whispercpp 1.3.3 → 1.3.4

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 (586) hide show
  1. checksums.yaml +4 -4
  2. data/ext/ruby_whisper_params.c +55 -25
  3. data/ext/sources/CMakeLists.txt +1 -1
  4. data/ext/sources/bindings/javascript/package.json +1 -1
  5. data/ext/sources/build-xcframework.sh +24 -0
  6. data/ext/sources/examples/CMakeLists.txt +1 -0
  7. data/ext/sources/examples/addon.node/addon.cpp +19 -19
  8. data/ext/sources/examples/addon.node/index.js +7 -5
  9. data/ext/sources/examples/bench/bench.cpp +26 -16
  10. data/ext/sources/examples/bench.wasm/index-tmpl.html +10 -9
  11. data/ext/sources/examples/cli/cli.cpp +4 -2
  12. data/ext/sources/examples/command/command.cpp +26 -24
  13. data/ext/sources/examples/command.wasm/index-tmpl.html +5 -4
  14. data/ext/sources/examples/common-ggml.cpp +2 -0
  15. data/ext/sources/examples/lsp/lsp.cpp +19 -17
  16. data/ext/sources/examples/server/server.cpp +24 -13
  17. data/ext/sources/examples/server.py +6 -1
  18. data/ext/sources/examples/stream/stream.cpp +4 -2
  19. data/ext/sources/examples/stream.wasm/emscripten.cpp +6 -6
  20. data/ext/sources/examples/stream.wasm/index-tmpl.html +82 -5
  21. data/ext/sources/examples/talk-llama/CMakeLists.txt +2 -2
  22. data/ext/sources/examples/talk-llama/llama-adapter.cpp +101 -4
  23. data/ext/sources/examples/talk-llama/llama-adapter.h +6 -0
  24. data/ext/sources/examples/talk-llama/llama-arch.cpp +588 -15
  25. data/ext/sources/examples/talk-llama/llama-arch.h +58 -1
  26. data/ext/sources/examples/talk-llama/llama-batch.cpp +103 -71
  27. data/ext/sources/examples/talk-llama/llama-batch.h +31 -18
  28. data/ext/sources/examples/talk-llama/llama-chat.cpp +120 -5
  29. data/ext/sources/examples/talk-llama/llama-chat.h +7 -0
  30. data/ext/sources/examples/talk-llama/llama-context.cpp +460 -357
  31. data/ext/sources/examples/talk-llama/llama-context.h +44 -29
  32. data/ext/sources/examples/talk-llama/llama-cparams.h +4 -4
  33. data/ext/sources/examples/talk-llama/llama-graph.cpp +543 -271
  34. data/ext/sources/examples/talk-llama/llama-graph.h +278 -168
  35. data/ext/sources/examples/talk-llama/llama-hparams.cpp +118 -4
  36. data/ext/sources/examples/talk-llama/llama-hparams.h +61 -15
  37. data/ext/sources/examples/talk-llama/llama-impl.h +2 -0
  38. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +326 -0
  39. data/ext/sources/examples/talk-llama/{llama-kv-cache-unified-iswa.h → llama-kv-cache-iswa.h} +38 -29
  40. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +2020 -0
  41. data/ext/sources/examples/talk-llama/llama-kv-cache.h +358 -27
  42. data/ext/sources/examples/talk-llama/llama-kv-cells.h +80 -28
  43. data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +56 -36
  44. data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +30 -29
  45. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +48 -19
  46. data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +13 -14
  47. data/ext/sources/examples/talk-llama/llama-memory.h +16 -10
  48. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +2 -0
  49. data/ext/sources/examples/talk-llama/llama-model-loader.h +3 -2
  50. data/ext/sources/examples/talk-llama/llama-model.cpp +7165 -2336
  51. data/ext/sources/examples/talk-llama/llama-model.h +60 -9
  52. data/ext/sources/examples/talk-llama/llama-quant.cpp +48 -10
  53. data/ext/sources/examples/talk-llama/llama-sampling.cpp +226 -126
  54. data/ext/sources/examples/talk-llama/llama-vocab.cpp +440 -13
  55. data/ext/sources/examples/talk-llama/llama-vocab.h +45 -0
  56. data/ext/sources/examples/talk-llama/llama.cpp +65 -10
  57. data/ext/sources/examples/talk-llama/llama.h +95 -177
  58. data/ext/sources/examples/talk-llama/talk-llama.cpp +9 -6
  59. data/ext/sources/examples/talk-llama/unicode.cpp +207 -0
  60. data/ext/sources/examples/talk-llama/unicode.h +45 -0
  61. data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +4 -2
  62. data/ext/sources/examples/whisper.wasm/index-tmpl.html +17 -16
  63. data/ext/sources/ggml/CMakeLists.txt +59 -31
  64. data/ext/sources/ggml/cmake/ggml-config.cmake.in +132 -93
  65. data/ext/sources/ggml/include/ggml-backend.h +17 -1
  66. data/ext/sources/ggml/include/ggml-cpu.h +1 -1
  67. data/ext/sources/ggml/include/ggml-metal.h +1 -6
  68. data/ext/sources/ggml/include/ggml-opt.h +25 -6
  69. data/ext/sources/ggml/include/ggml-webgpu.h +19 -0
  70. data/ext/sources/ggml/include/ggml-zdnn.h +17 -0
  71. data/ext/sources/ggml/include/ggml.h +221 -16
  72. data/ext/sources/ggml/src/CMakeLists.txt +17 -2
  73. data/ext/sources/ggml/src/ggml-alloc.c +265 -141
  74. data/ext/sources/ggml/src/ggml-backend-impl.h +4 -1
  75. data/ext/sources/ggml/src/ggml-backend-reg.cpp +30 -13
  76. data/ext/sources/ggml/src/ggml-backend.cpp +221 -38
  77. data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +1 -1
  78. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +5 -4
  79. data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +14 -0
  80. data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +3 -1
  81. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +903 -717
  82. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +143 -25
  83. data/ext/sources/ggml/src/ggml-cann/common.h +143 -1
  84. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +488 -69
  85. data/ext/sources/ggml/src/ggml-common.h +17 -0
  86. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +40 -18
  87. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +4 -2
  88. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +132 -596
  89. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +14 -286
  90. data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +103 -582
  91. data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +162 -589
  92. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +265 -437
  93. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +3 -58
  94. data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +521 -353
  95. data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +54 -314
  96. data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +184 -675
  97. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +4679 -1657
  98. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +32 -2
  99. data/ext/sources/ggml/src/ggml-cpu/common.h +14 -0
  100. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +13 -6
  101. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +70 -42
  102. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +35 -28
  103. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +152 -18
  104. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +7 -1
  105. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +227 -97
  106. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +474 -1116
  107. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +1587 -1177
  108. data/ext/sources/ggml/src/ggml-cpu/ops.h +5 -8
  109. data/ext/sources/ggml/src/ggml-cpu/quants.c +35 -0
  110. data/ext/sources/ggml/src/ggml-cpu/quants.h +8 -0
  111. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +458 -47
  112. data/ext/sources/ggml/src/ggml-cpu/repack.h +22 -0
  113. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +89 -60
  114. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1024 -0
  115. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
  116. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
  117. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
  118. data/ext/sources/ggml/src/ggml-cpu/traits.cpp +2 -2
  119. data/ext/sources/ggml/src/ggml-cpu/traits.h +1 -1
  120. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +170 -26
  121. data/ext/sources/ggml/src/ggml-cpu/vec.h +506 -63
  122. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +20 -16
  123. data/ext/sources/ggml/src/ggml-cuda/add-id.cu +58 -0
  124. data/ext/sources/ggml/src/ggml-cuda/add-id.cuh +3 -0
  125. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +330 -191
  126. data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +2 -0
  127. data/ext/sources/ggml/src/ggml-cuda/common.cuh +250 -63
  128. data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cu +1 -4
  129. data/ext/sources/ggml/src/ggml-cuda/conv2d.cu +166 -0
  130. data/ext/sources/ggml/src/ggml-cuda/conv2d.cuh +5 -0
  131. data/ext/sources/ggml/src/ggml-cuda/convert.cu +95 -22
  132. data/ext/sources/ggml/src/ggml-cuda/convert.cuh +15 -0
  133. data/ext/sources/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
  134. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +64 -307
  135. data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cu +2 -14
  136. data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +14 -40
  137. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +498 -367
  138. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +137 -91
  139. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +755 -0
  140. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +3 -0
  141. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +593 -0
  142. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +86 -50
  143. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +185 -198
  144. data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
  145. data/ext/sources/ggml/src/ggml-cuda/getrows.cu +50 -39
  146. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +379 -107
  147. data/ext/sources/ggml/src/ggml-cuda/im2col.cu +196 -35
  148. data/ext/sources/ggml/src/ggml-cuda/im2col.cuh +1 -0
  149. data/ext/sources/ggml/src/ggml-cuda/mean.cu +56 -2
  150. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +198 -45
  151. data/ext/sources/ggml/src/ggml-cuda/mmf.cu +123 -0
  152. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +496 -0
  153. data/ext/sources/ggml/src/ggml-cuda/mmq.cu +206 -57
  154. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +1262 -721
  155. data/ext/sources/ggml/src/ggml-cuda/{mmv.cu → mmvf.cu} +53 -53
  156. data/ext/sources/ggml/src/ggml-cuda/{mmv.cuh → mmvf.cuh} +3 -3
  157. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +64 -73
  158. data/ext/sources/ggml/src/ggml-cuda/norm.cu +284 -12
  159. data/ext/sources/ggml/src/ggml-cuda/norm.cuh +7 -0
  160. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
  161. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
  162. data/ext/sources/ggml/src/ggml-cuda/pad.cu +46 -23
  163. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
  164. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
  165. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +12 -10
  166. data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +53 -0
  167. data/ext/sources/ggml/src/ggml-cuda/roll.cu +67 -0
  168. data/ext/sources/ggml/src/ggml-cuda/roll.cuh +5 -0
  169. data/ext/sources/ggml/src/ggml-cuda/rope.cu +21 -27
  170. data/ext/sources/ggml/src/ggml-cuda/scale.cu +14 -11
  171. data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +276 -0
  172. data/ext/sources/ggml/src/ggml-cuda/set-rows.cuh +7 -0
  173. data/ext/sources/ggml/src/ggml-cuda/softcap.cu +34 -0
  174. data/ext/sources/ggml/src/ggml-cuda/softcap.cuh +5 -0
  175. data/ext/sources/ggml/src/ggml-cuda/softmax.cu +126 -59
  176. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +10 -2
  177. data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +322 -100
  178. data/ext/sources/ggml/src/ggml-cuda/sum.cu +6 -10
  179. data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +21 -4
  180. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  181. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  182. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  183. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  184. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  185. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  186. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  187. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  188. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  189. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  190. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  191. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  192. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  193. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  194. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  195. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  196. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  197. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  198. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  199. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  200. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  201. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  202. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  203. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  204. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  205. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  206. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  207. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  208. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  209. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  210. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  211. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  212. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  213. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  214. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  215. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  216. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +21 -18
  217. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  218. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  219. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  220. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  221. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  222. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  223. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  224. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  225. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  226. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  227. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  228. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  229. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  230. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  231. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  232. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  233. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
  234. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +259 -0
  235. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +14 -0
  236. data/ext/sources/ggml/src/ggml-cuda/tsembd.cu +3 -3
  237. data/ext/sources/ggml/src/ggml-cuda/unary.cu +90 -0
  238. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +8 -0
  239. data/ext/sources/ggml/src/ggml-cuda/upscale.cu +92 -6
  240. data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +110 -22
  241. data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +4 -0
  242. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +58 -36
  243. data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +4 -3
  244. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +10 -2
  245. data/ext/sources/ggml/src/ggml-impl.h +119 -9
  246. data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +10 -7
  247. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
  248. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  249. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
  250. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +600 -0
  251. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +1376 -0
  252. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +226 -0
  253. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +1308 -0
  254. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +136 -63
  255. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +3158 -0
  256. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +82 -0
  257. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +718 -0
  258. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +2854 -1503
  259. data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +18 -8
  260. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +18 -0
  261. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +2510 -242
  262. data/ext/sources/ggml/src/ggml-opencl/kernels/add.cl +107 -0
  263. data/ext/sources/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
  264. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
  265. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
  266. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +84 -0
  267. data/ext/sources/ggml/src/ggml-opencl/kernels/div.cl +66 -0
  268. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
  269. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +370 -0
  270. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
  271. data/ext/sources/ggml/src/ggml-opencl/kernels/gelu.cl +27 -0
  272. data/ext/sources/ggml/src/ggml-opencl/kernels/glu.cl +177 -0
  273. data/ext/sources/ggml/src/ggml-opencl/kernels/group_norm.cl +49 -0
  274. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f16.cl +1 -1
  275. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f32.cl +1 -1
  276. data/ext/sources/ggml/src/ggml-opencl/kernels/mul.cl +73 -0
  277. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
  278. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +132 -0
  279. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +133 -0
  280. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
  281. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
  282. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
  283. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
  284. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
  285. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
  286. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
  287. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
  288. data/ext/sources/ggml/src/ggml-opencl/kernels/norm.cl +80 -0
  289. data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +79 -0
  290. data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +3 -2
  291. data/ext/sources/ggml/src/ggml-opencl/kernels/set_rows.cl +189 -0
  292. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +34 -13
  293. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +34 -13
  294. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f16.cl +34 -13
  295. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f32.cl +34 -13
  296. data/ext/sources/ggml/src/ggml-opencl/kernels/sub.cl +66 -0
  297. data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +20 -0
  298. data/ext/sources/ggml/src/ggml-opencl/kernels/tsembd.cl +2 -2
  299. data/ext/sources/ggml/src/ggml-opencl/kernels/upscale.cl +2 -3
  300. data/ext/sources/ggml/src/ggml-opt.cpp +97 -41
  301. data/ext/sources/ggml/src/ggml-quants.c +111 -16
  302. data/ext/sources/ggml/src/ggml-quants.h +6 -0
  303. data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +67 -47
  304. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +2 -0
  305. data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +15 -5
  306. data/ext/sources/ggml/src/ggml-sycl/binbcast.hpp +6 -0
  307. data/ext/sources/ggml/src/ggml-sycl/concat.cpp +25 -16
  308. data/ext/sources/ggml/src/ggml-sycl/conv.cpp +10 -4
  309. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +166 -99
  310. data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +72 -306
  311. data/ext/sources/ggml/src/ggml-sycl/cpy.hpp +213 -1
  312. data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +67 -49
  313. data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +1 -31
  314. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +79 -29
  315. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +2 -0
  316. data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +14 -26
  317. data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +9 -6
  318. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +328 -323
  319. data/ext/sources/ggml/src/ggml-sycl/gla.cpp +2 -2
  320. data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +2 -2
  321. data/ext/sources/ggml/src/ggml-sycl/mmq.cpp +80 -60
  322. data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +201 -132
  323. data/ext/sources/ggml/src/ggml-sycl/norm.cpp +74 -55
  324. data/ext/sources/ggml/src/ggml-sycl/quantize.hpp +133 -0
  325. data/ext/sources/ggml/src/ggml-sycl/quants.hpp +8 -9
  326. data/ext/sources/ggml/src/ggml-sycl/rope.cpp +35 -42
  327. data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +234 -0
  328. data/ext/sources/ggml/src/ggml-sycl/set_rows.hpp +8 -0
  329. data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +3 -3
  330. data/ext/sources/ggml/src/ggml-sycl/tsembd.cpp +12 -6
  331. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +2 -6
  332. data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +16 -12
  333. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +3492 -883
  334. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +41 -1
  335. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +42 -0
  336. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +13 -4
  337. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +39 -29
  338. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +349 -0
  339. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +2 -2
  340. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +66 -12
  341. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp +154 -0
  342. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp +21 -0
  343. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +1 -1
  344. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +2 -1
  345. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +6 -5
  346. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +4 -2
  347. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +32 -0
  348. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +1 -1
  349. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
  350. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +1 -1
  351. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +1 -1
  352. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
  353. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +21 -0
  354. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +69 -24
  355. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +60 -20
  356. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +98 -42
  357. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +64 -27
  358. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +74 -13
  359. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
  360. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
  361. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
  362. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp +4 -17
  363. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +19 -10
  364. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +25 -15
  365. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.comp +4 -0
  366. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
  367. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
  368. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +18 -14
  369. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +126 -0
  370. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +65 -1
  371. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +11 -7
  372. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +140 -0
  373. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +144 -531
  374. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +206 -38
  375. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.comp +556 -0
  376. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +12 -5
  377. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +15 -9
  378. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +111 -0
  379. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +22 -0
  380. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +24 -3
  381. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +53 -3
  382. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +55 -11
  383. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +65 -0
  384. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
  385. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +1 -4
  386. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -9
  387. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -9
  388. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -9
  389. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.comp +5 -0
  390. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +1 -1
  391. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +29 -7
  392. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +4 -0
  393. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +17 -0
  394. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +38 -5
  395. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +14 -0
  396. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +4 -3
  397. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.comp +101 -9
  398. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +69 -5
  399. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/utils.comp +25 -0
  400. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +335 -77
  401. data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +54 -0
  402. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1558 -0
  403. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add.tmpl.wgsl +44 -0
  404. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_in_place.tmpl.wgsl +41 -0
  405. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
  406. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
  407. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +60 -0
  408. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +124 -0
  409. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
  410. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
  411. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul.tmpl.wgsl +44 -0
  412. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_in_place.tmpl.wgsl +41 -0
  413. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +907 -0
  414. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +57 -0
  415. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_in_place.wgsl +48 -0
  416. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +81 -0
  417. data/ext/sources/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
  418. data/ext/sources/ggml/src/ggml-zdnn/common.hpp +59 -0
  419. data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +628 -0
  420. data/ext/sources/ggml/src/ggml-zdnn/mmf.cpp +80 -0
  421. data/ext/sources/ggml/src/ggml-zdnn/mmf.hpp +12 -0
  422. data/ext/sources/ggml/src/ggml-zdnn/utils.cpp +79 -0
  423. data/ext/sources/ggml/src/ggml-zdnn/utils.hpp +19 -0
  424. data/ext/sources/ggml/src/ggml.c +478 -98
  425. data/ext/sources/ggml/src/gguf.cpp +8 -1
  426. data/ext/sources/src/whisper.cpp +23 -46
  427. data/ext/sources/tests/CMakeLists.txt +8 -1
  428. data/ext/sources/tests/test-vad-full.cpp +3 -3
  429. data/ext/sources/tests/test-vad.cpp +2 -2
  430. data/lib/whisper/model/uri.rb +1 -1
  431. data/sig/whisper.rbs +7 -0
  432. data/test/test_params.rb +8 -0
  433. data/test/test_whisper.rb +1 -1
  434. data/whispercpp.gemspec +1 -1
  435. metadata +164 -157
  436. data/ext/sources/examples/talk-llama/llama-kv-cache-unified-iswa.cpp +0 -279
  437. data/ext/sources/examples/talk-llama/llama-kv-cache-unified.cpp +0 -1841
  438. data/ext/sources/examples/talk-llama/llama-kv-cache-unified.h +0 -303
  439. data/ext/sources/ggml/include/ggml-kompute.h +0 -50
  440. data/ext/sources/ggml/src/ggml-amx/CMakeLists.txt +0 -107
  441. data/ext/sources/ggml/src/ggml-amx/common.h +0 -94
  442. data/ext/sources/ggml/src/ggml-amx/ggml-amx.cpp +0 -446
  443. data/ext/sources/ggml/src/ggml-amx/mmq.cpp +0 -2510
  444. data/ext/sources/ggml/src/ggml-amx/mmq.h +0 -17
  445. data/ext/sources/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
  446. data/ext/sources/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
  447. data/ext/sources/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
  448. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
  449. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
  450. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
  451. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
  452. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
  453. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
  454. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
  455. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -357
  456. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
  457. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -365
  458. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
  459. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -482
  460. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -472
  461. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
  462. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
  463. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
  464. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
  465. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
  466. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
  467. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
  468. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
  469. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
  470. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
  471. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
  472. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
  473. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
  474. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
  475. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
  476. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
  477. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
  478. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
  479. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
  480. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
  481. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
  482. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
  483. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
  484. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
  485. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
  486. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
  487. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
  488. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
  489. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
  490. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
  491. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
  492. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
  493. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
  494. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
  495. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
  496. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
  497. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
  498. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
  499. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
  500. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
  501. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
  502. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
  503. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
  504. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
  505. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
  506. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
  507. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
  508. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
  509. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
  510. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
  511. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
  512. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
  513. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
  514. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
  515. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
  516. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
  517. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
  518. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
  519. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
  520. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
  521. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
  522. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
  523. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
  524. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
  525. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
  526. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
  527. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
  528. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
  529. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
  530. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
  531. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
  532. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
  533. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
  534. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
  535. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
  536. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
  537. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
  538. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
  539. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
  540. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
  541. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
  542. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
  543. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
  544. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
  545. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
  546. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
  547. data/ext/sources/ggml/src/ggml-kompute/CMakeLists.txt +0 -166
  548. data/ext/sources/ggml/src/ggml-kompute/ggml-kompute.cpp +0 -2251
  549. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/common.comp +0 -112
  550. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +0 -58
  551. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +0 -25
  552. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +0 -52
  553. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +0 -52
  554. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +0 -52
  555. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +0 -52
  556. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +0 -30
  557. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +0 -22
  558. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +0 -17
  559. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +0 -31
  560. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +0 -31
  561. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +0 -38
  562. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +0 -39
  563. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +0 -44
  564. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +0 -52
  565. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +0 -69
  566. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +0 -51
  567. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +0 -33
  568. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +0 -35
  569. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +0 -140
  570. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +0 -106
  571. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +0 -73
  572. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +0 -52
  573. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +0 -28
  574. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +0 -84
  575. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +0 -21
  576. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +0 -53
  577. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +0 -52
  578. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +0 -52
  579. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +0 -52
  580. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +0 -52
  581. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +0 -19
  582. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +0 -23
  583. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +0 -22
  584. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +0 -72
  585. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +0 -71
  586. data/ext/sources/ggml/src/ggml-metal/ggml-metal.m +0 -6280
@@ -20,8 +20,11 @@
20
20
  #define N_R0_Q5_1 4
21
21
  #define N_SG_Q5_1 2
22
22
 
23
- #define N_R0_Q8_0 4
24
- #define N_SG_Q8_0 2
23
+ #define N_R0_Q8_0 2
24
+ #define N_SG_Q8_0 4
25
+
26
+ #define N_R0_MXFP4 2
27
+ #define N_SG_MXFP4 2
25
28
 
26
29
  #define N_R0_Q2_K 4
27
30
  #define N_SG_Q2_K 2
@@ -29,13 +32,13 @@
29
32
  #define N_R0_Q3_K 2
30
33
  #define N_SG_Q3_K 2
31
34
 
32
- #define N_R0_Q4_K 4
35
+ #define N_R0_Q4_K 2
33
36
  #define N_SG_Q4_K 2
34
37
 
35
38
  #define N_R0_Q5_K 2
36
39
  #define N_SG_Q5_K 2
37
40
 
38
- #define N_R0_Q6_K 1
41
+ #define N_R0_Q6_K 2
39
42
  #define N_SG_Q6_K 2
40
43
 
41
44
  #define N_R0_IQ1_S 4
@@ -65,6 +68,13 @@
65
68
  #define N_R0_IQ4_XS 2
66
69
  #define N_SG_IQ4_XS 2
67
70
 
71
+ // function constants offsets
72
+ #define FC_FLASH_ATTN_EXT 100
73
+ #define FC_FLASH_ATTN_EXT_VEC 200
74
+ #define FC_FLASH_ATTN_EXT_VEC_REDUCE 300
75
+ #define FC_MUL_MV 400
76
+ #define FC_MUL_MM 500
77
+
68
78
  // kernel argument structs
69
79
  //
70
80
  // - element counters (e.g. ne00) typically use int32_t to reduce register usage
@@ -126,8 +136,18 @@ typedef struct {
126
136
  uint64_t nb2;
127
137
  uint64_t nb3;
128
138
  uint64_t offs;
139
+ uint64_t o1[8];
129
140
  } ggml_metal_kargs_bin;
130
141
 
142
+ typedef struct {
143
+ int64_t ne0;
144
+ int64_t ne1;
145
+ size_t nb01;
146
+ size_t nb02;
147
+ size_t nb11;
148
+ size_t nb21;
149
+ } ggml_metal_kargs_add_id;
150
+
131
151
  typedef struct {
132
152
  int32_t ne00;
133
153
  int32_t ne01;
@@ -147,6 +167,16 @@ typedef struct {
147
167
  uint64_t nb3;
148
168
  } ggml_metal_kargs_repeat;
149
169
 
170
+ typedef struct {
171
+ float scale;
172
+ float bias;
173
+ } ggml_metal_kargs_scale;
174
+
175
+ typedef struct {
176
+ float min;
177
+ float max;
178
+ } ggml_metal_kargs_clamp;
179
+
150
180
  typedef struct {
151
181
  int64_t ne00;
152
182
  int64_t ne01;
@@ -223,23 +253,68 @@ typedef struct {
223
253
  int32_t ne11;
224
254
  int32_t ne_12_2; // assume K and V are same shape
225
255
  int32_t ne_12_3;
256
+ int32_t ns10;
226
257
  uint64_t nb11;
227
258
  uint64_t nb12;
228
259
  uint64_t nb13;
260
+ int32_t ns20;
229
261
  uint64_t nb21;
230
262
  uint64_t nb22;
231
263
  uint64_t nb23;
264
+ int32_t ne32;
265
+ int32_t ne33;
232
266
  uint64_t nb31;
267
+ uint64_t nb32;
268
+ uint64_t nb33;
233
269
  int32_t ne1;
234
270
  int32_t ne2;
271
+ int32_t ne3;
235
272
  float scale;
236
273
  float max_bias;
237
274
  float m0;
238
275
  float m1;
239
- uint16_t n_head_log2;
276
+ int32_t n_head_log2;
240
277
  float logit_softcap;
241
278
  } ggml_metal_kargs_flash_attn_ext;
242
279
 
280
+ typedef struct {
281
+ int32_t ne01;
282
+ int32_t ne02;
283
+ int32_t ne03;
284
+ uint64_t nb01;
285
+ uint64_t nb02;
286
+ uint64_t nb03;
287
+ int32_t ne11;
288
+ int32_t ne_12_2; // assume K and V are same shape
289
+ int32_t ne_12_3;
290
+ int32_t ns10;
291
+ uint64_t nb11;
292
+ uint64_t nb12;
293
+ uint64_t nb13;
294
+ int32_t ns20;
295
+ uint64_t nb21;
296
+ uint64_t nb22;
297
+ uint64_t nb23;
298
+ int32_t ne32;
299
+ int32_t ne33;
300
+ uint64_t nb31;
301
+ uint64_t nb32;
302
+ uint64_t nb33;
303
+ int32_t ne1;
304
+ int32_t ne2;
305
+ int32_t ne3;
306
+ float scale;
307
+ float max_bias;
308
+ float m0;
309
+ float m1;
310
+ int32_t n_head_log2;
311
+ float logit_softcap;
312
+ } ggml_metal_kargs_flash_attn_ext_vec;
313
+
314
+ typedef struct {
315
+ int32_t nrows;
316
+ } ggml_metal_kargs_flash_attn_ext_vec_reduce;
317
+
243
318
  typedef struct {
244
319
  int32_t ne00;
245
320
  int32_t ne02;
@@ -274,6 +349,7 @@ typedef struct {
274
349
  uint64_t nb13;
275
350
  int32_t ne0;
276
351
  int32_t ne1;
352
+ int32_t nr0;
277
353
  int16_t r2;
278
354
  int16_t r3;
279
355
  } ggml_metal_kargs_mul_mv;
@@ -297,46 +373,34 @@ typedef struct {
297
373
  int32_t ne1;
298
374
  int16_t r2;
299
375
  int16_t r3;
300
- int16_t nsg;
301
- int16_t nxpsg;
302
- int16_t r1ptg;
303
376
  } ggml_metal_kargs_mul_mv_ext;
304
377
 
305
378
  typedef struct {
379
+ int32_t ne02;
306
380
  int32_t ne10;
307
381
  int32_t ne11; // n_expert_used (bcast)
308
382
  uint64_t nb11;
309
383
  uint64_t nb12;
310
- int32_t neh11; // n_tokens
311
- uint64_t nbh11;
384
+ int32_t ne21; // n_tokens
312
385
  int32_t ne20; // n_expert_used
313
386
  uint64_t nb21;
314
387
  } ggml_metal_kargs_mul_mm_id_map0;
315
388
 
316
- typedef struct {
317
- int32_t ne20; // n_expert_used
318
- int32_t neh0;
319
- int32_t neh1;
320
- uint64_t nbh1;
321
- uint64_t nbh2;
322
- int32_t ne0;
323
- uint64_t nb1;
324
- uint64_t nb2;
325
- } ggml_metal_kargs_mul_mm_id_map1;
326
-
327
389
  typedef struct {
328
390
  int32_t ne00;
329
391
  int32_t ne02;
330
392
  uint64_t nb01;
331
393
  uint64_t nb02;
332
394
  uint64_t nb03;
333
- int32_t neh12;
334
- uint64_t nbh10;
335
- uint64_t nbh11;
336
- uint64_t nbh12;
337
- uint64_t nbh13;
338
- int32_t neh0;
339
- int32_t neh1;
395
+ int32_t ne11;
396
+ uint64_t nb10;
397
+ uint64_t nb11;
398
+ uint64_t nb12;
399
+ uint64_t nb13;
400
+ int32_t ne20;
401
+ int32_t ne21;
402
+ int32_t ne0;
403
+ int32_t ne1;
340
404
  int16_t r2;
341
405
  int16_t r3;
342
406
  } ggml_metal_kargs_mul_mm_id;
@@ -361,22 +425,26 @@ typedef struct {
361
425
  int32_t ne0;
362
426
  int32_t ne1;
363
427
  uint64_t nb1;
428
+ int32_t nr0;
364
429
  } ggml_metal_kargs_mul_mv_id;
365
430
 
431
+ // NORM
432
+ // RMS_NORM
366
433
  typedef struct {
367
434
  int32_t ne00;
368
- int32_t ne00_4;
369
- uint64_t nb01;
435
+ int32_t ne00_t;
436
+ uint64_t nb1;
437
+ uint64_t nb2;
438
+ uint64_t nb3;
370
439
  float eps;
440
+ int32_t nef1[3];
441
+ int32_t nef2[3];
442
+ int32_t nef3[3];
443
+ uint64_t nbf1[3];
444
+ uint64_t nbf2[3];
445
+ uint64_t nbf3[3];
371
446
  } ggml_metal_kargs_norm;
372
447
 
373
- typedef struct {
374
- int32_t ne00;
375
- int32_t ne00_4;
376
- uint64_t nb01;
377
- float eps;
378
- } ggml_metal_kargs_rms_norm;
379
-
380
448
  typedef struct {
381
449
  int32_t ne00;
382
450
  int32_t ne00_4;
@@ -391,7 +459,7 @@ typedef struct {
391
459
  uint64_t nb00;
392
460
  uint64_t nb01;
393
461
  uint64_t nb02;
394
- int32_t n_groups;
462
+ int32_t ngrp;
395
463
  float eps;
396
464
  } ggml_metal_kargs_group_norm;
397
465
 
@@ -431,6 +499,8 @@ typedef struct{
431
499
  uint64_t nb1;
432
500
  int32_t i00;
433
501
  int32_t i10;
502
+ float alpha;
503
+ float limit;
434
504
  } ggml_metal_kargs_glu;
435
505
 
436
506
  typedef struct {
@@ -442,14 +512,6 @@ typedef struct {
442
512
  uint64_t nb01;
443
513
  uint64_t nb02;
444
514
  uint64_t nb03;
445
- int64_t ne10;
446
- int64_t ne11;
447
- int64_t ne12;
448
- int64_t ne13;
449
- uint64_t nb10;
450
- uint64_t nb11;
451
- uint64_t nb12;
452
- uint64_t nb13;
453
515
  int64_t ne0;
454
516
  int64_t ne1;
455
517
  int64_t ne2;
@@ -461,22 +523,28 @@ typedef struct {
461
523
  } ggml_metal_kargs_sum_rows;
462
524
 
463
525
  typedef struct {
464
- int64_t ne00;
465
- int64_t ne01;
466
- int64_t ne02;
526
+ int32_t ne00;
527
+ int32_t ne01;
528
+ int32_t ne02;
529
+ uint64_t nb01;
530
+ uint64_t nb02;
531
+ uint64_t nb03;
532
+ int32_t ne11;
533
+ int32_t ne12;
534
+ int32_t ne13;
535
+ uint64_t nb11;
536
+ uint64_t nb12;
537
+ uint64_t nb13;
538
+ uint64_t nb1;
539
+ uint64_t nb2;
540
+ uint64_t nb3;
467
541
  float scale;
468
542
  float max_bias;
469
543
  float m0;
470
544
  float m1;
471
- uint32_t n_head_log2;
545
+ int32_t n_head_log2;
472
546
  } ggml_metal_kargs_soft_max;
473
547
 
474
- typedef struct {
475
- int64_t ne00;
476
- int64_t ne01;
477
- int n_past;
478
- } ggml_metal_kargs_diag_mask_inf;
479
-
480
548
  typedef struct {
481
549
  int64_t ne00;
482
550
  int64_t ne01;
@@ -499,26 +567,26 @@ typedef struct {
499
567
  typedef struct {
500
568
  int64_t d_state;
501
569
  int64_t d_inner;
570
+ int64_t n_head;
571
+ int64_t n_group;
502
572
  int64_t n_seq_tokens;
503
573
  int64_t n_seqs;
504
- uint64_t nb00;
574
+ uint64_t s_off;
505
575
  uint64_t nb01;
506
576
  uint64_t nb02;
507
- uint64_t nb10;
577
+ uint64_t nb03;
508
578
  uint64_t nb11;
509
579
  uint64_t nb12;
510
580
  uint64_t nb13;
511
- uint64_t nb20;
512
581
  uint64_t nb21;
513
582
  uint64_t nb22;
514
- uint64_t nb30;
515
583
  uint64_t nb31;
516
- uint64_t nb40;
517
584
  uint64_t nb41;
518
585
  uint64_t nb42;
519
- uint64_t nb50;
586
+ uint64_t nb43;
520
587
  uint64_t nb51;
521
588
  uint64_t nb52;
589
+ uint64_t nb53;
522
590
  } ggml_metal_kargs_ssm_scan;
523
591
 
524
592
  typedef struct {
@@ -643,7 +711,12 @@ typedef struct {
643
711
  int64_t IW;
644
712
  int64_t OH;
645
713
  int64_t OW;
646
- int64_t parallel_elements;
714
+ int64_t np;
647
715
  } ggml_metal_kargs_pool_2d;
648
716
 
717
+ typedef struct {
718
+ int64_t ne00;
719
+ uint64_t nb01;
720
+ } ggml_metal_kargs_argmax;
721
+
649
722
  #endif // GGML_METAL_IMPL