mlx-cpu 0.30.1__py3-none-manylinux_2_35_x86_64.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (231) hide show
  1. mlx/__main__.py +27 -0
  2. mlx/_reprlib_fix.py +16 -0
  3. mlx/extension.py +88 -0
  4. mlx/include/mlx/3rdparty/pocketfft.h +3581 -0
  5. mlx/include/mlx/allocator.h +73 -0
  6. mlx/include/mlx/array.h +645 -0
  7. mlx/include/mlx/backend/common/binary.h +97 -0
  8. mlx/include/mlx/backend/common/broadcasting.h +11 -0
  9. mlx/include/mlx/backend/common/buffer_cache.h +157 -0
  10. mlx/include/mlx/backend/common/compiled.h +77 -0
  11. mlx/include/mlx/backend/common/copy.h +50 -0
  12. mlx/include/mlx/backend/common/hadamard.h +109 -0
  13. mlx/include/mlx/backend/common/matmul.h +67 -0
  14. mlx/include/mlx/backend/common/reduce.h +59 -0
  15. mlx/include/mlx/backend/common/slicing.h +20 -0
  16. mlx/include/mlx/backend/common/ternary.h +85 -0
  17. mlx/include/mlx/backend/common/unary.h +29 -0
  18. mlx/include/mlx/backend/common/utils.h +205 -0
  19. mlx/include/mlx/backend/cpu/arange.h +28 -0
  20. mlx/include/mlx/backend/cpu/available.h +9 -0
  21. mlx/include/mlx/backend/cpu/binary.h +517 -0
  22. mlx/include/mlx/backend/cpu/binary_ops.h +98 -0
  23. mlx/include/mlx/backend/cpu/binary_two.h +166 -0
  24. mlx/include/mlx/backend/cpu/compiled_preamble.h +12 -0
  25. mlx/include/mlx/backend/cpu/copy.h +36 -0
  26. mlx/include/mlx/backend/cpu/encoder.h +67 -0
  27. mlx/include/mlx/backend/cpu/eval.h +12 -0
  28. mlx/include/mlx/backend/cpu/gemm.h +26 -0
  29. mlx/include/mlx/backend/cpu/gemms/simd_gemm.h +139 -0
  30. mlx/include/mlx/backend/cpu/jit_compiler.h +20 -0
  31. mlx/include/mlx/backend/cpu/lapack.h +80 -0
  32. mlx/include/mlx/backend/cpu/simd/accelerate_fp16_simd.h +56 -0
  33. mlx/include/mlx/backend/cpu/simd/accelerate_simd.h +329 -0
  34. mlx/include/mlx/backend/cpu/simd/base_simd.h +295 -0
  35. mlx/include/mlx/backend/cpu/simd/math.h +193 -0
  36. mlx/include/mlx/backend/cpu/simd/neon_fp16_simd.h +212 -0
  37. mlx/include/mlx/backend/cpu/simd/simd.h +4 -0
  38. mlx/include/mlx/backend/cpu/simd/type.h +11 -0
  39. mlx/include/mlx/backend/cpu/slicing.h +21 -0
  40. mlx/include/mlx/backend/cpu/ternary.h +154 -0
  41. mlx/include/mlx/backend/cpu/threefry.h +21 -0
  42. mlx/include/mlx/backend/cpu/unary.h +281 -0
  43. mlx/include/mlx/backend/cpu/unary_ops.h +180 -0
  44. mlx/include/mlx/backend/cuda/allocator.h +89 -0
  45. mlx/include/mlx/backend/cuda/conv/conv.h +126 -0
  46. mlx/include/mlx/backend/cuda/cublas_utils.h +96 -0
  47. mlx/include/mlx/backend/cuda/cuda.h +10 -0
  48. mlx/include/mlx/backend/cuda/cuda_utils.h +89 -0
  49. mlx/include/mlx/backend/cuda/cudnn_utils.h +171 -0
  50. mlx/include/mlx/backend/cuda/device/config.h +12 -0
  51. mlx/include/mlx/backend/cuda/device.h +189 -0
  52. mlx/include/mlx/backend/cuda/event.h +78 -0
  53. mlx/include/mlx/backend/cuda/gemms/cublas_gemm.h +114 -0
  54. mlx/include/mlx/backend/cuda/gemms/gemv.h +24 -0
  55. mlx/include/mlx/backend/cuda/jit_module.h +119 -0
  56. mlx/include/mlx/backend/cuda/lru_cache.h +189 -0
  57. mlx/include/mlx/backend/cuda/quantized/cublas_qqmm.h +88 -0
  58. mlx/include/mlx/backend/cuda/quantized/cuda_fp4.h +83 -0
  59. mlx/include/mlx/backend/cuda/quantized/qqmm_utils.h +30 -0
  60. mlx/include/mlx/backend/cuda/quantized/quantized.h +45 -0
  61. mlx/include/mlx/backend/cuda/utils.h +46 -0
  62. mlx/include/mlx/backend/cuda/worker.h +55 -0
  63. mlx/include/mlx/backend/gpu/available.h +9 -0
  64. mlx/include/mlx/backend/gpu/copy.h +57 -0
  65. mlx/include/mlx/backend/gpu/eval.h +18 -0
  66. mlx/include/mlx/backend/gpu/slicing.h +36 -0
  67. mlx/include/mlx/backend/metal/allocator.h +79 -0
  68. mlx/include/mlx/backend/metal/binary.h +33 -0
  69. mlx/include/mlx/backend/metal/device.h +283 -0
  70. mlx/include/mlx/backend/metal/jit/includes.h +57 -0
  71. mlx/include/mlx/backend/metal/jit/indexing.h +76 -0
  72. mlx/include/mlx/backend/metal/kernels/arange.h +9 -0
  73. mlx/include/mlx/backend/metal/kernels/atomic.h +345 -0
  74. mlx/include/mlx/backend/metal/kernels/bf16.h +16 -0
  75. mlx/include/mlx/backend/metal/kernels/bf16_math.h +380 -0
  76. mlx/include/mlx/backend/metal/kernels/binary.h +199 -0
  77. mlx/include/mlx/backend/metal/kernels/binary_ops.h +326 -0
  78. mlx/include/mlx/backend/metal/kernels/binary_two.h +244 -0
  79. mlx/include/mlx/backend/metal/kernels/cexpf.h +134 -0
  80. mlx/include/mlx/backend/metal/kernels/complex.h +173 -0
  81. mlx/include/mlx/backend/metal/kernels/copy.h +276 -0
  82. mlx/include/mlx/backend/metal/kernels/defines.h +24 -0
  83. mlx/include/mlx/backend/metal/kernels/erf.h +69 -0
  84. mlx/include/mlx/backend/metal/kernels/expm1f.h +90 -0
  85. mlx/include/mlx/backend/metal/kernels/fft/radix.h +328 -0
  86. mlx/include/mlx/backend/metal/kernels/fft/readwrite.h +624 -0
  87. mlx/include/mlx/backend/metal/kernels/fft.h +486 -0
  88. mlx/include/mlx/backend/metal/kernels/fp4.h +59 -0
  89. mlx/include/mlx/backend/metal/kernels/fp8.h +82 -0
  90. mlx/include/mlx/backend/metal/kernels/fp_quantized.h +1804 -0
  91. mlx/include/mlx/backend/metal/kernels/fp_quantized_nax.h +1059 -0
  92. mlx/include/mlx/backend/metal/kernels/gemv_masked.h +827 -0
  93. mlx/include/mlx/backend/metal/kernels/hadamard.h +182 -0
  94. mlx/include/mlx/backend/metal/kernels/indexing/gather.h +51 -0
  95. mlx/include/mlx/backend/metal/kernels/indexing/gather_axis.h +44 -0
  96. mlx/include/mlx/backend/metal/kernels/indexing/gather_front.h +24 -0
  97. mlx/include/mlx/backend/metal/kernels/indexing/indexing.h +23 -0
  98. mlx/include/mlx/backend/metal/kernels/indexing/masked_scatter.h +38 -0
  99. mlx/include/mlx/backend/metal/kernels/indexing/scatter.h +59 -0
  100. mlx/include/mlx/backend/metal/kernels/indexing/scatter_axis.h +52 -0
  101. mlx/include/mlx/backend/metal/kernels/logsumexp.h +140 -0
  102. mlx/include/mlx/backend/metal/kernels/quantized.h +2502 -0
  103. mlx/include/mlx/backend/metal/kernels/quantized_nax.h +1705 -0
  104. mlx/include/mlx/backend/metal/kernels/quantized_utils.h +90 -0
  105. mlx/include/mlx/backend/metal/kernels/reduce.h +5 -0
  106. mlx/include/mlx/backend/metal/kernels/reduce_utils.h +6 -0
  107. mlx/include/mlx/backend/metal/kernels/reduction/ops.h +275 -0
  108. mlx/include/mlx/backend/metal/kernels/reduction/reduce_all.h +66 -0
  109. mlx/include/mlx/backend/metal/kernels/reduction/reduce_col.h +398 -0
  110. mlx/include/mlx/backend/metal/kernels/reduction/reduce_init.h +8 -0
  111. mlx/include/mlx/backend/metal/kernels/reduction/reduce_row.h +369 -0
  112. mlx/include/mlx/backend/metal/kernels/scan.h +514 -0
  113. mlx/include/mlx/backend/metal/kernels/sdpa_vector.h +415 -0
  114. mlx/include/mlx/backend/metal/kernels/softmax.h +190 -0
  115. mlx/include/mlx/backend/metal/kernels/sort.h +715 -0
  116. mlx/include/mlx/backend/metal/kernels/steel/attn/attn.h +296 -0
  117. mlx/include/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention.h +476 -0
  118. mlx/include/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention_nax.h +481 -0
  119. mlx/include/mlx/backend/metal/kernels/steel/attn/loader.h +264 -0
  120. mlx/include/mlx/backend/metal/kernels/steel/attn/mma.h +750 -0
  121. mlx/include/mlx/backend/metal/kernels/steel/attn/nax.h +1076 -0
  122. mlx/include/mlx/backend/metal/kernels/steel/attn/params.h +44 -0
  123. mlx/include/mlx/backend/metal/kernels/steel/attn/transforms.h +71 -0
  124. mlx/include/mlx/backend/metal/kernels/steel/conv/conv.h +13 -0
  125. mlx/include/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv.h +176 -0
  126. mlx/include/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv_general.h +225 -0
  127. mlx/include/mlx/backend/metal/kernels/steel/conv/loader.h +6 -0
  128. mlx/include/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_l.h +451 -0
  129. mlx/include/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_n.h +319 -0
  130. mlx/include/mlx/backend/metal/kernels/steel/conv/loaders/loader_general.h +381 -0
  131. mlx/include/mlx/backend/metal/kernels/steel/conv/params.h +62 -0
  132. mlx/include/mlx/backend/metal/kernels/steel/defines.h +7 -0
  133. mlx/include/mlx/backend/metal/kernels/steel/gemm/gemm.h +295 -0
  134. mlx/include/mlx/backend/metal/kernels/steel/gemm/gemm_nax.h +156 -0
  135. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused.h +346 -0
  136. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused_nax.h +207 -0
  137. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather.h +459 -0
  138. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather_nax.h +132 -0
  139. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_masked.h +719 -0
  140. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_segmented.h +266 -0
  141. mlx/include/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk.h +227 -0
  142. mlx/include/mlx/backend/metal/kernels/steel/gemm/loader.h +137 -0
  143. mlx/include/mlx/backend/metal/kernels/steel/gemm/mma.h +1146 -0
  144. mlx/include/mlx/backend/metal/kernels/steel/gemm/nax.h +1084 -0
  145. mlx/include/mlx/backend/metal/kernels/steel/gemm/params.h +64 -0
  146. mlx/include/mlx/backend/metal/kernels/steel/gemm/transforms.h +72 -0
  147. mlx/include/mlx/backend/metal/kernels/steel/utils/integral_constant.h +134 -0
  148. mlx/include/mlx/backend/metal/kernels/steel/utils/type_traits.h +55 -0
  149. mlx/include/mlx/backend/metal/kernels/steel/utils.h +42 -0
  150. mlx/include/mlx/backend/metal/kernels/ternary.h +145 -0
  151. mlx/include/mlx/backend/metal/kernels/ternary_ops.h +10 -0
  152. mlx/include/mlx/backend/metal/kernels/unary.h +63 -0
  153. mlx/include/mlx/backend/metal/kernels/unary_ops.h +454 -0
  154. mlx/include/mlx/backend/metal/kernels/utils.h +444 -0
  155. mlx/include/mlx/backend/metal/matmul.h +144 -0
  156. mlx/include/mlx/backend/metal/metal.h +22 -0
  157. mlx/include/mlx/backend/metal/reduce.h +41 -0
  158. mlx/include/mlx/backend/metal/resident.h +32 -0
  159. mlx/include/mlx/backend/metal/scan.h +17 -0
  160. mlx/include/mlx/backend/metal/ternary.h +21 -0
  161. mlx/include/mlx/backend/metal/unary.h +21 -0
  162. mlx/include/mlx/backend/metal/utils.h +84 -0
  163. mlx/include/mlx/backend/no_gpu/apple_memory.h +16 -0
  164. mlx/include/mlx/backend/no_gpu/linux_memory.h +22 -0
  165. mlx/include/mlx/compile.h +44 -0
  166. mlx/include/mlx/compile_impl.h +69 -0
  167. mlx/include/mlx/device.h +31 -0
  168. mlx/include/mlx/distributed/distributed.h +60 -0
  169. mlx/include/mlx/distributed/distributed_impl.h +59 -0
  170. mlx/include/mlx/distributed/jaccl/jaccl.h +12 -0
  171. mlx/include/mlx/distributed/mpi/mpi.h +12 -0
  172. mlx/include/mlx/distributed/mpi/mpi_declarations.h +28 -0
  173. mlx/include/mlx/distributed/nccl/nccl.h +12 -0
  174. mlx/include/mlx/distributed/ops.h +56 -0
  175. mlx/include/mlx/distributed/primitives.h +156 -0
  176. mlx/include/mlx/distributed/reduction_ops.h +38 -0
  177. mlx/include/mlx/distributed/ring/ring.h +12 -0
  178. mlx/include/mlx/distributed/utils.h +67 -0
  179. mlx/include/mlx/dtype.h +115 -0
  180. mlx/include/mlx/dtype_utils.h +119 -0
  181. mlx/include/mlx/einsum.h +22 -0
  182. mlx/include/mlx/event.h +58 -0
  183. mlx/include/mlx/export.h +136 -0
  184. mlx/include/mlx/export_impl.h +98 -0
  185. mlx/include/mlx/fast.h +102 -0
  186. mlx/include/mlx/fast_primitives.h +427 -0
  187. mlx/include/mlx/fence.h +39 -0
  188. mlx/include/mlx/fft.h +167 -0
  189. mlx/include/mlx/graph_utils.h +66 -0
  190. mlx/include/mlx/io/gguf.h +20 -0
  191. mlx/include/mlx/io/load.h +175 -0
  192. mlx/include/mlx/io.h +61 -0
  193. mlx/include/mlx/linalg.h +111 -0
  194. mlx/include/mlx/memory.h +78 -0
  195. mlx/include/mlx/mlx.h +25 -0
  196. mlx/include/mlx/ops.h +1627 -0
  197. mlx/include/mlx/primitives.h +2524 -0
  198. mlx/include/mlx/random.h +282 -0
  199. mlx/include/mlx/scheduler.h +188 -0
  200. mlx/include/mlx/small_vector.h +540 -0
  201. mlx/include/mlx/stream.h +41 -0
  202. mlx/include/mlx/threadpool.h +133 -0
  203. mlx/include/mlx/transforms.h +229 -0
  204. mlx/include/mlx/transforms_impl.h +86 -0
  205. mlx/include/mlx/types/bf16.h +187 -0
  206. mlx/include/mlx/types/complex.h +113 -0
  207. mlx/include/mlx/types/fp16.h +234 -0
  208. mlx/include/mlx/types/half_types.h +58 -0
  209. mlx/include/mlx/types/limits.h +70 -0
  210. mlx/include/mlx/utils.h +175 -0
  211. mlx/include/mlx/version.h +20 -0
  212. mlx/lib/libmlx.so +0 -0
  213. mlx/py.typed +1 -0
  214. mlx/share/cmake/MLX/FindNCCL.cmake +54 -0
  215. mlx/share/cmake/MLX/Findnvpl.cmake +3 -0
  216. mlx/share/cmake/MLX/MLXConfig.cmake +66 -0
  217. mlx/share/cmake/MLX/MLXConfigVersion.cmake +65 -0
  218. mlx/share/cmake/MLX/MLXTargets-release.cmake +19 -0
  219. mlx/share/cmake/MLX/MLXTargets.cmake +106 -0
  220. mlx/share/cmake/MLX/extension.cmake +50 -0
  221. mlx/utils.py +325 -0
  222. mlx_cpu-0.30.1.dist-info/METADATA +142 -0
  223. mlx_cpu-0.30.1.dist-info/RECORD +231 -0
  224. mlx_cpu-0.30.1.dist-info/WHEEL +5 -0
  225. mlx_cpu-0.30.1.dist-info/licenses/LICENSE +21 -0
  226. mlx_cpu-0.30.1.dist-info/sboms/auditwheel.cdx.json +1 -0
  227. mlx_cpu-0.30.1.dist-info/top_level.txt +1 -0
  228. mlx_cpu.libs/libblas-bd8a282c.so.3.10.0 +0 -0
  229. mlx_cpu.libs/libgfortran-3ec47101.so.5.0.0 +0 -0
  230. mlx_cpu.libs/liblapack-86b2c207.so.3.10.0 +0 -0
  231. mlx_cpu.libs/libquadmath-67d31475.so.0.0.0 +0 -0
mlx/__main__.py ADDED
@@ -0,0 +1,27 @@
1
+ import argparse
2
+
3
+
4
+ def main() -> None:
5
+ from mlx.core import __version__
6
+
7
+ parser = argparse.ArgumentParser()
8
+ parser.add_argument(
9
+ "--version",
10
+ action="version",
11
+ version=__version__,
12
+ help="Print the version number.",
13
+ )
14
+ parser.add_argument(
15
+ "--cmake-dir",
16
+ action="store_true",
17
+ help="Print the path to the MLX CMake module directory.",
18
+ )
19
+ args = parser.parse_args()
20
+ if args.cmake_dir:
21
+ from pathlib import Path
22
+
23
+ print(Path(__file__).parent)
24
+
25
+
26
+ if __name__ == "__main__":
27
+ main()
mlx/_reprlib_fix.py ADDED
@@ -0,0 +1,16 @@
1
+ # Copyright © 2023 Apple Inc.
2
+
3
+ import array
4
+ import reprlib
5
+
6
+ _old_repr_array = reprlib.Repr.repr_array
7
+
8
+
9
+ def repr_array(self, x, maxlevel):
10
+ if isinstance(x, array.array):
11
+ return _old_repr_array(self, x, maxlevel)
12
+ else:
13
+ return self.repr_instance(x, maxlevel)
14
+
15
+
16
+ reprlib.Repr.repr_array = repr_array
mlx/extension.py ADDED
@@ -0,0 +1,88 @@
1
+ # Copyright © 2023 Apple Inc.
2
+
3
+ import os
4
+ import re
5
+ import subprocess
6
+ import sys
7
+ from pathlib import Path
8
+
9
+ from setuptools import Extension
10
+ from setuptools.command.build_ext import build_ext
11
+
12
+ import mlx
13
+
14
+ _MLX_PATH = str(mlx.__path__[0])
15
+
16
+
17
+ # A CMakeExtension needs a sourcedir instead of a file list.
18
+ class CMakeExtension(Extension):
19
+ def __init__(self, name: str, sourcedir: str = "") -> None:
20
+ super().__init__(name, sources=[])
21
+ self.sourcedir = os.fspath(Path(sourcedir).resolve())
22
+
23
+
24
+ class CMakeBuild(build_ext):
25
+ def build_extension(self, ext: CMakeExtension) -> None:
26
+ # Must be in this form due to bug in .resolve() only fixed in Python 3.10+
27
+ ext_fullpath = Path.cwd() / self.get_ext_fullpath(ext.name) # type: ignore[no-untyped-call]
28
+ extdir = ext_fullpath.parent.resolve()
29
+
30
+ debug = int(os.environ.get("DEBUG", 0)) if self.debug is None else self.debug
31
+ cfg = "Debug" if debug else "Release"
32
+
33
+ # Set Python_EXECUTABLE instead if you use PYBIND11_FINDPYTHON
34
+ # EXAMPLE_VERSION_INFO shows you how to pass a value into the C++ code
35
+ # from Python.
36
+ cmake_args = [
37
+ f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={extdir}{os.sep}",
38
+ f"-DCMAKE_BUILD_TYPE={cfg}",
39
+ "-DBUILD_SHARED_LIBS=ON",
40
+ ]
41
+ build_args = []
42
+ # Adding CMake arguments set as environment variable
43
+ # (needed e.g. to build for ARM OSx on conda-forge)
44
+ if "CMAKE_ARGS" in os.environ:
45
+ cmake_args += [item for item in os.environ["CMAKE_ARGS"].split(" ") if item]
46
+
47
+ if sys.platform.startswith("darwin"):
48
+ # Cross-compile support for macOS - respect ARCHFLAGS if set
49
+ archs = re.findall(r"-arch (\S+)", os.environ.get("ARCHFLAGS", ""))
50
+ if archs:
51
+ cmake_args += ["-DCMAKE_OSX_ARCHITECTURES={}".format(";".join(archs))]
52
+
53
+ # Set CMAKE_BUILD_PARALLEL_LEVEL to control the parallel build level
54
+ # across all generators.
55
+ if "CMAKE_BUILD_PARALLEL_LEVEL" not in os.environ:
56
+ build_args += [f"-j{os.cpu_count()}"]
57
+
58
+ build_temp = Path(self.build_temp) / ext.name
59
+ if not build_temp.exists():
60
+ build_temp.mkdir(parents=True)
61
+
62
+ # Make sure cmake can find MLX
63
+ os.environ["MLX_DIR"] = _MLX_PATH
64
+
65
+ subprocess.run(
66
+ ["cmake", ext.sourcedir, *cmake_args], cwd=build_temp, check=True
67
+ )
68
+ subprocess.run(
69
+ ["cmake", "--build", ".", *build_args], cwd=build_temp, check=True
70
+ )
71
+
72
+ def run(self) -> None:
73
+ super().run()
74
+
75
+ # Based on https://github.com/pypa/setuptools/blob/main/setuptools/command/build_ext.py#L102
76
+ if self.inplace:
77
+ for ext in self.extensions:
78
+ if isinstance(ext, CMakeExtension):
79
+ # Resolve inplace package dir
80
+ build_py = self.get_finalized_command("build_py")
81
+ inplace_file, regular_file = self._get_inplace_equivalent(
82
+ build_py, ext
83
+ )
84
+
85
+ inplace_dir = str(Path(inplace_file).parent.resolve())
86
+ regular_dir = str(Path(regular_file).parent.resolve())
87
+
88
+ self.copy_tree(regular_dir, inplace_dir)