Commit Graph

  • 0de5988f92 Custom VJP and checkpointing (#541) Angelos Katharopoulos 2024-01-30 16:04:45 -08:00
  • 143e2690d5 Fix SGD implementation (#473) Jacket 2024-01-30 17:50:46 -06:00
  • 375446453e Update Compute Pipeline Creation API (#581) Jagrit Digani 2024-01-30 15:42:36 -08:00
  • 1895d34c20 Fix log1p with inf inputs (#592) Angelos Katharopoulos 2024-01-30 14:02:50 -08:00
  • 09b9275027 Make shape a tuple (#591) Awni Hannun 2024-01-30 13:11:01 -08:00
  • d3a9005454 Softshrink mapping + op (#552) Andre Slavescu 2024-01-30 15:56:28 -05:00
  • 3f7aba8498 Implement diagonal operator (#562) Jacket 2024-01-30 11:45:48 -06:00
  • 65d0b8df9f Fix binary op dispatch (#584) Angelos Katharopoulos 2024-01-29 19:36:17 -08:00
  • 3c2f192345 Propagate nans in binary ops (#579) Awni Hannun 2024-01-29 11:19:38 -08:00
  • 37d98ba6ff No gil eval (#565) Angelos Katharopoulos 2024-01-26 22:03:52 -08:00
  • 8993382aaa Buffer Donation (#519) Awni Hannun 2024-01-26 16:30:33 -08:00
  • 07f35c9d8a Fix a few issues: docs for flatten, erf, dequantize validation (#560) Awni Hannun 2024-01-26 15:16:46 -08:00
  • bf17ab5002 Add more checks and clearer error messages to conv operations (#563) Jagrit Digani 2024-01-26 15:13:26 -08:00
  • 8fa6b322b9 Compile front-end (#476) Awni Hannun 2024-01-26 13:45:30 -08:00
  • 874b739f3c Fix cache key in RoPE (#561) David Koski 2024-01-26 13:10:02 -08:00
  • 077c1ee64a QR factorization (#310) taher 2024-01-26 09:27:31 -08:00
  • 2463496471 [Fix] mx.allclose bug with infinite values (#539) Rifur13 2024-01-25 23:47:06 -05:00
  • 87b7fa9ba2 Bump the version (#554) v0.0.11 Angelos Katharopoulos 2024-01-25 11:01:05 -08:00
  • 624065c074 Fix package installation for CI (#521) Danilo Peixoto 2024-01-25 14:43:34 -03:00
  • f27ec5e097 More helpful error message in vjp transform + concate bug (#543) Awni Hannun 2024-01-24 09:58:33 -08:00
  • f30e63353a Minor updates to address a few issues (#537) Awni Hannun 2024-01-23 22:24:41 -08:00
  • 4fe2fa2a64 GGUF: Avoid dequantization when format is compatible (#426) Juarez Bochi 2024-01-23 18:43:57 -05:00
  • 37fc9db82c Added Adafactor (#415) Hazem Essam 2024-01-24 01:11:27 +02:00
  • 755dcf6137 Enable cross_entropy loss to handle dense targets (#517) AtomicVar 2024-01-24 04:17:22 +08:00
  • 6b4b30e3fc Common neural network initializers nn.initializers (#456) LeonEricsson 2024-01-23 15:47:20 +01:00
  • 86e0c79467 remove stale benchmarks (#527) Awni Hannun 2024-01-22 22:17:58 -08:00
  • 98c37d3a22 use axes in tensordot (#525) Awni Hannun 2024-01-22 21:17:00 -08:00
  • f326dd8334 Update README.md (#524) Sugato Ray 2024-01-22 23:53:54 -05:00
  • 6d3bee3364 Fix oob reads in gemv kernel (#523) Jagrit Digani 2024-01-22 12:06:04 -08:00
  • ecb174ca9d Type annotations for mlx.core module (#512) Danilo Peixoto 2024-01-21 17:53:12 -03:00
  • 7a34e46677 Quantize with groups of 32 (#511) Awni Hannun 2024-01-21 06:19:05 -08:00
  • 92c22c1ea3 feat: Update isort version to 5.13.2 (#514) Nripesh Niketan 2024-01-21 18:11:48 +04:00
  • d52383367a format (#510) Awni Hannun 2024-01-20 10:33:46 -08:00
  • 363d3add6d Add ValuError message for Adamax (#508) Arda Orçun 2024-01-20 18:56:15 +03:00
  • b207c2c86b Power VJP fix for 0 (#505) Awni Hannun 2024-01-20 01:17:40 -08:00
  • 6bf779e72b fix array from list for > 32 bit types (#501) Awni Hannun 2024-01-19 15:49:25 -08:00
  • ddf50113c5 GGUF: Load and save metadata (#446) Juarez Bochi 2024-01-19 23:06:05 +01:00
  • 6589c869d6 Added MSE message (#500) Arda Orçun 2024-01-19 17:27:50 +03:00
  • f6feb61f92 feat: add support for saving safetensors in the save_weights (#497) Anchen 2024-01-19 06:19:33 -08:00
  • c4ec836523 fix isinf for integer types (#494) Awni Hannun 2024-01-19 05:31:10 -08:00
  • 550d4bf7c0 Update binary_cross_entropy function to handle both logits and probabilities (#492) AtomicVar 2024-01-19 11:22:23 +08:00
  • f6e911ced0 version bump (#490) v0.0.10 Awni Hannun 2024-01-18 12:00:24 -08:00
  • 3d99a8d31d Fix format / build (#489) Awni Hannun 2024-01-18 10:01:59 -08:00
  • a749a91c75 Support disable metal buffer cache to prevent performance degradation caused by large memory caching (#390) Ethan 2024-01-19 00:33:34 +08:00
  • 49a52610b7 Added formatter structure and a boolean value formatter (#354) toji 2024-01-18 21:19:41 +05:30
  • d1fef34138 Add Gaussian NLL loss function (#477) AtomicVar 2024-01-18 22:44:44 +08:00
  • 9c111f176d Fix split optimization for array iterator (#484) Angelos Katharopoulos 2024-01-18 05:50:25 -08:00
  • 78e5f2d17d usage doc for function transformations (#481) Awni Hannun 2024-01-17 17:10:53 -08:00
  • 90c234b7ac Fix round to round half-cases to even (#482) Angelos Katharopoulos 2024-01-17 15:27:23 -08:00
  • 135fd796d2 Fix detach for multi-output primitives (#480) Angelos Katharopoulos 2024-01-17 14:08:07 -08:00
  • 78102a47ad Update GEMM (#424) Jagrit Digani 2024-01-17 12:42:39 -08:00
  • 556cdf0e06 Resolves build issues with the extension example (#419) Diogo 2024-01-17 15:07:05 -05:00
  • 275db7221a Command buffer reports errors (#479) Awni Hannun 2024-01-17 11:53:30 -08:00
  • 4a9012cba0 Sort some APIs docs by names (a-z) (#472) AtomicVar 2024-01-17 11:37:50 +08:00
  • a2bf7693dd Primitive's VJP takes outputs as input (#475) Awni Hannun 2024-01-16 19:03:53 -08:00
  • d8fabaa12b Split multi output (#461) Angelos Katharopoulos 2024-01-16 13:33:55 -08:00
  • 4e290d282f feat: add time based seed to random.h (#457) Avikant Srivastava 2024-01-16 21:02:28 +05:30
  • e72458a3fa implemented isposinf and isneginf in one PR (#470) Yashraj Singh 2024-01-16 20:18:07 +05:30
  • a2ffea683a Fix eye for larger matrices (#463) Awni Hannun 2024-01-16 00:51:24 -08:00
  • c15fe3e61b Allow arbitrary first dimension in quantization kernels. (#458) Angelos Katharopoulos 2024-01-16 00:46:21 -08:00
  • f44c132f4a Add scatter_min VJP (#462) Tristan Bilot 2024-01-16 09:37:40 +01:00
  • 92a2fdd577 Adds isinf (#445) Matthew Ernst 2024-01-15 19:50:44 -08:00
  • 6022d4129e scatter_max vjp + bindings + tests (#431) Tristan Bilot 2024-01-14 23:12:15 +01:00
  • 4bc446be08 Use a dummy primitive to only sync with one output (#453) Awni Hannun 2024-01-14 14:09:40 -08:00
  • 41cc7bdfdb Fix stub generation, change graph exporting for arrows to go to outputs (#455) Awni Hannun 2024-01-14 14:06:16 -08:00
  • 6e81c3e164 Sync only with outputs we need to sync with (#447) Awni Hannun 2024-01-13 01:47:25 -08:00
  • 2e29d0815b Add tile op (#438) Diogo 2024-01-13 02:03:16 -05:00
  • 1b71487e1f docs (#444) Awni Hannun 2024-01-12 13:34:16 -08:00
  • 1416e7b664 Add isnan (#423) Ayush Shridhar 2024-01-12 14:16:48 -05:00
  • 29081204d1 array.swapaxes should point to swapaxes free function (#441) davidkoski 2024-01-12 11:06:16 -08:00
  • 006d01ba42 Fix packaging of gguflib (#435) v0.0.9 Angelos Katharopoulos 2024-01-11 13:56:03 -08:00
  • 46dc24d835 version bump (#433) Awni Hannun 2024-01-11 12:29:35 -08:00
  • c9934fe8a4 Metal validation (#432) Awni Hannun 2024-01-11 11:57:24 -08:00
  • 975e265f74 feat: Add numpy constants (#428) Avikant Srivastava 2024-01-11 20:17:29 +05:30
  • c92a134b0d more docs (#421) Awni Hannun 2024-01-10 14:04:12 -08:00
  • 3b4f066dac Correct types for vjp + tests (#418) Awni Hannun 2024-01-10 13:32:37 -08:00
  • b7f905787e GGUF support (#350) Juarez Bochi 2024-01-10 16:22:48 -05:00
  • e3e933c6bc Add type hint for Module (#412) Chunyang Wen 2024-01-11 03:23:42 +08:00
  • 1d90a76d63 in place ops behave in place, fix some overloads (#411) Awni Hannun 2024-01-09 16:05:38 -08:00
  • 961435a243 Scatter vjp (#394) Angelos Katharopoulos 2024-01-09 13:36:51 -08:00
  • e9ca65c939 Fix BN stats to not expand shape (#409) Awni Hannun 2024-01-09 11:54:51 -08:00
  • 753867123d Fix data_types.rst uint64 (#406) Dwayne Robinson 2024-01-09 06:40:10 -08:00
  • f099ebe535 Multi output primitives (#330) Awni Hannun 2024-01-08 16:39:08 -08:00
  • f45f70f133 Update mlx-example link for llms llama in llama-inference.rst (#405) BigsnarfDude 2024-01-08 17:29:53 -07:00
  • 0b8aeddac6 Additoinal losses (#336) YUN, Junwoo 2024-01-09 07:01:13 +09:00
  • 432ee5650b Update cpp tests with allclose and doctest::Approx for numerical tolerance (#401) Jagrit Digani 2024-01-08 09:35:05 -08:00
  • 73321b8097 feat: add logicalAnd and logicalOR (#386) Nripesh Niketan 2024-01-08 19:00:05 +04:00
  • 022a944367 Added GLU activation function and Gated activation function (#329) Hazem Essam 2024-01-08 16:13:16 +02:00
  • 026ef9aae4 Update Install Instructions (#397) Chris Costes 2024-01-07 22:11:04 -05:00
  • a611b0bc82 Removes the retain_graph flag (#385) Angelos Katharopoulos 2024-01-07 15:16:51 -08:00
  • 449b43762e Add inner / outer op (#348) Diogo 2024-01-07 12:01:09 -05:00
  • 6ea6b4258d Fix style check (#395) Angelos Katharopoulos 2024-01-07 05:54:58 -08:00
  • 48f6ca8c3a Add theta cache for Rope and mask cache for ALiBi (#375) Anchen 2024-01-07 00:22:58 -08:00
  • c6d2878c1a safely divide for 0 size inputs (#388) Awni Hannun 2024-01-07 00:19:54 -08:00
  • b34bf5d52b fix saving for non-contiguous arrays (#389) Awni Hannun 2024-01-06 12:44:02 -08:00
  • 608bd43604 Move the matmul type check in the op (#384) Angelos Katharopoulos 2024-01-05 19:10:13 -08:00
  • 4c48f6460d Fix segfault from buffer protocol and tests (#383) Angelos Katharopoulos 2024-01-05 18:17:44 -08:00
  • 1331fa19f6 Make array conform to the Python Buffer Protocol (#323) Daniel Strobusch 2024-01-06 00:58:33 +01:00
  • dfdb284e16 make behaviour of dtype arguments consistent and compliant to numpy (#379) Daniel Strobusch 2024-01-05 18:37:46 +01:00
  • d8f41a5c0f support python mlx.array creation from list of mlx.array's (#325) mutexuan 2024-01-05 10:53:33 +08:00