mlx 0.30.7

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 (599) hide show
  1. checksums.yaml +7 -0
  2. data/ext/mlx/extconf.rb +94 -0
  3. data/ext/mlx/native.cpp +8027 -0
  4. data/lib/mlx/core.rb +1678 -0
  5. data/lib/mlx/distributed_utils/common.rb +116 -0
  6. data/lib/mlx/distributed_utils/config.rb +600 -0
  7. data/lib/mlx/distributed_utils/launch.rb +490 -0
  8. data/lib/mlx/extension.rb +24 -0
  9. data/lib/mlx/nn/base.rb +388 -0
  10. data/lib/mlx/nn/init.rb +140 -0
  11. data/lib/mlx/nn/layers/activations.rb +336 -0
  12. data/lib/mlx/nn/layers/base.rb +6 -0
  13. data/lib/mlx/nn/layers/containers.rb +20 -0
  14. data/lib/mlx/nn/layers/convolution.rb +120 -0
  15. data/lib/mlx/nn/layers/convolution_transpose.rb +114 -0
  16. data/lib/mlx/nn/layers/distributed.rb +309 -0
  17. data/lib/mlx/nn/layers/dropout.rb +75 -0
  18. data/lib/mlx/nn/layers/embedding.rb +28 -0
  19. data/lib/mlx/nn/layers/linear.rb +79 -0
  20. data/lib/mlx/nn/layers/normalization.rb +216 -0
  21. data/lib/mlx/nn/layers/pooling.rb +167 -0
  22. data/lib/mlx/nn/layers/positional_encoding.rb +126 -0
  23. data/lib/mlx/nn/layers/quantized.rb +215 -0
  24. data/lib/mlx/nn/layers/recurrent.rb +135 -0
  25. data/lib/mlx/nn/layers/transformer.rb +330 -0
  26. data/lib/mlx/nn/layers/upsample.rb +97 -0
  27. data/lib/mlx/nn/layers.rb +18 -0
  28. data/lib/mlx/nn/losses.rb +251 -0
  29. data/lib/mlx/nn/utils.rb +167 -0
  30. data/lib/mlx/nn.rb +12 -0
  31. data/lib/mlx/optimizers/optimizers.rb +808 -0
  32. data/lib/mlx/optimizers/schedulers.rb +62 -0
  33. data/lib/mlx/optimizers.rb +9 -0
  34. data/lib/mlx/utils.rb +171 -0
  35. data/lib/mlx/version.rb +5 -0
  36. data/lib/mlx.rb +64 -0
  37. data/mlx/CMakeLists.txt +449 -0
  38. data/mlx/cmake/FindCUDNN.cmake +177 -0
  39. data/mlx/cmake/FindNCCL.cmake +54 -0
  40. data/mlx/cmake/Findnvpl.cmake +3 -0
  41. data/mlx/cmake/extension.cmake +50 -0
  42. data/mlx/mlx/3rdparty/.clang-format +2 -0
  43. data/mlx/mlx/3rdparty/pocketfft.h +3581 -0
  44. data/mlx/mlx/CMakeLists.txt +107 -0
  45. data/mlx/mlx/allocator.h +75 -0
  46. data/mlx/mlx/api.h +29 -0
  47. data/mlx/mlx/array.cpp +354 -0
  48. data/mlx/mlx/array.h +647 -0
  49. data/mlx/mlx/backend/common/CMakeLists.txt +9 -0
  50. data/mlx/mlx/backend/common/binary.h +97 -0
  51. data/mlx/mlx/backend/common/broadcasting.cpp +24 -0
  52. data/mlx/mlx/backend/common/broadcasting.h +11 -0
  53. data/mlx/mlx/backend/common/buffer_cache.h +158 -0
  54. data/mlx/mlx/backend/common/common.cpp +305 -0
  55. data/mlx/mlx/backend/common/compiled.cpp +243 -0
  56. data/mlx/mlx/backend/common/compiled.h +77 -0
  57. data/mlx/mlx/backend/common/copy.h +50 -0
  58. data/mlx/mlx/backend/common/hadamard.h +109 -0
  59. data/mlx/mlx/backend/common/load.cpp +57 -0
  60. data/mlx/mlx/backend/common/matmul.h +67 -0
  61. data/mlx/mlx/backend/common/reduce.cpp +154 -0
  62. data/mlx/mlx/backend/common/reduce.h +59 -0
  63. data/mlx/mlx/backend/common/slicing.cpp +71 -0
  64. data/mlx/mlx/backend/common/slicing.h +20 -0
  65. data/mlx/mlx/backend/common/ternary.h +85 -0
  66. data/mlx/mlx/backend/common/unary.h +29 -0
  67. data/mlx/mlx/backend/common/utils.cpp +231 -0
  68. data/mlx/mlx/backend/common/utils.h +205 -0
  69. data/mlx/mlx/backend/cpu/CMakeLists.txt +88 -0
  70. data/mlx/mlx/backend/cpu/arange.h +28 -0
  71. data/mlx/mlx/backend/cpu/arg_reduce.cpp +124 -0
  72. data/mlx/mlx/backend/cpu/binary.cpp +269 -0
  73. data/mlx/mlx/backend/cpu/binary.h +517 -0
  74. data/mlx/mlx/backend/cpu/binary_ops.h +98 -0
  75. data/mlx/mlx/backend/cpu/binary_two.h +166 -0
  76. data/mlx/mlx/backend/cpu/cholesky.cpp +85 -0
  77. data/mlx/mlx/backend/cpu/compiled.cpp +357 -0
  78. data/mlx/mlx/backend/cpu/compiled_preamble.h +12 -0
  79. data/mlx/mlx/backend/cpu/conv.cpp +1351 -0
  80. data/mlx/mlx/backend/cpu/copy.cpp +386 -0
  81. data/mlx/mlx/backend/cpu/copy.h +36 -0
  82. data/mlx/mlx/backend/cpu/device_info.cpp +113 -0
  83. data/mlx/mlx/backend/cpu/device_info.h +28 -0
  84. data/mlx/mlx/backend/cpu/distributed.cpp +103 -0
  85. data/mlx/mlx/backend/cpu/eig.cpp +281 -0
  86. data/mlx/mlx/backend/cpu/eigh.cpp +241 -0
  87. data/mlx/mlx/backend/cpu/encoder.cpp +16 -0
  88. data/mlx/mlx/backend/cpu/encoder.h +67 -0
  89. data/mlx/mlx/backend/cpu/eval.cpp +40 -0
  90. data/mlx/mlx/backend/cpu/eval.h +12 -0
  91. data/mlx/mlx/backend/cpu/fft.cpp +120 -0
  92. data/mlx/mlx/backend/cpu/gemm.h +26 -0
  93. data/mlx/mlx/backend/cpu/gemms/bnns.cpp +214 -0
  94. data/mlx/mlx/backend/cpu/gemms/cblas.cpp +134 -0
  95. data/mlx/mlx/backend/cpu/gemms/simd_bf16.cpp +45 -0
  96. data/mlx/mlx/backend/cpu/gemms/simd_fp16.cpp +45 -0
  97. data/mlx/mlx/backend/cpu/gemms/simd_gemm.h +139 -0
  98. data/mlx/mlx/backend/cpu/hadamard.cpp +121 -0
  99. data/mlx/mlx/backend/cpu/indexing.cpp +854 -0
  100. data/mlx/mlx/backend/cpu/inverse.cpp +160 -0
  101. data/mlx/mlx/backend/cpu/jit_compiler.cpp +166 -0
  102. data/mlx/mlx/backend/cpu/jit_compiler.h +20 -0
  103. data/mlx/mlx/backend/cpu/lapack.h +80 -0
  104. data/mlx/mlx/backend/cpu/logsumexp.cpp +139 -0
  105. data/mlx/mlx/backend/cpu/luf.cpp +120 -0
  106. data/mlx/mlx/backend/cpu/make_compiled_preamble.ps1 +38 -0
  107. data/mlx/mlx/backend/cpu/make_compiled_preamble.sh +41 -0
  108. data/mlx/mlx/backend/cpu/masked_mm.cpp +608 -0
  109. data/mlx/mlx/backend/cpu/matmul.cpp +166 -0
  110. data/mlx/mlx/backend/cpu/primitives.cpp +478 -0
  111. data/mlx/mlx/backend/cpu/qrf.cpp +147 -0
  112. data/mlx/mlx/backend/cpu/quantized.cpp +1370 -0
  113. data/mlx/mlx/backend/cpu/reduce.cpp +587 -0
  114. data/mlx/mlx/backend/cpu/scan.cpp +338 -0
  115. data/mlx/mlx/backend/cpu/select.cpp +95 -0
  116. data/mlx/mlx/backend/cpu/simd/accelerate_fp16_simd.h +56 -0
  117. data/mlx/mlx/backend/cpu/simd/accelerate_simd.h +329 -0
  118. data/mlx/mlx/backend/cpu/simd/base_simd.h +319 -0
  119. data/mlx/mlx/backend/cpu/simd/math.h +193 -0
  120. data/mlx/mlx/backend/cpu/simd/neon_fp16_simd.h +212 -0
  121. data/mlx/mlx/backend/cpu/simd/simd.h +4 -0
  122. data/mlx/mlx/backend/cpu/simd/type.h +11 -0
  123. data/mlx/mlx/backend/cpu/slicing.h +21 -0
  124. data/mlx/mlx/backend/cpu/softmax.cpp +170 -0
  125. data/mlx/mlx/backend/cpu/sort.cpp +481 -0
  126. data/mlx/mlx/backend/cpu/svd.cpp +289 -0
  127. data/mlx/mlx/backend/cpu/ternary.h +154 -0
  128. data/mlx/mlx/backend/cpu/threefry.cpp +31 -0
  129. data/mlx/mlx/backend/cpu/threefry.h +21 -0
  130. data/mlx/mlx/backend/cpu/unary.cpp +238 -0
  131. data/mlx/mlx/backend/cpu/unary.h +281 -0
  132. data/mlx/mlx/backend/cpu/unary_ops.h +175 -0
  133. data/mlx/mlx/backend/cuda/CMakeLists.txt +265 -0
  134. data/mlx/mlx/backend/cuda/allocator.cpp +451 -0
  135. data/mlx/mlx/backend/cuda/allocator.h +94 -0
  136. data/mlx/mlx/backend/cuda/arange.cu +68 -0
  137. data/mlx/mlx/backend/cuda/arg_reduce.cu +189 -0
  138. data/mlx/mlx/backend/cuda/bin2h.cmake +150 -0
  139. data/mlx/mlx/backend/cuda/binary/CMakeLists.txt +21 -0
  140. data/mlx/mlx/backend/cuda/binary/add.cu +7 -0
  141. data/mlx/mlx/backend/cuda/binary/arctan2.cu +7 -0
  142. data/mlx/mlx/backend/cuda/binary/binary.cuh +383 -0
  143. data/mlx/mlx/backend/cuda/binary/bitwise_binary.cu +27 -0
  144. data/mlx/mlx/backend/cuda/binary/divide.cu +7 -0
  145. data/mlx/mlx/backend/cuda/binary/equal.cu +15 -0
  146. data/mlx/mlx/backend/cuda/binary/greater.cu +7 -0
  147. data/mlx/mlx/backend/cuda/binary/greater_equal.cu +7 -0
  148. data/mlx/mlx/backend/cuda/binary/less.cu +7 -0
  149. data/mlx/mlx/backend/cuda/binary/less_equal.cu +7 -0
  150. data/mlx/mlx/backend/cuda/binary/log_add_exp.cu +7 -0
  151. data/mlx/mlx/backend/cuda/binary/logical_and.cu +7 -0
  152. data/mlx/mlx/backend/cuda/binary/logical_or.cu +7 -0
  153. data/mlx/mlx/backend/cuda/binary/maximum.cu +7 -0
  154. data/mlx/mlx/backend/cuda/binary/minimum.cu +7 -0
  155. data/mlx/mlx/backend/cuda/binary/multiply.cu +7 -0
  156. data/mlx/mlx/backend/cuda/binary/not_equal.cu +7 -0
  157. data/mlx/mlx/backend/cuda/binary/power.cu +7 -0
  158. data/mlx/mlx/backend/cuda/binary/remainder.cu +7 -0
  159. data/mlx/mlx/backend/cuda/binary/subtract.cu +7 -0
  160. data/mlx/mlx/backend/cuda/binary_two.cu +412 -0
  161. data/mlx/mlx/backend/cuda/compiled.cpp +357 -0
  162. data/mlx/mlx/backend/cuda/conv/conv.h +126 -0
  163. data/mlx/mlx/backend/cuda/conv/gemm_conv.cu +217 -0
  164. data/mlx/mlx/backend/cuda/conv/gemm_grouped_conv.cu +231 -0
  165. data/mlx/mlx/backend/cuda/conv.cpp +403 -0
  166. data/mlx/mlx/backend/cuda/copy/copy.cuh +55 -0
  167. data/mlx/mlx/backend/cuda/copy/copy_contiguous.cu +88 -0
  168. data/mlx/mlx/backend/cuda/copy/copy_general.cu +171 -0
  169. data/mlx/mlx/backend/cuda/copy/copy_general_dynamic.cu +118 -0
  170. data/mlx/mlx/backend/cuda/copy/copy_general_input.cu +229 -0
  171. data/mlx/mlx/backend/cuda/copy.cu +132 -0
  172. data/mlx/mlx/backend/cuda/cublas_utils.cpp +222 -0
  173. data/mlx/mlx/backend/cuda/cublas_utils.h +95 -0
  174. data/mlx/mlx/backend/cuda/cuda.h +21 -0
  175. data/mlx/mlx/backend/cuda/cuda_utils.h +90 -0
  176. data/mlx/mlx/backend/cuda/cudnn_utils.cpp +133 -0
  177. data/mlx/mlx/backend/cuda/cudnn_utils.h +187 -0
  178. data/mlx/mlx/backend/cuda/custom_kernel.cpp +379 -0
  179. data/mlx/mlx/backend/cuda/cutlass_utils.cuh +46 -0
  180. data/mlx/mlx/backend/cuda/delayload.cpp +80 -0
  181. data/mlx/mlx/backend/cuda/device/atomic_ops.cuh +63 -0
  182. data/mlx/mlx/backend/cuda/device/binary_ops.cuh +300 -0
  183. data/mlx/mlx/backend/cuda/device/cast_op.cuh +118 -0
  184. data/mlx/mlx/backend/cuda/device/complex.cuh +60 -0
  185. data/mlx/mlx/backend/cuda/device/config.h +12 -0
  186. data/mlx/mlx/backend/cuda/device/fp16_math.cuh +96 -0
  187. data/mlx/mlx/backend/cuda/device/gather.cuh +53 -0
  188. data/mlx/mlx/backend/cuda/device/gather_axis.cuh +65 -0
  189. data/mlx/mlx/backend/cuda/device/indexing.cuh +30 -0
  190. data/mlx/mlx/backend/cuda/device/scatter.cuh +68 -0
  191. data/mlx/mlx/backend/cuda/device/scatter_axis.cuh +67 -0
  192. data/mlx/mlx/backend/cuda/device/scatter_ops.cuh +44 -0
  193. data/mlx/mlx/backend/cuda/device/ternary_ops.cuh +13 -0
  194. data/mlx/mlx/backend/cuda/device/unary_ops.cuh +350 -0
  195. data/mlx/mlx/backend/cuda/device/utils.cuh +464 -0
  196. data/mlx/mlx/backend/cuda/device.cpp +522 -0
  197. data/mlx/mlx/backend/cuda/device.h +195 -0
  198. data/mlx/mlx/backend/cuda/device_info.cpp +232 -0
  199. data/mlx/mlx/backend/cuda/distributed.cu +121 -0
  200. data/mlx/mlx/backend/cuda/eval.cpp +66 -0
  201. data/mlx/mlx/backend/cuda/event.cu +415 -0
  202. data/mlx/mlx/backend/cuda/event.h +79 -0
  203. data/mlx/mlx/backend/cuda/fence.cpp +42 -0
  204. data/mlx/mlx/backend/cuda/gemms/cublas_gemm.cpp +233 -0
  205. data/mlx/mlx/backend/cuda/gemms/cublas_gemm.h +114 -0
  206. data/mlx/mlx/backend/cuda/gemms/cublas_gemm_batched_12_0.cpp +77 -0
  207. data/mlx/mlx/backend/cuda/gemms/cublas_gemm_batched_12_9.cu +329 -0
  208. data/mlx/mlx/backend/cuda/gemms/gemv.cu +327 -0
  209. data/mlx/mlx/backend/cuda/gemms/gemv.h +34 -0
  210. data/mlx/mlx/backend/cuda/gemms/grouped_gemm.h +25 -0
  211. data/mlx/mlx/backend/cuda/gemms/grouped_gemm_unaligned.cu +358 -0
  212. data/mlx/mlx/backend/cuda/indexing.cpp +434 -0
  213. data/mlx/mlx/backend/cuda/jit_module.cpp +443 -0
  214. data/mlx/mlx/backend/cuda/jit_module.h +120 -0
  215. data/mlx/mlx/backend/cuda/kernel_utils.cu +52 -0
  216. data/mlx/mlx/backend/cuda/kernel_utils.cuh +148 -0
  217. data/mlx/mlx/backend/cuda/layer_norm.cu +417 -0
  218. data/mlx/mlx/backend/cuda/load.cpp +60 -0
  219. data/mlx/mlx/backend/cuda/logsumexp.cu +161 -0
  220. data/mlx/mlx/backend/cuda/lru_cache.h +190 -0
  221. data/mlx/mlx/backend/cuda/matmul.cpp +373 -0
  222. data/mlx/mlx/backend/cuda/no_cuda.cpp +47 -0
  223. data/mlx/mlx/backend/cuda/primitives.cpp +46 -0
  224. data/mlx/mlx/backend/cuda/quantized/affine_quantize.cu +329 -0
  225. data/mlx/mlx/backend/cuda/quantized/convert_fp8.cu +19 -0
  226. data/mlx/mlx/backend/cuda/quantized/cublas_qqmm.cpp +206 -0
  227. data/mlx/mlx/backend/cuda/quantized/cublas_qqmm.h +88 -0
  228. data/mlx/mlx/backend/cuda/quantized/cuda_fp4.h +100 -0
  229. data/mlx/mlx/backend/cuda/quantized/fp_quantize.cu +496 -0
  230. data/mlx/mlx/backend/cuda/quantized/mxfp8_quantize.cuh +32 -0
  231. data/mlx/mlx/backend/cuda/quantized/no_qqmm_impl.cpp +26 -0
  232. data/mlx/mlx/backend/cuda/quantized/nvfp4_quantize.cuh +334 -0
  233. data/mlx/mlx/backend/cuda/quantized/qmv.cu +304 -0
  234. data/mlx/mlx/backend/cuda/quantized/qmv.h +21 -0
  235. data/mlx/mlx/backend/cuda/quantized/qqmm.cpp +158 -0
  236. data/mlx/mlx/backend/cuda/quantized/qqmm_impl.cpp +50 -0
  237. data/mlx/mlx/backend/cuda/quantized/qqmm_impl.h +26 -0
  238. data/mlx/mlx/backend/cuda/quantized/qqmm_utils.cu +227 -0
  239. data/mlx/mlx/backend/cuda/quantized/qqmm_utils.h +30 -0
  240. data/mlx/mlx/backend/cuda/quantized/quantized.cpp +85 -0
  241. data/mlx/mlx/backend/cuda/quantized/quantized.h +53 -0
  242. data/mlx/mlx/backend/cuda/quantized/quantized_utils.cuh +88 -0
  243. data/mlx/mlx/backend/cuda/quantized/quantized_utils.h +50 -0
  244. data/mlx/mlx/backend/cuda/random.cu +202 -0
  245. data/mlx/mlx/backend/cuda/reduce/all_reduce.cu +159 -0
  246. data/mlx/mlx/backend/cuda/reduce/col_reduce.cu +510 -0
  247. data/mlx/mlx/backend/cuda/reduce/init_reduce.cu +50 -0
  248. data/mlx/mlx/backend/cuda/reduce/reduce.cuh +71 -0
  249. data/mlx/mlx/backend/cuda/reduce/reduce_ops.cuh +211 -0
  250. data/mlx/mlx/backend/cuda/reduce/reduce_utils.cuh +145 -0
  251. data/mlx/mlx/backend/cuda/reduce/row_reduce.cu +361 -0
  252. data/mlx/mlx/backend/cuda/reduce.cu +73 -0
  253. data/mlx/mlx/backend/cuda/rms_norm.cu +536 -0
  254. data/mlx/mlx/backend/cuda/rope.cu +429 -0
  255. data/mlx/mlx/backend/cuda/scaled_dot_product_attention.cpp +681 -0
  256. data/mlx/mlx/backend/cuda/scaled_dot_product_attention.cu +796 -0
  257. data/mlx/mlx/backend/cuda/scan.cu +468 -0
  258. data/mlx/mlx/backend/cuda/slicing.cpp +111 -0
  259. data/mlx/mlx/backend/cuda/softmax.cu +162 -0
  260. data/mlx/mlx/backend/cuda/sort.cu +1076 -0
  261. data/mlx/mlx/backend/cuda/steel/defines.cuh +9 -0
  262. data/mlx/mlx/backend/cuda/steel/gemm.cuh +101 -0
  263. data/mlx/mlx/backend/cuda/steel/mma.cuh +117 -0
  264. data/mlx/mlx/backend/cuda/steel/tiles.cuh +450 -0
  265. data/mlx/mlx/backend/cuda/steel/utils.cuh +89 -0
  266. data/mlx/mlx/backend/cuda/ternary.cu +271 -0
  267. data/mlx/mlx/backend/cuda/unary/CMakeLists.txt +34 -0
  268. data/mlx/mlx/backend/cuda/unary/abs.cu +7 -0
  269. data/mlx/mlx/backend/cuda/unary/arccos.cu +7 -0
  270. data/mlx/mlx/backend/cuda/unary/arccosh.cu +7 -0
  271. data/mlx/mlx/backend/cuda/unary/arcsin.cu +7 -0
  272. data/mlx/mlx/backend/cuda/unary/arcsinh.cu +7 -0
  273. data/mlx/mlx/backend/cuda/unary/arctan.cu +7 -0
  274. data/mlx/mlx/backend/cuda/unary/arctanh.cu +7 -0
  275. data/mlx/mlx/backend/cuda/unary/bitwise_invert.cu +7 -0
  276. data/mlx/mlx/backend/cuda/unary/ceil.cu +7 -0
  277. data/mlx/mlx/backend/cuda/unary/conjugate.cu +7 -0
  278. data/mlx/mlx/backend/cuda/unary/cos.cu +7 -0
  279. data/mlx/mlx/backend/cuda/unary/cosh.cu +7 -0
  280. data/mlx/mlx/backend/cuda/unary/erf.cu +7 -0
  281. data/mlx/mlx/backend/cuda/unary/erf_inv.cu +7 -0
  282. data/mlx/mlx/backend/cuda/unary/exp.cu +7 -0
  283. data/mlx/mlx/backend/cuda/unary/expm1.cu +7 -0
  284. data/mlx/mlx/backend/cuda/unary/floor.cu +7 -0
  285. data/mlx/mlx/backend/cuda/unary/imag.cu +7 -0
  286. data/mlx/mlx/backend/cuda/unary/log.cu +21 -0
  287. data/mlx/mlx/backend/cuda/unary/log1p.cu +7 -0
  288. data/mlx/mlx/backend/cuda/unary/logical_not.cu +7 -0
  289. data/mlx/mlx/backend/cuda/unary/negative.cu +7 -0
  290. data/mlx/mlx/backend/cuda/unary/real.cu +7 -0
  291. data/mlx/mlx/backend/cuda/unary/round.cu +18 -0
  292. data/mlx/mlx/backend/cuda/unary/sigmoid.cu +7 -0
  293. data/mlx/mlx/backend/cuda/unary/sign.cu +7 -0
  294. data/mlx/mlx/backend/cuda/unary/sin.cu +7 -0
  295. data/mlx/mlx/backend/cuda/unary/sinh.cu +7 -0
  296. data/mlx/mlx/backend/cuda/unary/sqrt.cu +15 -0
  297. data/mlx/mlx/backend/cuda/unary/square.cu +7 -0
  298. data/mlx/mlx/backend/cuda/unary/tan.cu +7 -0
  299. data/mlx/mlx/backend/cuda/unary/tanh.cu +7 -0
  300. data/mlx/mlx/backend/cuda/unary/unary.cuh +224 -0
  301. data/mlx/mlx/backend/cuda/utils.cpp +116 -0
  302. data/mlx/mlx/backend/cuda/utils.h +49 -0
  303. data/mlx/mlx/backend/cuda/vector_types.cuh +48 -0
  304. data/mlx/mlx/backend/cuda/worker.cpp +79 -0
  305. data/mlx/mlx/backend/cuda/worker.h +55 -0
  306. data/mlx/mlx/backend/gpu/CMakeLists.txt +5 -0
  307. data/mlx/mlx/backend/gpu/copy.cpp +89 -0
  308. data/mlx/mlx/backend/gpu/copy.h +57 -0
  309. data/mlx/mlx/backend/gpu/device_info.h +36 -0
  310. data/mlx/mlx/backend/gpu/eval.h +18 -0
  311. data/mlx/mlx/backend/gpu/primitives.cpp +307 -0
  312. data/mlx/mlx/backend/gpu/slicing.cpp +44 -0
  313. data/mlx/mlx/backend/gpu/slicing.h +36 -0
  314. data/mlx/mlx/backend/metal/CMakeLists.txt +144 -0
  315. data/mlx/mlx/backend/metal/allocator.cpp +279 -0
  316. data/mlx/mlx/backend/metal/allocator.h +79 -0
  317. data/mlx/mlx/backend/metal/binary.cpp +257 -0
  318. data/mlx/mlx/backend/metal/binary.h +33 -0
  319. data/mlx/mlx/backend/metal/compiled.cpp +471 -0
  320. data/mlx/mlx/backend/metal/conv.cpp +1118 -0
  321. data/mlx/mlx/backend/metal/copy.cpp +235 -0
  322. data/mlx/mlx/backend/metal/custom_kernel.cpp +430 -0
  323. data/mlx/mlx/backend/metal/device.cpp +816 -0
  324. data/mlx/mlx/backend/metal/device.h +289 -0
  325. data/mlx/mlx/backend/metal/device_info.cpp +58 -0
  326. data/mlx/mlx/backend/metal/distributed.cpp +38 -0
  327. data/mlx/mlx/backend/metal/eval.cpp +97 -0
  328. data/mlx/mlx/backend/metal/event.cpp +62 -0
  329. data/mlx/mlx/backend/metal/fence.cpp +162 -0
  330. data/mlx/mlx/backend/metal/fft.cpp +807 -0
  331. data/mlx/mlx/backend/metal/hadamard.cpp +198 -0
  332. data/mlx/mlx/backend/metal/indexing.cpp +727 -0
  333. data/mlx/mlx/backend/metal/jit/includes.h +58 -0
  334. data/mlx/mlx/backend/metal/jit/indexing.h +76 -0
  335. data/mlx/mlx/backend/metal/jit_kernels.cpp +1118 -0
  336. data/mlx/mlx/backend/metal/kernels/CMakeLists.txt +193 -0
  337. data/mlx/mlx/backend/metal/kernels/arange.h +9 -0
  338. data/mlx/mlx/backend/metal/kernels/arange.metal +20 -0
  339. data/mlx/mlx/backend/metal/kernels/arg_reduce.metal +182 -0
  340. data/mlx/mlx/backend/metal/kernels/atomic.h +345 -0
  341. data/mlx/mlx/backend/metal/kernels/bf16.h +16 -0
  342. data/mlx/mlx/backend/metal/kernels/bf16_math.h +380 -0
  343. data/mlx/mlx/backend/metal/kernels/binary.h +199 -0
  344. data/mlx/mlx/backend/metal/kernels/binary.metal +109 -0
  345. data/mlx/mlx/backend/metal/kernels/binary_ops.h +330 -0
  346. data/mlx/mlx/backend/metal/kernels/binary_two.h +244 -0
  347. data/mlx/mlx/backend/metal/kernels/binary_two.metal +54 -0
  348. data/mlx/mlx/backend/metal/kernels/cexpf.h +134 -0
  349. data/mlx/mlx/backend/metal/kernels/complex.h +173 -0
  350. data/mlx/mlx/backend/metal/kernels/conv.metal +701 -0
  351. data/mlx/mlx/backend/metal/kernels/copy.h +276 -0
  352. data/mlx/mlx/backend/metal/kernels/copy.metal +75 -0
  353. data/mlx/mlx/backend/metal/kernels/defines.h +24 -0
  354. data/mlx/mlx/backend/metal/kernels/erf.h +69 -0
  355. data/mlx/mlx/backend/metal/kernels/expm1f.h +90 -0
  356. data/mlx/mlx/backend/metal/kernels/fence.metal +52 -0
  357. data/mlx/mlx/backend/metal/kernels/fft/radix.h +328 -0
  358. data/mlx/mlx/backend/metal/kernels/fft/readwrite.h +624 -0
  359. data/mlx/mlx/backend/metal/kernels/fft.h +486 -0
  360. data/mlx/mlx/backend/metal/kernels/fft.metal +67 -0
  361. data/mlx/mlx/backend/metal/kernels/fp4.h +48 -0
  362. data/mlx/mlx/backend/metal/kernels/fp8.h +80 -0
  363. data/mlx/mlx/backend/metal/kernels/fp_quantized.h +1850 -0
  364. data/mlx/mlx/backend/metal/kernels/fp_quantized.metal +153 -0
  365. data/mlx/mlx/backend/metal/kernels/fp_quantized_nax.h +1044 -0
  366. data/mlx/mlx/backend/metal/kernels/fp_quantized_nax.metal +79 -0
  367. data/mlx/mlx/backend/metal/kernels/gemv.metal +868 -0
  368. data/mlx/mlx/backend/metal/kernels/gemv_masked.h +827 -0
  369. data/mlx/mlx/backend/metal/kernels/gemv_masked.metal +76 -0
  370. data/mlx/mlx/backend/metal/kernels/hadamard.h +182 -0
  371. data/mlx/mlx/backend/metal/kernels/indexing/gather.h +51 -0
  372. data/mlx/mlx/backend/metal/kernels/indexing/gather_axis.h +44 -0
  373. data/mlx/mlx/backend/metal/kernels/indexing/gather_front.h +24 -0
  374. data/mlx/mlx/backend/metal/kernels/indexing/indexing.h +23 -0
  375. data/mlx/mlx/backend/metal/kernels/indexing/masked_scatter.h +41 -0
  376. data/mlx/mlx/backend/metal/kernels/indexing/scatter.h +59 -0
  377. data/mlx/mlx/backend/metal/kernels/indexing/scatter_axis.h +52 -0
  378. data/mlx/mlx/backend/metal/kernels/layer_norm.metal +433 -0
  379. data/mlx/mlx/backend/metal/kernels/logging.h +26 -0
  380. data/mlx/mlx/backend/metal/kernels/logsumexp.h +140 -0
  381. data/mlx/mlx/backend/metal/kernels/logsumexp.metal +18 -0
  382. data/mlx/mlx/backend/metal/kernels/quantized.h +2508 -0
  383. data/mlx/mlx/backend/metal/kernels/quantized.metal +144 -0
  384. data/mlx/mlx/backend/metal/kernels/quantized_nax.h +1705 -0
  385. data/mlx/mlx/backend/metal/kernels/quantized_nax.metal +106 -0
  386. data/mlx/mlx/backend/metal/kernels/quantized_utils.h +90 -0
  387. data/mlx/mlx/backend/metal/kernels/random.metal +103 -0
  388. data/mlx/mlx/backend/metal/kernels/reduce.h +5 -0
  389. data/mlx/mlx/backend/metal/kernels/reduce.metal +169 -0
  390. data/mlx/mlx/backend/metal/kernels/reduce_utils.h +6 -0
  391. data/mlx/mlx/backend/metal/kernels/reduction/ops.h +275 -0
  392. data/mlx/mlx/backend/metal/kernels/reduction/reduce_all.h +66 -0
  393. data/mlx/mlx/backend/metal/kernels/reduction/reduce_col.h +398 -0
  394. data/mlx/mlx/backend/metal/kernels/reduction/reduce_init.h +8 -0
  395. data/mlx/mlx/backend/metal/kernels/reduction/reduce_row.h +369 -0
  396. data/mlx/mlx/backend/metal/kernels/rms_norm.metal +391 -0
  397. data/mlx/mlx/backend/metal/kernels/rope.metal +229 -0
  398. data/mlx/mlx/backend/metal/kernels/scaled_dot_product_attention.metal +44 -0
  399. data/mlx/mlx/backend/metal/kernels/scan.h +514 -0
  400. data/mlx/mlx/backend/metal/kernels/scan.metal +109 -0
  401. data/mlx/mlx/backend/metal/kernels/sdpa_vector.h +394 -0
  402. data/mlx/mlx/backend/metal/kernels/softmax.h +190 -0
  403. data/mlx/mlx/backend/metal/kernels/softmax.metal +24 -0
  404. data/mlx/mlx/backend/metal/kernels/sort.h +719 -0
  405. data/mlx/mlx/backend/metal/kernels/sort.metal +80 -0
  406. data/mlx/mlx/backend/metal/kernels/steel/attn/attn.h +296 -0
  407. data/mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention.h +471 -0
  408. data/mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention.metal +27 -0
  409. data/mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention_nax.h +481 -0
  410. data/mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention_nax.metal +28 -0
  411. data/mlx/mlx/backend/metal/kernels/steel/attn/loader.h +264 -0
  412. data/mlx/mlx/backend/metal/kernels/steel/attn/mma.h +750 -0
  413. data/mlx/mlx/backend/metal/kernels/steel/attn/nax.h +1076 -0
  414. data/mlx/mlx/backend/metal/kernels/steel/attn/params.h +44 -0
  415. data/mlx/mlx/backend/metal/kernels/steel/attn/transforms.h +71 -0
  416. data/mlx/mlx/backend/metal/kernels/steel/conv/conv.h +13 -0
  417. data/mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv.h +176 -0
  418. data/mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv.metal +56 -0
  419. data/mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv_general.h +225 -0
  420. data/mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv_general.metal +47 -0
  421. data/mlx/mlx/backend/metal/kernels/steel/conv/loader.h +6 -0
  422. data/mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_l.h +451 -0
  423. data/mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_n.h +319 -0
  424. data/mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_general.h +381 -0
  425. data/mlx/mlx/backend/metal/kernels/steel/conv/params.h +62 -0
  426. data/mlx/mlx/backend/metal/kernels/steel/defines.h +7 -0
  427. data/mlx/mlx/backend/metal/kernels/steel/gemm/gemm.h +295 -0
  428. data/mlx/mlx/backend/metal/kernels/steel/gemm/gemm_nax.h +157 -0
  429. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused.h +346 -0
  430. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused.metal +34 -0
  431. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused_nax.h +219 -0
  432. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused_nax.metal +30 -0
  433. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather.h +459 -0
  434. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather.metal +59 -0
  435. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather_nax.h +143 -0
  436. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather_nax.metal +37 -0
  437. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_masked.h +719 -0
  438. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_masked.metal +76 -0
  439. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_segmented.h +266 -0
  440. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_segmented.metal +43 -0
  441. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk.h +227 -0
  442. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk.metal +76 -0
  443. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk_nax.h +152 -0
  444. data/mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk_nax.metal +30 -0
  445. data/mlx/mlx/backend/metal/kernels/steel/gemm/loader.h +137 -0
  446. data/mlx/mlx/backend/metal/kernels/steel/gemm/mma.h +1146 -0
  447. data/mlx/mlx/backend/metal/kernels/steel/gemm/nax.h +1084 -0
  448. data/mlx/mlx/backend/metal/kernels/steel/gemm/params.h +65 -0
  449. data/mlx/mlx/backend/metal/kernels/steel/gemm/transforms.h +72 -0
  450. data/mlx/mlx/backend/metal/kernels/steel/utils/integral_constant.h +134 -0
  451. data/mlx/mlx/backend/metal/kernels/steel/utils/type_traits.h +55 -0
  452. data/mlx/mlx/backend/metal/kernels/steel/utils.h +42 -0
  453. data/mlx/mlx/backend/metal/kernels/ternary.h +145 -0
  454. data/mlx/mlx/backend/metal/kernels/ternary.metal +48 -0
  455. data/mlx/mlx/backend/metal/kernels/ternary_ops.h +10 -0
  456. data/mlx/mlx/backend/metal/kernels/unary.h +63 -0
  457. data/mlx/mlx/backend/metal/kernels/unary.metal +115 -0
  458. data/mlx/mlx/backend/metal/kernels/unary_ops.h +454 -0
  459. data/mlx/mlx/backend/metal/kernels/utils.h +445 -0
  460. data/mlx/mlx/backend/metal/kernels.h +375 -0
  461. data/mlx/mlx/backend/metal/logsumexp.cpp +95 -0
  462. data/mlx/mlx/backend/metal/make_compiled_preamble.sh +120 -0
  463. data/mlx/mlx/backend/metal/matmul.cpp +2572 -0
  464. data/mlx/mlx/backend/metal/matmul.h +144 -0
  465. data/mlx/mlx/backend/metal/metal.cpp +50 -0
  466. data/mlx/mlx/backend/metal/metal.h +25 -0
  467. data/mlx/mlx/backend/metal/no_metal.cpp +42 -0
  468. data/mlx/mlx/backend/metal/nojit_kernels.cpp +414 -0
  469. data/mlx/mlx/backend/metal/normalization.cpp +433 -0
  470. data/mlx/mlx/backend/metal/primitives.cpp +242 -0
  471. data/mlx/mlx/backend/metal/quantized.cpp +1651 -0
  472. data/mlx/mlx/backend/metal/reduce.cpp +1038 -0
  473. data/mlx/mlx/backend/metal/reduce.h +41 -0
  474. data/mlx/mlx/backend/metal/resident.cpp +100 -0
  475. data/mlx/mlx/backend/metal/resident.h +32 -0
  476. data/mlx/mlx/backend/metal/rope.cpp +165 -0
  477. data/mlx/mlx/backend/metal/scaled_dot_product_attention.cpp +798 -0
  478. data/mlx/mlx/backend/metal/scan.cpp +145 -0
  479. data/mlx/mlx/backend/metal/scan.h +17 -0
  480. data/mlx/mlx/backend/metal/slicing.cpp +99 -0
  481. data/mlx/mlx/backend/metal/softmax.cpp +87 -0
  482. data/mlx/mlx/backend/metal/sort.cpp +368 -0
  483. data/mlx/mlx/backend/metal/ternary.cpp +160 -0
  484. data/mlx/mlx/backend/metal/ternary.h +21 -0
  485. data/mlx/mlx/backend/metal/unary.cpp +161 -0
  486. data/mlx/mlx/backend/metal/unary.h +21 -0
  487. data/mlx/mlx/backend/metal/utils.cpp +77 -0
  488. data/mlx/mlx/backend/metal/utils.h +99 -0
  489. data/mlx/mlx/backend/no_cpu/CMakeLists.txt +7 -0
  490. data/mlx/mlx/backend/no_cpu/compiled.cpp +24 -0
  491. data/mlx/mlx/backend/no_cpu/device_info.cpp +22 -0
  492. data/mlx/mlx/backend/no_cpu/primitives.cpp +146 -0
  493. data/mlx/mlx/backend/no_gpu/CMakeLists.txt +8 -0
  494. data/mlx/mlx/backend/no_gpu/allocator.cpp +134 -0
  495. data/mlx/mlx/backend/no_gpu/apple_memory.h +16 -0
  496. data/mlx/mlx/backend/no_gpu/device_info.cpp +22 -0
  497. data/mlx/mlx/backend/no_gpu/eval.cpp +24 -0
  498. data/mlx/mlx/backend/no_gpu/event.cpp +53 -0
  499. data/mlx/mlx/backend/no_gpu/fence.cpp +54 -0
  500. data/mlx/mlx/backend/no_gpu/linux_memory.h +22 -0
  501. data/mlx/mlx/backend/no_gpu/primitives.cpp +185 -0
  502. data/mlx/mlx/compile.cpp +1243 -0
  503. data/mlx/mlx/compile.h +45 -0
  504. data/mlx/mlx/compile_impl.h +70 -0
  505. data/mlx/mlx/device.cpp +72 -0
  506. data/mlx/mlx/device.h +56 -0
  507. data/mlx/mlx/distributed/CMakeLists.txt +14 -0
  508. data/mlx/mlx/distributed/distributed.cpp +197 -0
  509. data/mlx/mlx/distributed/distributed.h +61 -0
  510. data/mlx/mlx/distributed/distributed_impl.h +59 -0
  511. data/mlx/mlx/distributed/jaccl/CMakeLists.txt +12 -0
  512. data/mlx/mlx/distributed/jaccl/jaccl.cpp +178 -0
  513. data/mlx/mlx/distributed/jaccl/jaccl.h +12 -0
  514. data/mlx/mlx/distributed/jaccl/mesh.cpp +451 -0
  515. data/mlx/mlx/distributed/jaccl/mesh.h +122 -0
  516. data/mlx/mlx/distributed/jaccl/no_jaccl.cpp +20 -0
  517. data/mlx/mlx/distributed/jaccl/ring.cpp +692 -0
  518. data/mlx/mlx/distributed/jaccl/ring.h +178 -0
  519. data/mlx/mlx/distributed/jaccl/utils.cpp +329 -0
  520. data/mlx/mlx/distributed/jaccl/utils.h +342 -0
  521. data/mlx/mlx/distributed/mpi/CMakeLists.txt +5 -0
  522. data/mlx/mlx/distributed/mpi/mpi.cpp +501 -0
  523. data/mlx/mlx/distributed/mpi/mpi.h +12 -0
  524. data/mlx/mlx/distributed/mpi/mpi_declarations.h +28 -0
  525. data/mlx/mlx/distributed/mpi/no_mpi.cpp +20 -0
  526. data/mlx/mlx/distributed/nccl/CMakeLists.txt +26 -0
  527. data/mlx/mlx/distributed/nccl/nccl.cpp +443 -0
  528. data/mlx/mlx/distributed/nccl/nccl.h +12 -0
  529. data/mlx/mlx/distributed/nccl/nccl_stub/CMakeLists.txt +1 -0
  530. data/mlx/mlx/distributed/nccl/nccl_stub/nccl_stubs.cpp +54 -0
  531. data/mlx/mlx/distributed/nccl/no_nccl.cpp +20 -0
  532. data/mlx/mlx/distributed/ops.cpp +186 -0
  533. data/mlx/mlx/distributed/ops.h +57 -0
  534. data/mlx/mlx/distributed/primitives.cpp +95 -0
  535. data/mlx/mlx/distributed/primitives.h +156 -0
  536. data/mlx/mlx/distributed/reduction_ops.h +38 -0
  537. data/mlx/mlx/distributed/ring/CMakeLists.txt +5 -0
  538. data/mlx/mlx/distributed/ring/no_ring.cpp +20 -0
  539. data/mlx/mlx/distributed/ring/ring.cpp +870 -0
  540. data/mlx/mlx/distributed/ring/ring.h +12 -0
  541. data/mlx/mlx/distributed/utils.cpp +206 -0
  542. data/mlx/mlx/distributed/utils.h +67 -0
  543. data/mlx/mlx/dtype.cpp +197 -0
  544. data/mlx/mlx/dtype.h +116 -0
  545. data/mlx/mlx/dtype_utils.cpp +42 -0
  546. data/mlx/mlx/dtype_utils.h +119 -0
  547. data/mlx/mlx/einsum.cpp +941 -0
  548. data/mlx/mlx/einsum.h +23 -0
  549. data/mlx/mlx/event.h +58 -0
  550. data/mlx/mlx/export.cpp +1130 -0
  551. data/mlx/mlx/export.h +137 -0
  552. data/mlx/mlx/export_impl.h +99 -0
  553. data/mlx/mlx/fast.cpp +941 -0
  554. data/mlx/mlx/fast.h +103 -0
  555. data/mlx/mlx/fast_primitives.h +427 -0
  556. data/mlx/mlx/fence.h +39 -0
  557. data/mlx/mlx/fft.cpp +262 -0
  558. data/mlx/mlx/fft.h +159 -0
  559. data/mlx/mlx/graph_utils.cpp +175 -0
  560. data/mlx/mlx/graph_utils.h +67 -0
  561. data/mlx/mlx/io/CMakeLists.txt +25 -0
  562. data/mlx/mlx/io/gguf.cpp +470 -0
  563. data/mlx/mlx/io/gguf.h +20 -0
  564. data/mlx/mlx/io/gguf_quants.cpp +164 -0
  565. data/mlx/mlx/io/load.cpp +397 -0
  566. data/mlx/mlx/io/load.h +175 -0
  567. data/mlx/mlx/io/no_gguf.cpp +20 -0
  568. data/mlx/mlx/io/no_safetensors.cpp +37 -0
  569. data/mlx/mlx/io/safetensors.cpp +234 -0
  570. data/mlx/mlx/io.h +61 -0
  571. data/mlx/mlx/linalg.cpp +708 -0
  572. data/mlx/mlx/linalg.h +115 -0
  573. data/mlx/mlx/memory.h +80 -0
  574. data/mlx/mlx/mlx.h +25 -0
  575. data/mlx/mlx/ops.cpp +6094 -0
  576. data/mlx/mlx/ops.h +1610 -0
  577. data/mlx/mlx/primitives.cpp +5850 -0
  578. data/mlx/mlx/primitives.h +2525 -0
  579. data/mlx/mlx/random.cpp +492 -0
  580. data/mlx/mlx/random.h +283 -0
  581. data/mlx/mlx/scheduler.cpp +73 -0
  582. data/mlx/mlx/scheduler.h +189 -0
  583. data/mlx/mlx/small_vector.h +540 -0
  584. data/mlx/mlx/stream.h +42 -0
  585. data/mlx/mlx/threadpool.h +133 -0
  586. data/mlx/mlx/transforms.cpp +1065 -0
  587. data/mlx/mlx/transforms.h +231 -0
  588. data/mlx/mlx/transforms_impl.h +88 -0
  589. data/mlx/mlx/types/bf16.h +187 -0
  590. data/mlx/mlx/types/complex.h +113 -0
  591. data/mlx/mlx/types/fp16.h +234 -0
  592. data/mlx/mlx/types/half_types.h +58 -0
  593. data/mlx/mlx/types/limits.h +70 -0
  594. data/mlx/mlx/utils.cpp +302 -0
  595. data/mlx/mlx/utils.h +174 -0
  596. data/mlx/mlx/version.cpp +11 -0
  597. data/mlx/mlx/version.h +22 -0
  598. data/mlx/mlx.pc.in +52 -0
  599. metadata +643 -0
metadata ADDED
@@ -0,0 +1,643 @@
1
+ --- !ruby/object:Gem::Specification
2
+ name: mlx
3
+ version: !ruby/object:Gem::Version
4
+ version: 0.30.7
5
+ platform: ruby
6
+ authors:
7
+ - MLX Contributors
8
+ - Aleksey Skryl
9
+ bindir: bin
10
+ cert_chain: []
11
+ date: 2026-02-13 00:00:00.000000000 Z
12
+ dependencies: []
13
+ description: A Ruby wrapper for the native MLX machine learning runtime.
14
+ email:
15
+ - mlx@group.apple.com
16
+ - aleksey.skryl@gmail.com
17
+ executables: []
18
+ extensions:
19
+ - ext/mlx/extconf.rb
20
+ extra_rdoc_files: []
21
+ files:
22
+ - ext/mlx/extconf.rb
23
+ - ext/mlx/native.cpp
24
+ - lib/mlx.rb
25
+ - lib/mlx/core.rb
26
+ - lib/mlx/distributed_utils/common.rb
27
+ - lib/mlx/distributed_utils/config.rb
28
+ - lib/mlx/distributed_utils/launch.rb
29
+ - lib/mlx/extension.rb
30
+ - lib/mlx/nn.rb
31
+ - lib/mlx/nn/base.rb
32
+ - lib/mlx/nn/init.rb
33
+ - lib/mlx/nn/layers.rb
34
+ - lib/mlx/nn/layers/activations.rb
35
+ - lib/mlx/nn/layers/base.rb
36
+ - lib/mlx/nn/layers/containers.rb
37
+ - lib/mlx/nn/layers/convolution.rb
38
+ - lib/mlx/nn/layers/convolution_transpose.rb
39
+ - lib/mlx/nn/layers/distributed.rb
40
+ - lib/mlx/nn/layers/dropout.rb
41
+ - lib/mlx/nn/layers/embedding.rb
42
+ - lib/mlx/nn/layers/linear.rb
43
+ - lib/mlx/nn/layers/normalization.rb
44
+ - lib/mlx/nn/layers/pooling.rb
45
+ - lib/mlx/nn/layers/positional_encoding.rb
46
+ - lib/mlx/nn/layers/quantized.rb
47
+ - lib/mlx/nn/layers/recurrent.rb
48
+ - lib/mlx/nn/layers/transformer.rb
49
+ - lib/mlx/nn/layers/upsample.rb
50
+ - lib/mlx/nn/losses.rb
51
+ - lib/mlx/nn/utils.rb
52
+ - lib/mlx/optimizers.rb
53
+ - lib/mlx/optimizers/optimizers.rb
54
+ - lib/mlx/optimizers/schedulers.rb
55
+ - lib/mlx/utils.rb
56
+ - lib/mlx/version.rb
57
+ - mlx/CMakeLists.txt
58
+ - mlx/cmake/FindCUDNN.cmake
59
+ - mlx/cmake/FindNCCL.cmake
60
+ - mlx/cmake/Findnvpl.cmake
61
+ - mlx/cmake/extension.cmake
62
+ - mlx/mlx.pc.in
63
+ - mlx/mlx/3rdparty/.clang-format
64
+ - mlx/mlx/3rdparty/pocketfft.h
65
+ - mlx/mlx/CMakeLists.txt
66
+ - mlx/mlx/allocator.h
67
+ - mlx/mlx/api.h
68
+ - mlx/mlx/array.cpp
69
+ - mlx/mlx/array.h
70
+ - mlx/mlx/backend/common/CMakeLists.txt
71
+ - mlx/mlx/backend/common/binary.h
72
+ - mlx/mlx/backend/common/broadcasting.cpp
73
+ - mlx/mlx/backend/common/broadcasting.h
74
+ - mlx/mlx/backend/common/buffer_cache.h
75
+ - mlx/mlx/backend/common/common.cpp
76
+ - mlx/mlx/backend/common/compiled.cpp
77
+ - mlx/mlx/backend/common/compiled.h
78
+ - mlx/mlx/backend/common/copy.h
79
+ - mlx/mlx/backend/common/hadamard.h
80
+ - mlx/mlx/backend/common/load.cpp
81
+ - mlx/mlx/backend/common/matmul.h
82
+ - mlx/mlx/backend/common/reduce.cpp
83
+ - mlx/mlx/backend/common/reduce.h
84
+ - mlx/mlx/backend/common/slicing.cpp
85
+ - mlx/mlx/backend/common/slicing.h
86
+ - mlx/mlx/backend/common/ternary.h
87
+ - mlx/mlx/backend/common/unary.h
88
+ - mlx/mlx/backend/common/utils.cpp
89
+ - mlx/mlx/backend/common/utils.h
90
+ - mlx/mlx/backend/cpu/CMakeLists.txt
91
+ - mlx/mlx/backend/cpu/arange.h
92
+ - mlx/mlx/backend/cpu/arg_reduce.cpp
93
+ - mlx/mlx/backend/cpu/binary.cpp
94
+ - mlx/mlx/backend/cpu/binary.h
95
+ - mlx/mlx/backend/cpu/binary_ops.h
96
+ - mlx/mlx/backend/cpu/binary_two.h
97
+ - mlx/mlx/backend/cpu/cholesky.cpp
98
+ - mlx/mlx/backend/cpu/compiled.cpp
99
+ - mlx/mlx/backend/cpu/compiled_preamble.h
100
+ - mlx/mlx/backend/cpu/conv.cpp
101
+ - mlx/mlx/backend/cpu/copy.cpp
102
+ - mlx/mlx/backend/cpu/copy.h
103
+ - mlx/mlx/backend/cpu/device_info.cpp
104
+ - mlx/mlx/backend/cpu/device_info.h
105
+ - mlx/mlx/backend/cpu/distributed.cpp
106
+ - mlx/mlx/backend/cpu/eig.cpp
107
+ - mlx/mlx/backend/cpu/eigh.cpp
108
+ - mlx/mlx/backend/cpu/encoder.cpp
109
+ - mlx/mlx/backend/cpu/encoder.h
110
+ - mlx/mlx/backend/cpu/eval.cpp
111
+ - mlx/mlx/backend/cpu/eval.h
112
+ - mlx/mlx/backend/cpu/fft.cpp
113
+ - mlx/mlx/backend/cpu/gemm.h
114
+ - mlx/mlx/backend/cpu/gemms/bnns.cpp
115
+ - mlx/mlx/backend/cpu/gemms/cblas.cpp
116
+ - mlx/mlx/backend/cpu/gemms/simd_bf16.cpp
117
+ - mlx/mlx/backend/cpu/gemms/simd_fp16.cpp
118
+ - mlx/mlx/backend/cpu/gemms/simd_gemm.h
119
+ - mlx/mlx/backend/cpu/hadamard.cpp
120
+ - mlx/mlx/backend/cpu/indexing.cpp
121
+ - mlx/mlx/backend/cpu/inverse.cpp
122
+ - mlx/mlx/backend/cpu/jit_compiler.cpp
123
+ - mlx/mlx/backend/cpu/jit_compiler.h
124
+ - mlx/mlx/backend/cpu/lapack.h
125
+ - mlx/mlx/backend/cpu/logsumexp.cpp
126
+ - mlx/mlx/backend/cpu/luf.cpp
127
+ - mlx/mlx/backend/cpu/make_compiled_preamble.ps1
128
+ - mlx/mlx/backend/cpu/make_compiled_preamble.sh
129
+ - mlx/mlx/backend/cpu/masked_mm.cpp
130
+ - mlx/mlx/backend/cpu/matmul.cpp
131
+ - mlx/mlx/backend/cpu/primitives.cpp
132
+ - mlx/mlx/backend/cpu/qrf.cpp
133
+ - mlx/mlx/backend/cpu/quantized.cpp
134
+ - mlx/mlx/backend/cpu/reduce.cpp
135
+ - mlx/mlx/backend/cpu/scan.cpp
136
+ - mlx/mlx/backend/cpu/select.cpp
137
+ - mlx/mlx/backend/cpu/simd/accelerate_fp16_simd.h
138
+ - mlx/mlx/backend/cpu/simd/accelerate_simd.h
139
+ - mlx/mlx/backend/cpu/simd/base_simd.h
140
+ - mlx/mlx/backend/cpu/simd/math.h
141
+ - mlx/mlx/backend/cpu/simd/neon_fp16_simd.h
142
+ - mlx/mlx/backend/cpu/simd/simd.h
143
+ - mlx/mlx/backend/cpu/simd/type.h
144
+ - mlx/mlx/backend/cpu/slicing.h
145
+ - mlx/mlx/backend/cpu/softmax.cpp
146
+ - mlx/mlx/backend/cpu/sort.cpp
147
+ - mlx/mlx/backend/cpu/svd.cpp
148
+ - mlx/mlx/backend/cpu/ternary.h
149
+ - mlx/mlx/backend/cpu/threefry.cpp
150
+ - mlx/mlx/backend/cpu/threefry.h
151
+ - mlx/mlx/backend/cpu/unary.cpp
152
+ - mlx/mlx/backend/cpu/unary.h
153
+ - mlx/mlx/backend/cpu/unary_ops.h
154
+ - mlx/mlx/backend/cuda/CMakeLists.txt
155
+ - mlx/mlx/backend/cuda/allocator.cpp
156
+ - mlx/mlx/backend/cuda/allocator.h
157
+ - mlx/mlx/backend/cuda/arange.cu
158
+ - mlx/mlx/backend/cuda/arg_reduce.cu
159
+ - mlx/mlx/backend/cuda/bin2h.cmake
160
+ - mlx/mlx/backend/cuda/binary/CMakeLists.txt
161
+ - mlx/mlx/backend/cuda/binary/add.cu
162
+ - mlx/mlx/backend/cuda/binary/arctan2.cu
163
+ - mlx/mlx/backend/cuda/binary/binary.cuh
164
+ - mlx/mlx/backend/cuda/binary/bitwise_binary.cu
165
+ - mlx/mlx/backend/cuda/binary/divide.cu
166
+ - mlx/mlx/backend/cuda/binary/equal.cu
167
+ - mlx/mlx/backend/cuda/binary/greater.cu
168
+ - mlx/mlx/backend/cuda/binary/greater_equal.cu
169
+ - mlx/mlx/backend/cuda/binary/less.cu
170
+ - mlx/mlx/backend/cuda/binary/less_equal.cu
171
+ - mlx/mlx/backend/cuda/binary/log_add_exp.cu
172
+ - mlx/mlx/backend/cuda/binary/logical_and.cu
173
+ - mlx/mlx/backend/cuda/binary/logical_or.cu
174
+ - mlx/mlx/backend/cuda/binary/maximum.cu
175
+ - mlx/mlx/backend/cuda/binary/minimum.cu
176
+ - mlx/mlx/backend/cuda/binary/multiply.cu
177
+ - mlx/mlx/backend/cuda/binary/not_equal.cu
178
+ - mlx/mlx/backend/cuda/binary/power.cu
179
+ - mlx/mlx/backend/cuda/binary/remainder.cu
180
+ - mlx/mlx/backend/cuda/binary/subtract.cu
181
+ - mlx/mlx/backend/cuda/binary_two.cu
182
+ - mlx/mlx/backend/cuda/compiled.cpp
183
+ - mlx/mlx/backend/cuda/conv.cpp
184
+ - mlx/mlx/backend/cuda/conv/conv.h
185
+ - mlx/mlx/backend/cuda/conv/gemm_conv.cu
186
+ - mlx/mlx/backend/cuda/conv/gemm_grouped_conv.cu
187
+ - mlx/mlx/backend/cuda/copy.cu
188
+ - mlx/mlx/backend/cuda/copy/copy.cuh
189
+ - mlx/mlx/backend/cuda/copy/copy_contiguous.cu
190
+ - mlx/mlx/backend/cuda/copy/copy_general.cu
191
+ - mlx/mlx/backend/cuda/copy/copy_general_dynamic.cu
192
+ - mlx/mlx/backend/cuda/copy/copy_general_input.cu
193
+ - mlx/mlx/backend/cuda/cublas_utils.cpp
194
+ - mlx/mlx/backend/cuda/cublas_utils.h
195
+ - mlx/mlx/backend/cuda/cuda.h
196
+ - mlx/mlx/backend/cuda/cuda_utils.h
197
+ - mlx/mlx/backend/cuda/cudnn_utils.cpp
198
+ - mlx/mlx/backend/cuda/cudnn_utils.h
199
+ - mlx/mlx/backend/cuda/custom_kernel.cpp
200
+ - mlx/mlx/backend/cuda/cutlass_utils.cuh
201
+ - mlx/mlx/backend/cuda/delayload.cpp
202
+ - mlx/mlx/backend/cuda/device.cpp
203
+ - mlx/mlx/backend/cuda/device.h
204
+ - mlx/mlx/backend/cuda/device/atomic_ops.cuh
205
+ - mlx/mlx/backend/cuda/device/binary_ops.cuh
206
+ - mlx/mlx/backend/cuda/device/cast_op.cuh
207
+ - mlx/mlx/backend/cuda/device/complex.cuh
208
+ - mlx/mlx/backend/cuda/device/config.h
209
+ - mlx/mlx/backend/cuda/device/fp16_math.cuh
210
+ - mlx/mlx/backend/cuda/device/gather.cuh
211
+ - mlx/mlx/backend/cuda/device/gather_axis.cuh
212
+ - mlx/mlx/backend/cuda/device/indexing.cuh
213
+ - mlx/mlx/backend/cuda/device/scatter.cuh
214
+ - mlx/mlx/backend/cuda/device/scatter_axis.cuh
215
+ - mlx/mlx/backend/cuda/device/scatter_ops.cuh
216
+ - mlx/mlx/backend/cuda/device/ternary_ops.cuh
217
+ - mlx/mlx/backend/cuda/device/unary_ops.cuh
218
+ - mlx/mlx/backend/cuda/device/utils.cuh
219
+ - mlx/mlx/backend/cuda/device_info.cpp
220
+ - mlx/mlx/backend/cuda/distributed.cu
221
+ - mlx/mlx/backend/cuda/eval.cpp
222
+ - mlx/mlx/backend/cuda/event.cu
223
+ - mlx/mlx/backend/cuda/event.h
224
+ - mlx/mlx/backend/cuda/fence.cpp
225
+ - mlx/mlx/backend/cuda/gemms/cublas_gemm.cpp
226
+ - mlx/mlx/backend/cuda/gemms/cublas_gemm.h
227
+ - mlx/mlx/backend/cuda/gemms/cublas_gemm_batched_12_0.cpp
228
+ - mlx/mlx/backend/cuda/gemms/cublas_gemm_batched_12_9.cu
229
+ - mlx/mlx/backend/cuda/gemms/gemv.cu
230
+ - mlx/mlx/backend/cuda/gemms/gemv.h
231
+ - mlx/mlx/backend/cuda/gemms/grouped_gemm.h
232
+ - mlx/mlx/backend/cuda/gemms/grouped_gemm_unaligned.cu
233
+ - mlx/mlx/backend/cuda/indexing.cpp
234
+ - mlx/mlx/backend/cuda/jit_module.cpp
235
+ - mlx/mlx/backend/cuda/jit_module.h
236
+ - mlx/mlx/backend/cuda/kernel_utils.cu
237
+ - mlx/mlx/backend/cuda/kernel_utils.cuh
238
+ - mlx/mlx/backend/cuda/layer_norm.cu
239
+ - mlx/mlx/backend/cuda/load.cpp
240
+ - mlx/mlx/backend/cuda/logsumexp.cu
241
+ - mlx/mlx/backend/cuda/lru_cache.h
242
+ - mlx/mlx/backend/cuda/matmul.cpp
243
+ - mlx/mlx/backend/cuda/no_cuda.cpp
244
+ - mlx/mlx/backend/cuda/primitives.cpp
245
+ - mlx/mlx/backend/cuda/quantized/affine_quantize.cu
246
+ - mlx/mlx/backend/cuda/quantized/convert_fp8.cu
247
+ - mlx/mlx/backend/cuda/quantized/cublas_qqmm.cpp
248
+ - mlx/mlx/backend/cuda/quantized/cublas_qqmm.h
249
+ - mlx/mlx/backend/cuda/quantized/cuda_fp4.h
250
+ - mlx/mlx/backend/cuda/quantized/fp_quantize.cu
251
+ - mlx/mlx/backend/cuda/quantized/mxfp8_quantize.cuh
252
+ - mlx/mlx/backend/cuda/quantized/no_qqmm_impl.cpp
253
+ - mlx/mlx/backend/cuda/quantized/nvfp4_quantize.cuh
254
+ - mlx/mlx/backend/cuda/quantized/qmv.cu
255
+ - mlx/mlx/backend/cuda/quantized/qmv.h
256
+ - mlx/mlx/backend/cuda/quantized/qqmm.cpp
257
+ - mlx/mlx/backend/cuda/quantized/qqmm_impl.cpp
258
+ - mlx/mlx/backend/cuda/quantized/qqmm_impl.h
259
+ - mlx/mlx/backend/cuda/quantized/qqmm_utils.cu
260
+ - mlx/mlx/backend/cuda/quantized/qqmm_utils.h
261
+ - mlx/mlx/backend/cuda/quantized/quantized.cpp
262
+ - mlx/mlx/backend/cuda/quantized/quantized.h
263
+ - mlx/mlx/backend/cuda/quantized/quantized_utils.cuh
264
+ - mlx/mlx/backend/cuda/quantized/quantized_utils.h
265
+ - mlx/mlx/backend/cuda/random.cu
266
+ - mlx/mlx/backend/cuda/reduce.cu
267
+ - mlx/mlx/backend/cuda/reduce/all_reduce.cu
268
+ - mlx/mlx/backend/cuda/reduce/col_reduce.cu
269
+ - mlx/mlx/backend/cuda/reduce/init_reduce.cu
270
+ - mlx/mlx/backend/cuda/reduce/reduce.cuh
271
+ - mlx/mlx/backend/cuda/reduce/reduce_ops.cuh
272
+ - mlx/mlx/backend/cuda/reduce/reduce_utils.cuh
273
+ - mlx/mlx/backend/cuda/reduce/row_reduce.cu
274
+ - mlx/mlx/backend/cuda/rms_norm.cu
275
+ - mlx/mlx/backend/cuda/rope.cu
276
+ - mlx/mlx/backend/cuda/scaled_dot_product_attention.cpp
277
+ - mlx/mlx/backend/cuda/scaled_dot_product_attention.cu
278
+ - mlx/mlx/backend/cuda/scan.cu
279
+ - mlx/mlx/backend/cuda/slicing.cpp
280
+ - mlx/mlx/backend/cuda/softmax.cu
281
+ - mlx/mlx/backend/cuda/sort.cu
282
+ - mlx/mlx/backend/cuda/steel/defines.cuh
283
+ - mlx/mlx/backend/cuda/steel/gemm.cuh
284
+ - mlx/mlx/backend/cuda/steel/mma.cuh
285
+ - mlx/mlx/backend/cuda/steel/tiles.cuh
286
+ - mlx/mlx/backend/cuda/steel/utils.cuh
287
+ - mlx/mlx/backend/cuda/ternary.cu
288
+ - mlx/mlx/backend/cuda/unary/CMakeLists.txt
289
+ - mlx/mlx/backend/cuda/unary/abs.cu
290
+ - mlx/mlx/backend/cuda/unary/arccos.cu
291
+ - mlx/mlx/backend/cuda/unary/arccosh.cu
292
+ - mlx/mlx/backend/cuda/unary/arcsin.cu
293
+ - mlx/mlx/backend/cuda/unary/arcsinh.cu
294
+ - mlx/mlx/backend/cuda/unary/arctan.cu
295
+ - mlx/mlx/backend/cuda/unary/arctanh.cu
296
+ - mlx/mlx/backend/cuda/unary/bitwise_invert.cu
297
+ - mlx/mlx/backend/cuda/unary/ceil.cu
298
+ - mlx/mlx/backend/cuda/unary/conjugate.cu
299
+ - mlx/mlx/backend/cuda/unary/cos.cu
300
+ - mlx/mlx/backend/cuda/unary/cosh.cu
301
+ - mlx/mlx/backend/cuda/unary/erf.cu
302
+ - mlx/mlx/backend/cuda/unary/erf_inv.cu
303
+ - mlx/mlx/backend/cuda/unary/exp.cu
304
+ - mlx/mlx/backend/cuda/unary/expm1.cu
305
+ - mlx/mlx/backend/cuda/unary/floor.cu
306
+ - mlx/mlx/backend/cuda/unary/imag.cu
307
+ - mlx/mlx/backend/cuda/unary/log.cu
308
+ - mlx/mlx/backend/cuda/unary/log1p.cu
309
+ - mlx/mlx/backend/cuda/unary/logical_not.cu
310
+ - mlx/mlx/backend/cuda/unary/negative.cu
311
+ - mlx/mlx/backend/cuda/unary/real.cu
312
+ - mlx/mlx/backend/cuda/unary/round.cu
313
+ - mlx/mlx/backend/cuda/unary/sigmoid.cu
314
+ - mlx/mlx/backend/cuda/unary/sign.cu
315
+ - mlx/mlx/backend/cuda/unary/sin.cu
316
+ - mlx/mlx/backend/cuda/unary/sinh.cu
317
+ - mlx/mlx/backend/cuda/unary/sqrt.cu
318
+ - mlx/mlx/backend/cuda/unary/square.cu
319
+ - mlx/mlx/backend/cuda/unary/tan.cu
320
+ - mlx/mlx/backend/cuda/unary/tanh.cu
321
+ - mlx/mlx/backend/cuda/unary/unary.cuh
322
+ - mlx/mlx/backend/cuda/utils.cpp
323
+ - mlx/mlx/backend/cuda/utils.h
324
+ - mlx/mlx/backend/cuda/vector_types.cuh
325
+ - mlx/mlx/backend/cuda/worker.cpp
326
+ - mlx/mlx/backend/cuda/worker.h
327
+ - mlx/mlx/backend/gpu/CMakeLists.txt
328
+ - mlx/mlx/backend/gpu/copy.cpp
329
+ - mlx/mlx/backend/gpu/copy.h
330
+ - mlx/mlx/backend/gpu/device_info.h
331
+ - mlx/mlx/backend/gpu/eval.h
332
+ - mlx/mlx/backend/gpu/primitives.cpp
333
+ - mlx/mlx/backend/gpu/slicing.cpp
334
+ - mlx/mlx/backend/gpu/slicing.h
335
+ - mlx/mlx/backend/metal/CMakeLists.txt
336
+ - mlx/mlx/backend/metal/allocator.cpp
337
+ - mlx/mlx/backend/metal/allocator.h
338
+ - mlx/mlx/backend/metal/binary.cpp
339
+ - mlx/mlx/backend/metal/binary.h
340
+ - mlx/mlx/backend/metal/compiled.cpp
341
+ - mlx/mlx/backend/metal/conv.cpp
342
+ - mlx/mlx/backend/metal/copy.cpp
343
+ - mlx/mlx/backend/metal/custom_kernel.cpp
344
+ - mlx/mlx/backend/metal/device.cpp
345
+ - mlx/mlx/backend/metal/device.h
346
+ - mlx/mlx/backend/metal/device_info.cpp
347
+ - mlx/mlx/backend/metal/distributed.cpp
348
+ - mlx/mlx/backend/metal/eval.cpp
349
+ - mlx/mlx/backend/metal/event.cpp
350
+ - mlx/mlx/backend/metal/fence.cpp
351
+ - mlx/mlx/backend/metal/fft.cpp
352
+ - mlx/mlx/backend/metal/hadamard.cpp
353
+ - mlx/mlx/backend/metal/indexing.cpp
354
+ - mlx/mlx/backend/metal/jit/includes.h
355
+ - mlx/mlx/backend/metal/jit/indexing.h
356
+ - mlx/mlx/backend/metal/jit_kernels.cpp
357
+ - mlx/mlx/backend/metal/kernels.h
358
+ - mlx/mlx/backend/metal/kernels/CMakeLists.txt
359
+ - mlx/mlx/backend/metal/kernels/arange.h
360
+ - mlx/mlx/backend/metal/kernels/arange.metal
361
+ - mlx/mlx/backend/metal/kernels/arg_reduce.metal
362
+ - mlx/mlx/backend/metal/kernels/atomic.h
363
+ - mlx/mlx/backend/metal/kernels/bf16.h
364
+ - mlx/mlx/backend/metal/kernels/bf16_math.h
365
+ - mlx/mlx/backend/metal/kernels/binary.h
366
+ - mlx/mlx/backend/metal/kernels/binary.metal
367
+ - mlx/mlx/backend/metal/kernels/binary_ops.h
368
+ - mlx/mlx/backend/metal/kernels/binary_two.h
369
+ - mlx/mlx/backend/metal/kernels/binary_two.metal
370
+ - mlx/mlx/backend/metal/kernels/cexpf.h
371
+ - mlx/mlx/backend/metal/kernels/complex.h
372
+ - mlx/mlx/backend/metal/kernels/conv.metal
373
+ - mlx/mlx/backend/metal/kernels/copy.h
374
+ - mlx/mlx/backend/metal/kernels/copy.metal
375
+ - mlx/mlx/backend/metal/kernels/defines.h
376
+ - mlx/mlx/backend/metal/kernels/erf.h
377
+ - mlx/mlx/backend/metal/kernels/expm1f.h
378
+ - mlx/mlx/backend/metal/kernels/fence.metal
379
+ - mlx/mlx/backend/metal/kernels/fft.h
380
+ - mlx/mlx/backend/metal/kernels/fft.metal
381
+ - mlx/mlx/backend/metal/kernels/fft/radix.h
382
+ - mlx/mlx/backend/metal/kernels/fft/readwrite.h
383
+ - mlx/mlx/backend/metal/kernels/fp4.h
384
+ - mlx/mlx/backend/metal/kernels/fp8.h
385
+ - mlx/mlx/backend/metal/kernels/fp_quantized.h
386
+ - mlx/mlx/backend/metal/kernels/fp_quantized.metal
387
+ - mlx/mlx/backend/metal/kernels/fp_quantized_nax.h
388
+ - mlx/mlx/backend/metal/kernels/fp_quantized_nax.metal
389
+ - mlx/mlx/backend/metal/kernels/gemv.metal
390
+ - mlx/mlx/backend/metal/kernels/gemv_masked.h
391
+ - mlx/mlx/backend/metal/kernels/gemv_masked.metal
392
+ - mlx/mlx/backend/metal/kernels/hadamard.h
393
+ - mlx/mlx/backend/metal/kernels/indexing/gather.h
394
+ - mlx/mlx/backend/metal/kernels/indexing/gather_axis.h
395
+ - mlx/mlx/backend/metal/kernels/indexing/gather_front.h
396
+ - mlx/mlx/backend/metal/kernels/indexing/indexing.h
397
+ - mlx/mlx/backend/metal/kernels/indexing/masked_scatter.h
398
+ - mlx/mlx/backend/metal/kernels/indexing/scatter.h
399
+ - mlx/mlx/backend/metal/kernels/indexing/scatter_axis.h
400
+ - mlx/mlx/backend/metal/kernels/layer_norm.metal
401
+ - mlx/mlx/backend/metal/kernels/logging.h
402
+ - mlx/mlx/backend/metal/kernels/logsumexp.h
403
+ - mlx/mlx/backend/metal/kernels/logsumexp.metal
404
+ - mlx/mlx/backend/metal/kernels/quantized.h
405
+ - mlx/mlx/backend/metal/kernels/quantized.metal
406
+ - mlx/mlx/backend/metal/kernels/quantized_nax.h
407
+ - mlx/mlx/backend/metal/kernels/quantized_nax.metal
408
+ - mlx/mlx/backend/metal/kernels/quantized_utils.h
409
+ - mlx/mlx/backend/metal/kernels/random.metal
410
+ - mlx/mlx/backend/metal/kernels/reduce.h
411
+ - mlx/mlx/backend/metal/kernels/reduce.metal
412
+ - mlx/mlx/backend/metal/kernels/reduce_utils.h
413
+ - mlx/mlx/backend/metal/kernels/reduction/ops.h
414
+ - mlx/mlx/backend/metal/kernels/reduction/reduce_all.h
415
+ - mlx/mlx/backend/metal/kernels/reduction/reduce_col.h
416
+ - mlx/mlx/backend/metal/kernels/reduction/reduce_init.h
417
+ - mlx/mlx/backend/metal/kernels/reduction/reduce_row.h
418
+ - mlx/mlx/backend/metal/kernels/rms_norm.metal
419
+ - mlx/mlx/backend/metal/kernels/rope.metal
420
+ - mlx/mlx/backend/metal/kernels/scaled_dot_product_attention.metal
421
+ - mlx/mlx/backend/metal/kernels/scan.h
422
+ - mlx/mlx/backend/metal/kernels/scan.metal
423
+ - mlx/mlx/backend/metal/kernels/sdpa_vector.h
424
+ - mlx/mlx/backend/metal/kernels/softmax.h
425
+ - mlx/mlx/backend/metal/kernels/softmax.metal
426
+ - mlx/mlx/backend/metal/kernels/sort.h
427
+ - mlx/mlx/backend/metal/kernels/sort.metal
428
+ - mlx/mlx/backend/metal/kernels/steel/attn/attn.h
429
+ - mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention.h
430
+ - mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention.metal
431
+ - mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention_nax.h
432
+ - mlx/mlx/backend/metal/kernels/steel/attn/kernels/steel_attention_nax.metal
433
+ - mlx/mlx/backend/metal/kernels/steel/attn/loader.h
434
+ - mlx/mlx/backend/metal/kernels/steel/attn/mma.h
435
+ - mlx/mlx/backend/metal/kernels/steel/attn/nax.h
436
+ - mlx/mlx/backend/metal/kernels/steel/attn/params.h
437
+ - mlx/mlx/backend/metal/kernels/steel/attn/transforms.h
438
+ - mlx/mlx/backend/metal/kernels/steel/conv/conv.h
439
+ - mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv.h
440
+ - mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv.metal
441
+ - mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv_general.h
442
+ - mlx/mlx/backend/metal/kernels/steel/conv/kernels/steel_conv_general.metal
443
+ - mlx/mlx/backend/metal/kernels/steel/conv/loader.h
444
+ - mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_l.h
445
+ - mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_n.h
446
+ - mlx/mlx/backend/metal/kernels/steel/conv/loaders/loader_general.h
447
+ - mlx/mlx/backend/metal/kernels/steel/conv/params.h
448
+ - mlx/mlx/backend/metal/kernels/steel/defines.h
449
+ - mlx/mlx/backend/metal/kernels/steel/gemm/gemm.h
450
+ - mlx/mlx/backend/metal/kernels/steel/gemm/gemm_nax.h
451
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused.h
452
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused.metal
453
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused_nax.h
454
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_fused_nax.metal
455
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather.h
456
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather.metal
457
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather_nax.h
458
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_gather_nax.metal
459
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_masked.h
460
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_masked.metal
461
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_segmented.h
462
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_segmented.metal
463
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk.h
464
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk.metal
465
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk_nax.h
466
+ - mlx/mlx/backend/metal/kernels/steel/gemm/kernels/steel_gemm_splitk_nax.metal
467
+ - mlx/mlx/backend/metal/kernels/steel/gemm/loader.h
468
+ - mlx/mlx/backend/metal/kernels/steel/gemm/mma.h
469
+ - mlx/mlx/backend/metal/kernels/steel/gemm/nax.h
470
+ - mlx/mlx/backend/metal/kernels/steel/gemm/params.h
471
+ - mlx/mlx/backend/metal/kernels/steel/gemm/transforms.h
472
+ - mlx/mlx/backend/metal/kernels/steel/utils.h
473
+ - mlx/mlx/backend/metal/kernels/steel/utils/integral_constant.h
474
+ - mlx/mlx/backend/metal/kernels/steel/utils/type_traits.h
475
+ - mlx/mlx/backend/metal/kernels/ternary.h
476
+ - mlx/mlx/backend/metal/kernels/ternary.metal
477
+ - mlx/mlx/backend/metal/kernels/ternary_ops.h
478
+ - mlx/mlx/backend/metal/kernels/unary.h
479
+ - mlx/mlx/backend/metal/kernels/unary.metal
480
+ - mlx/mlx/backend/metal/kernels/unary_ops.h
481
+ - mlx/mlx/backend/metal/kernels/utils.h
482
+ - mlx/mlx/backend/metal/logsumexp.cpp
483
+ - mlx/mlx/backend/metal/make_compiled_preamble.sh
484
+ - mlx/mlx/backend/metal/matmul.cpp
485
+ - mlx/mlx/backend/metal/matmul.h
486
+ - mlx/mlx/backend/metal/metal.cpp
487
+ - mlx/mlx/backend/metal/metal.h
488
+ - mlx/mlx/backend/metal/no_metal.cpp
489
+ - mlx/mlx/backend/metal/nojit_kernels.cpp
490
+ - mlx/mlx/backend/metal/normalization.cpp
491
+ - mlx/mlx/backend/metal/primitives.cpp
492
+ - mlx/mlx/backend/metal/quantized.cpp
493
+ - mlx/mlx/backend/metal/reduce.cpp
494
+ - mlx/mlx/backend/metal/reduce.h
495
+ - mlx/mlx/backend/metal/resident.cpp
496
+ - mlx/mlx/backend/metal/resident.h
497
+ - mlx/mlx/backend/metal/rope.cpp
498
+ - mlx/mlx/backend/metal/scaled_dot_product_attention.cpp
499
+ - mlx/mlx/backend/metal/scan.cpp
500
+ - mlx/mlx/backend/metal/scan.h
501
+ - mlx/mlx/backend/metal/slicing.cpp
502
+ - mlx/mlx/backend/metal/softmax.cpp
503
+ - mlx/mlx/backend/metal/sort.cpp
504
+ - mlx/mlx/backend/metal/ternary.cpp
505
+ - mlx/mlx/backend/metal/ternary.h
506
+ - mlx/mlx/backend/metal/unary.cpp
507
+ - mlx/mlx/backend/metal/unary.h
508
+ - mlx/mlx/backend/metal/utils.cpp
509
+ - mlx/mlx/backend/metal/utils.h
510
+ - mlx/mlx/backend/no_cpu/CMakeLists.txt
511
+ - mlx/mlx/backend/no_cpu/compiled.cpp
512
+ - mlx/mlx/backend/no_cpu/device_info.cpp
513
+ - mlx/mlx/backend/no_cpu/primitives.cpp
514
+ - mlx/mlx/backend/no_gpu/CMakeLists.txt
515
+ - mlx/mlx/backend/no_gpu/allocator.cpp
516
+ - mlx/mlx/backend/no_gpu/apple_memory.h
517
+ - mlx/mlx/backend/no_gpu/device_info.cpp
518
+ - mlx/mlx/backend/no_gpu/eval.cpp
519
+ - mlx/mlx/backend/no_gpu/event.cpp
520
+ - mlx/mlx/backend/no_gpu/fence.cpp
521
+ - mlx/mlx/backend/no_gpu/linux_memory.h
522
+ - mlx/mlx/backend/no_gpu/primitives.cpp
523
+ - mlx/mlx/compile.cpp
524
+ - mlx/mlx/compile.h
525
+ - mlx/mlx/compile_impl.h
526
+ - mlx/mlx/device.cpp
527
+ - mlx/mlx/device.h
528
+ - mlx/mlx/distributed/CMakeLists.txt
529
+ - mlx/mlx/distributed/distributed.cpp
530
+ - mlx/mlx/distributed/distributed.h
531
+ - mlx/mlx/distributed/distributed_impl.h
532
+ - mlx/mlx/distributed/jaccl/CMakeLists.txt
533
+ - mlx/mlx/distributed/jaccl/jaccl.cpp
534
+ - mlx/mlx/distributed/jaccl/jaccl.h
535
+ - mlx/mlx/distributed/jaccl/mesh.cpp
536
+ - mlx/mlx/distributed/jaccl/mesh.h
537
+ - mlx/mlx/distributed/jaccl/no_jaccl.cpp
538
+ - mlx/mlx/distributed/jaccl/ring.cpp
539
+ - mlx/mlx/distributed/jaccl/ring.h
540
+ - mlx/mlx/distributed/jaccl/utils.cpp
541
+ - mlx/mlx/distributed/jaccl/utils.h
542
+ - mlx/mlx/distributed/mpi/CMakeLists.txt
543
+ - mlx/mlx/distributed/mpi/mpi.cpp
544
+ - mlx/mlx/distributed/mpi/mpi.h
545
+ - mlx/mlx/distributed/mpi/mpi_declarations.h
546
+ - mlx/mlx/distributed/mpi/no_mpi.cpp
547
+ - mlx/mlx/distributed/nccl/CMakeLists.txt
548
+ - mlx/mlx/distributed/nccl/nccl.cpp
549
+ - mlx/mlx/distributed/nccl/nccl.h
550
+ - mlx/mlx/distributed/nccl/nccl_stub/CMakeLists.txt
551
+ - mlx/mlx/distributed/nccl/nccl_stub/nccl_stubs.cpp
552
+ - mlx/mlx/distributed/nccl/no_nccl.cpp
553
+ - mlx/mlx/distributed/ops.cpp
554
+ - mlx/mlx/distributed/ops.h
555
+ - mlx/mlx/distributed/primitives.cpp
556
+ - mlx/mlx/distributed/primitives.h
557
+ - mlx/mlx/distributed/reduction_ops.h
558
+ - mlx/mlx/distributed/ring/CMakeLists.txt
559
+ - mlx/mlx/distributed/ring/no_ring.cpp
560
+ - mlx/mlx/distributed/ring/ring.cpp
561
+ - mlx/mlx/distributed/ring/ring.h
562
+ - mlx/mlx/distributed/utils.cpp
563
+ - mlx/mlx/distributed/utils.h
564
+ - mlx/mlx/dtype.cpp
565
+ - mlx/mlx/dtype.h
566
+ - mlx/mlx/dtype_utils.cpp
567
+ - mlx/mlx/dtype_utils.h
568
+ - mlx/mlx/einsum.cpp
569
+ - mlx/mlx/einsum.h
570
+ - mlx/mlx/event.h
571
+ - mlx/mlx/export.cpp
572
+ - mlx/mlx/export.h
573
+ - mlx/mlx/export_impl.h
574
+ - mlx/mlx/fast.cpp
575
+ - mlx/mlx/fast.h
576
+ - mlx/mlx/fast_primitives.h
577
+ - mlx/mlx/fence.h
578
+ - mlx/mlx/fft.cpp
579
+ - mlx/mlx/fft.h
580
+ - mlx/mlx/graph_utils.cpp
581
+ - mlx/mlx/graph_utils.h
582
+ - mlx/mlx/io.h
583
+ - mlx/mlx/io/CMakeLists.txt
584
+ - mlx/mlx/io/gguf.cpp
585
+ - mlx/mlx/io/gguf.h
586
+ - mlx/mlx/io/gguf_quants.cpp
587
+ - mlx/mlx/io/load.cpp
588
+ - mlx/mlx/io/load.h
589
+ - mlx/mlx/io/no_gguf.cpp
590
+ - mlx/mlx/io/no_safetensors.cpp
591
+ - mlx/mlx/io/safetensors.cpp
592
+ - mlx/mlx/linalg.cpp
593
+ - mlx/mlx/linalg.h
594
+ - mlx/mlx/memory.h
595
+ - mlx/mlx/mlx.h
596
+ - mlx/mlx/ops.cpp
597
+ - mlx/mlx/ops.h
598
+ - mlx/mlx/primitives.cpp
599
+ - mlx/mlx/primitives.h
600
+ - mlx/mlx/random.cpp
601
+ - mlx/mlx/random.h
602
+ - mlx/mlx/scheduler.cpp
603
+ - mlx/mlx/scheduler.h
604
+ - mlx/mlx/small_vector.h
605
+ - mlx/mlx/stream.h
606
+ - mlx/mlx/threadpool.h
607
+ - mlx/mlx/transforms.cpp
608
+ - mlx/mlx/transforms.h
609
+ - mlx/mlx/transforms_impl.h
610
+ - mlx/mlx/types/bf16.h
611
+ - mlx/mlx/types/complex.h
612
+ - mlx/mlx/types/fp16.h
613
+ - mlx/mlx/types/half_types.h
614
+ - mlx/mlx/types/limits.h
615
+ - mlx/mlx/utils.cpp
616
+ - mlx/mlx/utils.h
617
+ - mlx/mlx/version.cpp
618
+ - mlx/mlx/version.h
619
+ homepage: https://github.com/skryl/mlx-ruby
620
+ licenses:
621
+ - MIT
622
+ metadata:
623
+ homepage_uri: https://github.com/skryl/mlx-ruby
624
+ source_code_uri: https://github.com/skryl/mlx-ruby
625
+ github_uri: https://github.com/skryl
626
+ rdoc_options: []
627
+ require_paths:
628
+ - lib
629
+ required_ruby_version: !ruby/object:Gem::Requirement
630
+ requirements:
631
+ - - ">="
632
+ - !ruby/object:Gem::Version
633
+ version: '3.1'
634
+ required_rubygems_version: !ruby/object:Gem::Requirement
635
+ requirements:
636
+ - - ">="
637
+ - !ruby/object:Gem::Version
638
+ version: '0'
639
+ requirements: []
640
+ rubygems_version: 3.6.2
641
+ specification_version: 4
642
+ summary: Ruby bindings for the native MLX library
643
+ test_files: []