Bump the version to 0.12 (#1034) v0.12.0 Angelos Katharopoulos 2024-04-25 14:18:08 -07:00
  • 771575d27b Expose function to clear memory cache (#1032) Awni Hannun 2024-04-24 16:48:51 -07:00
  • 20a01bbd9f Simplifying and improving qmm (#1030) Angelos Katharopoulos 2024-04-24 13:07:45 -07:00
  • ec8578d41a Fix quantization of all 0s (#1028) Angelos Katharopoulos 2024-04-24 00:40:42 -07:00
  • d0dbfe0b97 Adds radians and degrees (#1011) Aneesh Shetty 2024-04-22 13:17:49 -05:00
  • 3d405fb3b1 Add synchronize function (#1006) Awni Hannun 2024-04-22 08:25:46 -07:00
  • b0012cdd0f Bump the patch version for the quants (#1018) v0.11.1 Angelos Katharopoulos 2024-04-19 20:28:34 -07:00
  • 84d61d27aa Make sure 0 is represented in the quantization (#1016) Angelos Katharopoulos 2024-04-19 19:47:26 -07:00
  • ed83908931 fix gguf loading quants (#1014) Awni Hannun 2024-04-19 12:24:07 -07:00
  • ef5f7d1aea Fix buffer protocol buffer size designation (#1010) Angelos Katharopoulos 2024-04-19 06:06:13 -07:00
  • 090ff659dc bump (#1007) v0.11.0 Awni Hannun 2024-04-18 13:18:43 -07:00
  • 85c8a91a27 Fix mask broadcasting bug and add relevant test (#1003) Jagrit Digani 2024-04-17 17:33:48 -07:00
  • 581b699ac9 avgpool, not maxpool (#1002) Piotr Rybiec 2024-04-17 17:26:22 +02:00
  • 8a0677d56d Shared events for synchronization + async eval (#998) Awni Hannun 2024-04-17 06:16:02 -07:00
  • b18468bf81 Masked mm (#978) Jagrit Digani 2024-04-16 14:45:39 -07:00
  • 107ba2891a gelu tanh approx (#989) Shiyu 2024-04-16 10:49:00 +08:00
  • cd9e184529 Quantize embedding (#994) Awni Hannun 2024-04-15 16:42:10 -07:00
  • 2e7c02d5cd Metal FFT for powers of 2 up to 2048 (#915) Alex Barron 2024-04-12 05:40:06 +01:00
  • ae18326533 No copy command encoder (#986) Awni Hannun 2024-04-11 21:15:36 -07:00
  • 91eba8e485 fix for grammatical typo in docs (#988) Alex Shepard 2024-04-11 17:02:06 -07:00
  • d07e295c62 bumpity bump (#987) v0.10.0 Awni Hannun 2024-04-11 12:48:52 -07:00
  • dce4bd74a4 Add ArrayDesc destructor to avoid possible stack overflow (#982) Angelos Katharopoulos 2024-04-11 11:37:02 -07:00
  • ffff671273 Update pre-commit hooks (#984) Nripesh Niketan 2024-04-11 18:27:53 +04:00
  • 12d4507ee3 Explicit barriers with concurrent dispatch (#977) Awni Hannun 2024-04-10 21:45:31 -07:00
  • 8580d997ff Try a stack-based DFS for eval (#980) Awni Hannun 2024-04-10 17:05:13 -07:00
  • 061cf9a4ce Upsample with bicubic interpolation (#967) Shiyu 2024-04-11 06:47:22 +08:00
  • 99abb9eff4 Async eval (#972) Awni Hannun 2024-04-09 18:34:00 -07:00