whispercpp 1.3.3 → 1.3.5

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 (963) hide show
  1. checksums.yaml +4 -4
  2. data/README.md +60 -43
  3. data/ext/extconf.rb +2 -2
  4. data/ext/ruby_whisper.c +14 -2
  5. data/ext/ruby_whisper.h +39 -0
  6. data/ext/ruby_whisper_context.c +22 -22
  7. data/ext/ruby_whisper_model.c +12 -12
  8. data/ext/ruby_whisper_params.c +79 -25
  9. data/ext/ruby_whisper_segment.c +84 -19
  10. data/ext/ruby_whisper_token.c +351 -0
  11. data/ext/ruby_whisper_transcribe.cpp +1 -1
  12. data/ext/ruby_whisper_vad_context.c +75 -0
  13. data/ext/ruby_whisper_vad_context_detect.cpp +50 -0
  14. data/ext/ruby_whisper_vad_segment.c +139 -0
  15. data/ext/ruby_whisper_vad_segments.c +106 -0
  16. data/ext/sources/CMakeLists.txt +4 -1
  17. data/ext/sources/bindings/javascript/package.json +1 -1
  18. data/ext/sources/cmake/arm64-apple-clang.cmake +16 -0
  19. data/ext/sources/cmake/arm64-windows-llvm.cmake +16 -0
  20. data/ext/sources/cmake/riscv64-spacemit-linux-gnu-gcc.cmake +29 -0
  21. data/ext/sources/cmake/x64-windows-llvm.cmake +5 -0
  22. data/ext/sources/examples/CMakeLists.txt +1 -0
  23. data/ext/sources/examples/addon.node/addon.cpp +19 -19
  24. data/ext/sources/examples/addon.node/index.js +7 -5
  25. data/ext/sources/examples/addon.node/vad-example.js +2 -2
  26. data/ext/sources/examples/bench/bench.cpp +26 -16
  27. data/ext/sources/examples/bench.wasm/index-tmpl.html +10 -9
  28. data/ext/sources/examples/cli/cli.cpp +122 -111
  29. data/ext/sources/examples/command/command.cpp +26 -24
  30. data/ext/sources/examples/command.wasm/index-tmpl.html +5 -4
  31. data/ext/sources/examples/common-ggml.cpp +2 -0
  32. data/ext/sources/examples/lsp/CMakeLists.txt +2 -1
  33. data/ext/sources/examples/lsp/lsp.cpp +19 -17
  34. data/ext/sources/examples/quantize/CMakeLists.txt +2 -1
  35. data/ext/sources/examples/server/server.cpp +34 -24
  36. data/ext/sources/examples/server.py +6 -1
  37. data/ext/sources/examples/stream/stream.cpp +4 -2
  38. data/ext/sources/examples/stream.wasm/emscripten.cpp +6 -6
  39. data/ext/sources/examples/stream.wasm/index-tmpl.html +82 -5
  40. data/ext/sources/examples/talk-llama/CMakeLists.txt +7 -3
  41. data/ext/sources/examples/talk-llama/llama-adapter.cpp +113 -7
  42. data/ext/sources/examples/talk-llama/llama-adapter.h +13 -1
  43. data/ext/sources/examples/talk-llama/llama-arch.cpp +2136 -1491
  44. data/ext/sources/examples/talk-llama/llama-arch.h +125 -3
  45. data/ext/sources/examples/talk-llama/llama-batch.cpp +174 -100
  46. data/ext/sources/examples/talk-llama/llama-batch.h +46 -20
  47. data/ext/sources/examples/talk-llama/llama-chat.cpp +199 -8
  48. data/ext/sources/examples/talk-llama/llama-chat.h +11 -0
  49. data/ext/sources/examples/talk-llama/llama-context.cpp +1213 -413
  50. data/ext/sources/examples/talk-llama/llama-context.h +99 -36
  51. data/ext/sources/examples/talk-llama/llama-cparams.h +5 -4
  52. data/ext/sources/examples/talk-llama/llama-grammar.cpp +288 -53
  53. data/ext/sources/examples/talk-llama/llama-grammar.h +22 -1
  54. data/ext/sources/examples/talk-llama/llama-graph.cpp +883 -294
  55. data/ext/sources/examples/talk-llama/llama-graph.h +361 -161
  56. data/ext/sources/examples/talk-llama/llama-hparams.cpp +144 -6
  57. data/ext/sources/examples/talk-llama/llama-hparams.h +100 -23
  58. data/ext/sources/examples/talk-llama/llama-impl.cpp +7 -3
  59. data/ext/sources/examples/talk-llama/llama-impl.h +3 -1
  60. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +328 -0
  61. data/ext/sources/examples/talk-llama/{llama-kv-cache-unified-iswa.h → llama-kv-cache-iswa.h} +38 -29
  62. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +2100 -0
  63. data/ext/sources/examples/talk-llama/llama-kv-cache.h +373 -27
  64. data/ext/sources/examples/talk-llama/llama-kv-cells.h +124 -30
  65. data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +63 -41
  66. data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +30 -29
  67. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +77 -35
  68. data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +15 -16
  69. data/ext/sources/examples/talk-llama/llama-memory.h +16 -10
  70. data/ext/sources/examples/talk-llama/llama-mmap.cpp +172 -37
  71. data/ext/sources/examples/talk-llama/llama-mmap.h +8 -3
  72. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +93 -9
  73. data/ext/sources/examples/talk-llama/llama-model-loader.h +9 -2
  74. data/ext/sources/examples/talk-llama/llama-model-saver.cpp +3 -0
  75. data/ext/sources/examples/talk-llama/llama-model.cpp +3369 -10145
  76. data/ext/sources/examples/talk-llama/llama-model.h +104 -12
  77. data/ext/sources/examples/talk-llama/llama-quant.cpp +53 -30
  78. data/ext/sources/examples/talk-llama/llama-sampling.cpp +1520 -324
  79. data/ext/sources/examples/talk-llama/llama-sampling.h +19 -7
  80. data/ext/sources/examples/talk-llama/llama-vocab.cpp +562 -39
  81. data/ext/sources/examples/talk-llama/llama-vocab.h +50 -0
  82. data/ext/sources/examples/talk-llama/llama.cpp +794 -12
  83. data/ext/sources/examples/talk-llama/llama.h +246 -190
  84. data/ext/sources/examples/talk-llama/models/afmoe.cpp +191 -0
  85. data/ext/sources/examples/talk-llama/models/apertus.cpp +125 -0
  86. data/ext/sources/examples/talk-llama/models/arcee.cpp +135 -0
  87. data/ext/sources/examples/talk-llama/models/arctic.cpp +138 -0
  88. data/ext/sources/examples/talk-llama/models/arwkv7.cpp +86 -0
  89. data/ext/sources/examples/talk-llama/models/baichuan.cpp +122 -0
  90. data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +144 -0
  91. data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +135 -0
  92. data/ext/sources/examples/talk-llama/models/bert.cpp +178 -0
  93. data/ext/sources/examples/talk-llama/models/bitnet.cpp +160 -0
  94. data/ext/sources/examples/talk-llama/models/bloom.cpp +101 -0
  95. data/ext/sources/examples/talk-llama/models/chameleon.cpp +178 -0
  96. data/ext/sources/examples/talk-llama/models/chatglm.cpp +132 -0
  97. data/ext/sources/examples/talk-llama/models/codeshell.cpp +111 -0
  98. data/ext/sources/examples/talk-llama/models/cogvlm.cpp +102 -0
  99. data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +134 -0
  100. data/ext/sources/examples/talk-llama/models/command-r.cpp +122 -0
  101. data/ext/sources/examples/talk-llama/models/dbrx.cpp +123 -0
  102. data/ext/sources/examples/talk-llama/models/deci.cpp +135 -0
  103. data/ext/sources/examples/talk-llama/models/deepseek.cpp +144 -0
  104. data/ext/sources/examples/talk-llama/models/deepseek2.cpp +259 -0
  105. data/ext/sources/examples/talk-llama/models/dots1.cpp +134 -0
  106. data/ext/sources/examples/talk-llama/models/dream.cpp +105 -0
  107. data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +150 -0
  108. data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +110 -0
  109. data/ext/sources/examples/talk-llama/models/exaone.cpp +114 -0
  110. data/ext/sources/examples/talk-llama/models/exaone4.cpp +123 -0
  111. data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +113 -0
  112. data/ext/sources/examples/talk-llama/models/falcon.cpp +120 -0
  113. data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +116 -0
  114. data/ext/sources/examples/talk-llama/models/gemma.cpp +112 -0
  115. data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +128 -0
  116. data/ext/sources/examples/talk-llama/models/gemma3.cpp +155 -0
  117. data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +384 -0
  118. data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +170 -0
  119. data/ext/sources/examples/talk-llama/models/glm4.cpp +150 -0
  120. data/ext/sources/examples/talk-llama/models/gpt2.cpp +105 -0
  121. data/ext/sources/examples/talk-llama/models/gptneox.cpp +144 -0
  122. data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +196 -0
  123. data/ext/sources/examples/talk-llama/models/granite.cpp +211 -0
  124. data/ext/sources/examples/talk-llama/models/graph-context-mamba.cpp +283 -0
  125. data/ext/sources/examples/talk-llama/models/grok.cpp +159 -0
  126. data/ext/sources/examples/talk-llama/models/grovemoe.cpp +141 -0
  127. data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +132 -0
  128. data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +154 -0
  129. data/ext/sources/examples/talk-llama/models/internlm2.cpp +120 -0
  130. data/ext/sources/examples/talk-llama/models/jais.cpp +86 -0
  131. data/ext/sources/examples/talk-llama/models/jamba.cpp +106 -0
  132. data/ext/sources/examples/talk-llama/models/lfm2.cpp +175 -0
  133. data/ext/sources/examples/talk-llama/models/llada-moe.cpp +122 -0
  134. data/ext/sources/examples/talk-llama/models/llada.cpp +99 -0
  135. data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +178 -0
  136. data/ext/sources/examples/talk-llama/models/llama.cpp +168 -0
  137. data/ext/sources/examples/talk-llama/models/maincoder.cpp +117 -0
  138. data/ext/sources/examples/talk-llama/models/mamba.cpp +55 -0
  139. data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +123 -0
  140. data/ext/sources/examples/talk-llama/models/minicpm3.cpp +199 -0
  141. data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +124 -0
  142. data/ext/sources/examples/talk-llama/models/mistral3.cpp +160 -0
  143. data/ext/sources/examples/talk-llama/models/models.h +569 -0
  144. data/ext/sources/examples/talk-llama/models/modern-bert.cpp +116 -0
  145. data/ext/sources/examples/talk-llama/models/mpt.cpp +126 -0
  146. data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +150 -0
  147. data/ext/sources/examples/talk-llama/models/nemotron.cpp +122 -0
  148. data/ext/sources/examples/talk-llama/models/neo-bert.cpp +104 -0
  149. data/ext/sources/examples/talk-llama/models/olmo.cpp +121 -0
  150. data/ext/sources/examples/talk-llama/models/olmo2.cpp +150 -0
  151. data/ext/sources/examples/talk-llama/models/olmoe.cpp +124 -0
  152. data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +127 -0
  153. data/ext/sources/examples/talk-llama/models/openelm.cpp +124 -0
  154. data/ext/sources/examples/talk-llama/models/orion.cpp +123 -0
  155. data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +121 -0
  156. data/ext/sources/examples/talk-llama/models/phi2.cpp +121 -0
  157. data/ext/sources/examples/talk-llama/models/phi3.cpp +152 -0
  158. data/ext/sources/examples/talk-llama/models/plamo.cpp +110 -0
  159. data/ext/sources/examples/talk-llama/models/plamo2.cpp +316 -0
  160. data/ext/sources/examples/talk-llama/models/plamo3.cpp +128 -0
  161. data/ext/sources/examples/talk-llama/models/plm.cpp +168 -0
  162. data/ext/sources/examples/talk-llama/models/qwen.cpp +108 -0
  163. data/ext/sources/examples/talk-llama/models/qwen2.cpp +126 -0
  164. data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +151 -0
  165. data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +117 -0
  166. data/ext/sources/examples/talk-llama/models/qwen3.cpp +117 -0
  167. data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +124 -0
  168. data/ext/sources/examples/talk-llama/models/qwen3next.cpp +873 -0
  169. data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +149 -0
  170. data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +141 -0
  171. data/ext/sources/examples/talk-llama/models/refact.cpp +94 -0
  172. data/ext/sources/examples/talk-llama/models/rnd1.cpp +126 -0
  173. data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +162 -0
  174. data/ext/sources/examples/talk-llama/models/rwkv6.cpp +94 -0
  175. data/ext/sources/examples/talk-llama/models/rwkv6qwen2.cpp +86 -0
  176. data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +135 -0
  177. data/ext/sources/examples/talk-llama/models/rwkv7.cpp +90 -0
  178. data/ext/sources/examples/talk-llama/models/seed-oss.cpp +124 -0
  179. data/ext/sources/examples/talk-llama/models/smallthinker.cpp +126 -0
  180. data/ext/sources/examples/talk-llama/models/smollm3.cpp +128 -0
  181. data/ext/sources/examples/talk-llama/models/stablelm.cpp +146 -0
  182. data/ext/sources/examples/talk-llama/models/starcoder.cpp +100 -0
  183. data/ext/sources/examples/talk-llama/models/starcoder2.cpp +121 -0
  184. data/ext/sources/examples/talk-llama/models/t5-dec.cpp +166 -0
  185. data/ext/sources/examples/talk-llama/models/t5-enc.cpp +96 -0
  186. data/ext/sources/examples/talk-llama/models/wavtokenizer-dec.cpp +149 -0
  187. data/ext/sources/examples/talk-llama/models/xverse.cpp +108 -0
  188. data/ext/sources/examples/talk-llama/talk-llama.cpp +9 -6
  189. data/ext/sources/examples/talk-llama/unicode.cpp +309 -16
  190. data/ext/sources/examples/talk-llama/unicode.h +45 -0
  191. data/ext/sources/examples/vad-speech-segments/CMakeLists.txt +1 -1
  192. data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +4 -2
  193. data/ext/sources/examples/whisper.wasm/index-tmpl.html +18 -17
  194. data/ext/sources/ggml/CMakeLists.txt +135 -79
  195. data/ext/sources/ggml/cmake/ggml-config.cmake.in +132 -93
  196. data/ext/sources/ggml/include/ggml-alloc.h +9 -0
  197. data/ext/sources/ggml/include/ggml-backend.h +21 -2
  198. data/ext/sources/ggml/include/ggml-cpu.h +2 -1
  199. data/ext/sources/ggml/include/ggml-hexagon.h +19 -0
  200. data/ext/sources/ggml/include/ggml-metal.h +1 -6
  201. data/ext/sources/ggml/include/ggml-opt.h +25 -6
  202. data/ext/sources/ggml/include/ggml-rpc.h +8 -11
  203. data/ext/sources/ggml/include/ggml-webgpu.h +19 -0
  204. data/ext/sources/ggml/include/ggml-zdnn.h +17 -0
  205. data/ext/sources/ggml/include/ggml-zendnn.h +22 -0
  206. data/ext/sources/ggml/include/ggml.h +406 -23
  207. data/ext/sources/ggml/src/CMakeLists.txt +99 -13
  208. data/ext/sources/ggml/src/ggml-alloc.c +368 -161
  209. data/ext/sources/ggml/src/ggml-backend-impl.h +5 -5
  210. data/ext/sources/ggml/src/ggml-backend-reg.cpp +55 -14
  211. data/ext/sources/ggml/src/ggml-backend.cpp +290 -57
  212. data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +17 -3
  213. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +10 -13
  214. data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +14 -0
  215. data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +59 -45
  216. data/ext/sources/ggml/src/ggml-cann/acl_tensor.h +138 -47
  217. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +2586 -1917
  218. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +348 -309
  219. data/ext/sources/ggml/src/ggml-cann/common.h +350 -133
  220. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +894 -625
  221. data/ext/sources/ggml/src/ggml-common.h +17 -0
  222. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +167 -75
  223. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +5 -2
  224. data/ext/sources/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +4 -0
  225. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +560 -622
  226. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +1002 -270
  227. data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +107 -587
  228. data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +162 -589
  229. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/cpu-feats.cpp +38 -0
  230. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +373 -486
  231. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +3 -58
  232. data/ext/sources/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
  233. data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +521 -353
  234. data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +54 -314
  235. data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +184 -675
  236. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +4682 -1660
  237. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +82 -4
  238. data/ext/sources/ggml/src/ggml-cpu/common.h +14 -0
  239. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +18 -9
  240. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +263 -111
  241. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +39 -28
  242. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +683 -82
  243. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +38 -43
  244. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +435 -119
  245. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm-ppc.h +333 -0
  246. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +1234 -1182
  247. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.h +6 -0
  248. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +2167 -1480
  249. data/ext/sources/ggml/src/ggml-cpu/ops.h +10 -12
  250. data/ext/sources/ggml/src/ggml-cpu/quants.c +35 -0
  251. data/ext/sources/ggml/src/ggml-cpu/quants.h +8 -0
  252. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +1132 -81
  253. data/ext/sources/ggml/src/ggml-cpu/repack.h +36 -0
  254. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +120 -93
  255. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1025 -0
  256. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
  257. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
  258. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
  259. data/ext/sources/ggml/src/ggml-cpu/traits.cpp +2 -2
  260. data/ext/sources/ggml/src/ggml-cpu/traits.h +1 -1
  261. data/ext/sources/ggml/src/ggml-cpu/unary-ops.cpp +151 -0
  262. data/ext/sources/ggml/src/ggml-cpu/unary-ops.h +7 -0
  263. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +294 -27
  264. data/ext/sources/ggml/src/ggml-cpu/vec.h +606 -48
  265. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +92 -17
  266. data/ext/sources/ggml/src/ggml-cuda/add-id.cu +58 -0
  267. data/ext/sources/ggml/src/ggml-cuda/add-id.cuh +3 -0
  268. data/ext/sources/ggml/src/ggml-cuda/argmax.cu +2 -2
  269. data/ext/sources/ggml/src/ggml-cuda/argsort.cu +123 -6
  270. data/ext/sources/ggml/src/ggml-cuda/argsort.cuh +16 -0
  271. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +330 -191
  272. data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +2 -0
  273. data/ext/sources/ggml/src/ggml-cuda/common.cuh +588 -128
  274. data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cu +1 -4
  275. data/ext/sources/ggml/src/ggml-cuda/conv2d.cu +166 -0
  276. data/ext/sources/ggml/src/ggml-cuda/conv2d.cuh +5 -0
  277. data/ext/sources/ggml/src/ggml-cuda/convert.cu +95 -22
  278. data/ext/sources/ggml/src/ggml-cuda/convert.cuh +25 -0
  279. data/ext/sources/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
  280. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +335 -485
  281. data/ext/sources/ggml/src/ggml-cuda/cpy.cuh +1 -5
  282. data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cu +2 -14
  283. data/ext/sources/ggml/src/ggml-cuda/cumsum.cu +307 -0
  284. data/ext/sources/ggml/src/ggml-cuda/cumsum.cuh +5 -0
  285. data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +14 -40
  286. data/ext/sources/ggml/src/ggml-cuda/diag.cu +77 -0
  287. data/ext/sources/ggml/src/ggml-cuda/diag.cuh +5 -0
  288. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +519 -378
  289. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +750 -637
  290. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +49 -0
  291. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +1244 -0
  292. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +586 -0
  293. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +98 -61
  294. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +48 -0
  295. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +230 -197
  296. data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
  297. data/ext/sources/ggml/src/ggml-cuda/fill.cu +37 -0
  298. data/ext/sources/ggml/src/ggml-cuda/fill.cuh +3 -0
  299. data/ext/sources/ggml/src/ggml-cuda/getrows.cu +50 -39
  300. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +1557 -294
  301. data/ext/sources/ggml/src/ggml-cuda/im2col.cu +196 -35
  302. data/ext/sources/ggml/src/ggml-cuda/im2col.cuh +1 -0
  303. data/ext/sources/ggml/src/ggml-cuda/mean.cu +57 -2
  304. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +915 -69
  305. data/ext/sources/ggml/src/ggml-cuda/mmf.cu +171 -0
  306. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +835 -0
  307. data/ext/sources/ggml/src/ggml-cuda/mmid.cu +164 -0
  308. data/ext/sources/ggml/src/ggml-cuda/mmid.cuh +5 -0
  309. data/ext/sources/ggml/src/ggml-cuda/mmq.cu +109 -67
  310. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +1601 -733
  311. data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +802 -0
  312. data/ext/sources/ggml/src/ggml-cuda/mmvf.cuh +12 -0
  313. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +286 -149
  314. data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +1 -1
  315. data/ext/sources/ggml/src/ggml-cuda/norm.cu +284 -12
  316. data/ext/sources/ggml/src/ggml-cuda/norm.cuh +7 -0
  317. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
  318. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
  319. data/ext/sources/ggml/src/ggml-cuda/pad.cu +86 -32
  320. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
  321. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
  322. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +163 -10
  323. data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +14 -0
  324. data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +53 -0
  325. data/ext/sources/ggml/src/ggml-cuda/roll.cu +67 -0
  326. data/ext/sources/ggml/src/ggml-cuda/roll.cuh +5 -0
  327. data/ext/sources/ggml/src/ggml-cuda/rope.cu +207 -98
  328. data/ext/sources/ggml/src/ggml-cuda/rope.cuh +2 -0
  329. data/ext/sources/ggml/src/ggml-cuda/scale.cu +14 -11
  330. data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +330 -0
  331. data/ext/sources/ggml/src/ggml-cuda/set-rows.cuh +7 -0
  332. data/ext/sources/ggml/src/ggml-cuda/set.cu +39 -0
  333. data/ext/sources/ggml/src/ggml-cuda/set.cuh +7 -0
  334. data/ext/sources/ggml/src/ggml-cuda/softcap.cu +34 -0
  335. data/ext/sources/ggml/src/ggml-cuda/softcap.cuh +5 -0
  336. data/ext/sources/ggml/src/ggml-cuda/softmax.cu +325 -61
  337. data/ext/sources/ggml/src/ggml-cuda/solve_tri.cu +275 -0
  338. data/ext/sources/ggml/src/ggml-cuda/solve_tri.cuh +3 -0
  339. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +14 -12
  340. data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +291 -104
  341. data/ext/sources/ggml/src/ggml-cuda/sum.cu +6 -10
  342. data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +21 -4
  343. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
  344. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
  345. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
  346. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
  347. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
  348. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
  349. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
  350. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
  351. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
  352. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  353. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  354. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  355. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  356. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  357. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  358. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  359. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  360. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  361. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  362. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  363. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  364. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  365. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  366. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  367. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  368. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  369. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  370. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  371. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  372. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  373. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  374. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  375. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  376. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  377. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  378. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  379. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  380. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  381. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  382. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  383. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  384. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  385. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  386. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  387. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  388. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +40 -19
  389. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  390. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  391. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  392. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  393. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  394. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  395. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  396. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  397. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  398. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  399. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  400. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  401. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  402. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  403. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  404. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  405. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
  406. data/ext/sources/ggml/src/ggml-cuda/top-k.cu +96 -0
  407. data/ext/sources/ggml/src/ggml-cuda/top-k.cuh +3 -0
  408. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +351 -0
  409. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +21 -0
  410. data/ext/sources/ggml/src/ggml-cuda/tri.cu +136 -0
  411. data/ext/sources/ggml/src/ggml-cuda/tri.cuh +5 -0
  412. data/ext/sources/ggml/src/ggml-cuda/tsembd.cu +3 -3
  413. data/ext/sources/ggml/src/ggml-cuda/unary.cu +189 -5
  414. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +44 -0
  415. data/ext/sources/ggml/src/ggml-cuda/upscale.cu +248 -6
  416. data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +110 -22
  417. data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +8 -0
  418. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +70 -37
  419. data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +10 -3
  420. data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +80 -0
  421. data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3151 -0
  422. data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +44 -0
  423. data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +682 -0
  424. data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +360 -0
  425. data/ext/sources/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
  426. data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +566 -0
  427. data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +112 -0
  428. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +35 -0
  429. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-dma.c +63 -0
  430. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-dma.h +157 -0
  431. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +165 -0
  432. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +92 -0
  433. data/ext/sources/ggml/src/ggml-hexagon/htp/htp_iface.idl +16 -0
  434. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.c +94 -0
  435. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.c +72 -0
  436. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +49 -0
  437. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.c +1020 -0
  438. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +1353 -0
  439. data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +1001 -0
  440. data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +2503 -0
  441. data/ext/sources/ggml/src/ggml-hexagon/htp/ops-utils.h +149 -0
  442. data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +487 -0
  443. data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +168 -0
  444. data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +402 -0
  445. data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +287 -0
  446. data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.c +297 -0
  447. data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
  448. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.c +454 -0
  449. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.h +221 -0
  450. data/ext/sources/ggml/src/ggml-hexagon/op-desc.h +153 -0
  451. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +16 -13
  452. data/ext/sources/ggml/src/ggml-impl.h +186 -15
  453. data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +10 -7
  454. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
  455. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  456. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
  457. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +609 -0
  458. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +1743 -0
  459. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +273 -0
  460. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +1686 -0
  461. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +356 -61
  462. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +4161 -0
  463. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +94 -0
  464. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +724 -0
  465. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +4495 -1876
  466. data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +21 -9
  467. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +29 -0
  468. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +4005 -427
  469. data/ext/sources/ggml/src/ggml-opencl/kernels/add.cl +107 -0
  470. data/ext/sources/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
  471. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
  472. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
  473. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +147 -0
  474. data/ext/sources/ggml/src/ggml-opencl/kernels/div.cl +66 -0
  475. data/ext/sources/ggml/src/ggml-opencl/kernels/expm1.cl +82 -0
  476. data/ext/sources/ggml/src/ggml-opencl/kernels/fill.cl +17 -0
  477. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
  478. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +371 -0
  479. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
  480. data/ext/sources/ggml/src/ggml-opencl/kernels/gelu.cl +27 -0
  481. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
  482. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
  483. data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +36 -12
  484. data/ext/sources/ggml/src/ggml-opencl/kernels/glu.cl +177 -0
  485. data/ext/sources/ggml/src/ggml-opencl/kernels/group_norm.cl +49 -0
  486. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f16.cl +1 -1
  487. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f32.cl +1 -1
  488. data/ext/sources/ggml/src/ggml-opencl/kernels/mean.cl +39 -0
  489. data/ext/sources/ggml/src/ggml-opencl/kernels/mul.cl +73 -0
  490. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
  491. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
  492. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +146 -0
  493. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +147 -0
  494. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
  495. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
  496. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
  497. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
  498. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
  499. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
  500. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
  501. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
  502. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
  503. data/ext/sources/ggml/src/ggml-opencl/kernels/norm.cl +80 -0
  504. data/ext/sources/ggml/src/ggml-opencl/kernels/pad.cl +29 -20
  505. data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +94 -0
  506. data/ext/sources/ggml/src/ggml-opencl/kernels/rope.cl +50 -24
  507. data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +3 -2
  508. data/ext/sources/ggml/src/ggml-opencl/kernels/set_rows.cl +208 -0
  509. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +34 -13
  510. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +34 -13
  511. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f16.cl +34 -13
  512. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f32.cl +34 -13
  513. data/ext/sources/ggml/src/ggml-opencl/kernels/softplus.cl +88 -0
  514. data/ext/sources/ggml/src/ggml-opencl/kernels/sqr.cl +53 -0
  515. data/ext/sources/ggml/src/ggml-opencl/kernels/sqrt.cl +53 -0
  516. data/ext/sources/ggml/src/ggml-opencl/kernels/ssm_conv.cl +77 -0
  517. data/ext/sources/ggml/src/ggml-opencl/kernels/sub.cl +66 -0
  518. data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +33 -0
  519. data/ext/sources/ggml/src/ggml-opencl/kernels/tsembd.cl +2 -2
  520. data/ext/sources/ggml/src/ggml-opencl/kernels/upscale.cl +2 -3
  521. data/ext/sources/ggml/src/ggml-opt.cpp +97 -41
  522. data/ext/sources/ggml/src/ggml-quants.c +111 -16
  523. data/ext/sources/ggml/src/ggml-quants.h +6 -0
  524. data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +497 -195
  525. data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +48 -3
  526. data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +77 -0
  527. data/ext/sources/ggml/src/ggml-sycl/add-id.hpp +8 -0
  528. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +8 -0
  529. data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +6 -5
  530. data/ext/sources/ggml/src/ggml-sycl/common.hpp +117 -15
  531. data/ext/sources/ggml/src/ggml-sycl/concat.cpp +50 -30
  532. data/ext/sources/ggml/src/ggml-sycl/conv.cpp +10 -4
  533. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +200 -99
  534. data/ext/sources/ggml/src/ggml-sycl/count-equal.cpp +79 -0
  535. data/ext/sources/ggml/src/ggml-sycl/count-equal.hpp +9 -0
  536. data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +72 -309
  537. data/ext/sources/ggml/src/ggml-sycl/cpy.hpp +213 -1
  538. data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +18 -0
  539. data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +67 -49
  540. data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +77 -34
  541. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +397 -314
  542. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +12 -2
  543. data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +14 -26
  544. data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +9 -6
  545. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +643 -413
  546. data/ext/sources/ggml/src/ggml-sycl/gla.cpp +2 -2
  547. data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +2 -2
  548. data/ext/sources/ggml/src/ggml-sycl/mmq.cpp +80 -60
  549. data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +223 -132
  550. data/ext/sources/ggml/src/ggml-sycl/norm.cpp +230 -55
  551. data/ext/sources/ggml/src/ggml-sycl/norm.hpp +2 -0
  552. data/ext/sources/ggml/src/ggml-sycl/pad.cpp +97 -0
  553. data/ext/sources/ggml/src/ggml-sycl/pad.hpp +24 -0
  554. data/ext/sources/ggml/src/ggml-sycl/pad_reflect_1d.cpp +100 -0
  555. data/ext/sources/ggml/src/ggml-sycl/pad_reflect_1d.hpp +10 -0
  556. data/ext/sources/ggml/src/ggml-sycl/presets.hpp +2 -0
  557. data/ext/sources/ggml/src/ggml-sycl/quantize.hpp +133 -0
  558. data/ext/sources/ggml/src/ggml-sycl/quants.hpp +8 -9
  559. data/ext/sources/ggml/src/ggml-sycl/repeat_back.cpp +76 -0
  560. data/ext/sources/ggml/src/ggml-sycl/repeat_back.hpp +8 -0
  561. data/ext/sources/ggml/src/ggml-sycl/roll.cpp +122 -0
  562. data/ext/sources/ggml/src/ggml-sycl/roll.hpp +20 -0
  563. data/ext/sources/ggml/src/ggml-sycl/rope.cpp +65 -59
  564. data/ext/sources/ggml/src/ggml-sycl/set.cpp +73 -0
  565. data/ext/sources/ggml/src/ggml-sycl/set.hpp +5 -0
  566. data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +234 -0
  567. data/ext/sources/ggml/src/ggml-sycl/set_rows.hpp +8 -0
  568. data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +330 -165
  569. data/ext/sources/ggml/src/ggml-sycl/softmax.hpp +4 -0
  570. data/ext/sources/ggml/src/ggml-sycl/ssm_conv.cpp +127 -0
  571. data/ext/sources/ggml/src/ggml-sycl/ssm_conv.hpp +5 -0
  572. data/ext/sources/ggml/src/ggml-sycl/tsembd.cpp +12 -6
  573. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +60 -6
  574. data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +16 -12
  575. data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +38 -18
  576. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +7398 -2635
  577. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/abs.comp +21 -0
  578. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +2 -2
  579. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +43 -3
  580. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add1.comp +28 -0
  581. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +42 -0
  582. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/arange.comp +20 -0
  583. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +15 -6
  584. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +56 -39
  585. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort_large.comp +114 -0
  586. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ceil.comp +22 -0
  587. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +2 -2
  588. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +2 -2
  589. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +2 -2
  590. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +1 -1
  591. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +347 -0
  592. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +1 -1
  593. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +2 -2
  594. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +5 -5
  595. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +67 -13
  596. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_transpose.comp +67 -0
  597. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +2 -2
  598. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +2 -2
  599. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/count_experts.comp +51 -0
  600. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum.comp +83 -0
  601. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass1.comp +60 -0
  602. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass2.comp +66 -0
  603. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +1 -1
  604. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs.comp → dequant_funcs.glsl} +158 -16
  605. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs_cm2.comp → dequant_funcs_cm2.glsl} +38 -3
  606. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_head.comp → dequant_head.glsl} +1 -1
  607. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
  608. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +1 -1
  609. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +2 -2
  610. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +1 -1
  611. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +3 -2
  612. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +7 -6
  613. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +5 -3
  614. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +1 -1
  615. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +1 -1
  616. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +32 -0
  617. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +4 -4
  618. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +2 -2
  619. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +1 -1
  620. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +1 -1
  621. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +4 -4
  622. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +1 -1
  623. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +1 -1
  624. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +4 -4
  625. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +2 -2
  626. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +1 -1
  627. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +29 -0
  628. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +1 -1
  629. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +2 -2
  630. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +21 -0
  631. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/fill.comp +19 -0
  632. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +103 -36
  633. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +220 -0
  634. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +139 -45
  635. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +113 -38
  636. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +75 -14
  637. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/floor.comp +22 -0
  638. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +2 -2
  639. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
  640. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
  641. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +2 -2
  642. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
  643. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +2 -2
  644. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_binary_head.comp → generic_binary_head.glsl} +19 -17
  645. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_head.comp → generic_head.glsl} +2 -0
  646. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_unary_head.comp → generic_unary_head.glsl} +7 -0
  647. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +21 -12
  648. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +28 -18
  649. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{glu_head.comp → glu_head.glsl} +4 -0
  650. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +2 -2
  651. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
  652. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
  653. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +33 -17
  654. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +125 -0
  655. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +2 -2
  656. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +2 -2
  657. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +18 -0
  658. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +2 -2
  659. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +2 -2
  660. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +227 -0
  661. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +35 -0
  662. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +71 -21
  663. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +41 -25
  664. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +2 -2
  665. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +44 -26
  666. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +2 -2
  667. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +2 -2
  668. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +2 -2
  669. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +20 -14
  670. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +9 -7
  671. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +4 -6
  672. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +2 -2
  673. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +4 -6
  674. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +4 -6
  675. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +2 -2
  676. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +143 -0
  677. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +494 -0
  678. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +144 -556
  679. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +230 -51
  680. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +566 -0
  681. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +72 -0
  682. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +90 -223
  683. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +454 -0
  684. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +78 -0
  685. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +195 -0
  686. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/neg.comp +20 -0
  687. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +2 -2
  688. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +2 -2
  689. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +22 -0
  690. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +41 -5
  691. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +1 -1
  692. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +59 -9
  693. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +2 -2
  694. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +2 -2
  695. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +2 -2
  696. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +2 -2
  697. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +104 -14
  698. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +2 -2
  699. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +65 -0
  700. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
  701. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +234 -0
  702. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +20 -0
  703. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +6 -52
  704. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +6 -35
  705. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +6 -35
  706. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +28 -0
  707. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +6 -39
  708. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/round.comp +29 -0
  709. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.glsl +5 -0
  710. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +3 -3
  711. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +2 -2
  712. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +2 -2
  713. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +2 -2
  714. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +2 -2
  715. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +30 -8
  716. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +6 -2
  717. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large1.comp +62 -0
  718. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large2.comp +79 -0
  719. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large3.comp +65 -0
  720. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large_common.glsl +53 -0
  721. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp +23 -0
  722. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/solve_tri.comp +81 -0
  723. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +17 -0
  724. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +2 -2
  725. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +44 -0
  726. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_scan.comp +124 -0
  727. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/step.comp +22 -0
  728. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +2 -2
  729. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +16 -6
  730. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.glsl +25 -0
  731. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +2 -2
  732. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +14 -0
  733. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +2 -2
  734. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +5 -4
  735. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_argsort.comp +118 -0
  736. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp +213 -0
  737. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_nary_search.comp +246 -0
  738. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +43 -0
  739. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/trunc.comp +22 -0
  740. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{types.comp → types.glsl} +435 -24
  741. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +148 -6
  742. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/utils.glsl +25 -0
  743. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +619 -177
  744. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/xielu.comp +35 -0
  745. data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +80 -0
  746. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +169 -0
  747. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +3087 -0
  748. data/ext/sources/ggml/src/ggml-webgpu/pre_wgsl.hpp +778 -0
  749. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +188 -0
  750. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
  751. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
  752. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +101 -0
  753. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +147 -0
  754. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +591 -0
  755. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
  756. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +323 -0
  757. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
  758. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +907 -0
  759. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +97 -0
  760. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl +247 -0
  761. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.tmpl.wgsl +302 -0
  762. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +267 -0
  763. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +123 -0
  764. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rope.tmpl.wgsl +295 -0
  765. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl +90 -0
  766. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.tmpl.wgsl +112 -0
  767. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +81 -0
  768. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.tmpl.wgsl +345 -0
  769. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary_op.wgsl +483 -0
  770. data/ext/sources/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
  771. data/ext/sources/ggml/src/ggml-zdnn/common.hpp +59 -0
  772. data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +628 -0
  773. data/ext/sources/ggml/src/ggml-zdnn/mmf.cpp +80 -0
  774. data/ext/sources/ggml/src/ggml-zdnn/mmf.hpp +12 -0
  775. data/ext/sources/ggml/src/ggml-zdnn/utils.cpp +79 -0
  776. data/ext/sources/ggml/src/ggml-zdnn/utils.hpp +19 -0
  777. data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +92 -0
  778. data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +466 -0
  779. data/ext/sources/ggml/src/ggml.c +901 -129
  780. data/ext/sources/ggml/src/gguf.cpp +8 -1
  781. data/ext/sources/include/whisper.h +1 -0
  782. data/ext/sources/src/CMakeLists.txt +3 -1
  783. data/ext/sources/src/whisper.cpp +124 -81
  784. data/ext/sources/tests/CMakeLists.txt +8 -1
  785. data/ext/sources/tests/test-vad-full.cpp +7 -5
  786. data/ext/sources/tests/test-vad.cpp +3 -3
  787. data/extsources.rb +1 -0
  788. data/lib/whisper/model/uri.rb +17 -18
  789. data/sig/whisper.rbs +126 -2
  790. data/test/test_params.rb +24 -8
  791. data/test/test_segment.rb +0 -1
  792. data/test/test_token.rb +70 -0
  793. data/test/test_vad.rb +1 -1
  794. data/test/test_vad_context.rb +50 -0
  795. data/test/test_vad_segment.rb +19 -0
  796. data/test/test_vad_segments.rb +16 -0
  797. data/test/test_whisper.rb +8 -1
  798. data/whispercpp.gemspec +1 -1
  799. metadata +439 -179
  800. data/ext/sources/build-xcframework.sh +0 -547
  801. data/ext/sources/examples/talk-llama/llama-kv-cache-unified-iswa.cpp +0 -279
  802. data/ext/sources/examples/talk-llama/llama-kv-cache-unified.cpp +0 -1841
  803. data/ext/sources/examples/talk-llama/llama-kv-cache-unified.h +0 -303
  804. data/ext/sources/ggml/include/ggml-kompute.h +0 -50
  805. data/ext/sources/ggml/src/ggml-amx/CMakeLists.txt +0 -107
  806. data/ext/sources/ggml/src/ggml-amx/common.h +0 -94
  807. data/ext/sources/ggml/src/ggml-amx/ggml-amx.cpp +0 -446
  808. data/ext/sources/ggml/src/ggml-amx/mmq.cpp +0 -2510
  809. data/ext/sources/ggml/src/ggml-amx/mmq.h +0 -17
  810. data/ext/sources/ggml/src/ggml-cann/Doxyfile +0 -2579
  811. data/ext/sources/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
  812. data/ext/sources/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
  813. data/ext/sources/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
  814. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
  815. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
  816. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
  817. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
  818. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
  819. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
  820. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
  821. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -357
  822. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
  823. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -365
  824. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
  825. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -482
  826. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -472
  827. data/ext/sources/ggml/src/ggml-cuda/mmv.cu +0 -506
  828. data/ext/sources/ggml/src/ggml-cuda/mmv.cuh +0 -11
  829. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
  830. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
  831. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
  832. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
  833. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
  834. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
  835. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
  836. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
  837. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
  838. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
  839. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
  840. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
  841. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
  842. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
  843. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
  844. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
  845. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
  846. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
  847. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
  848. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
  849. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
  850. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
  851. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
  852. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
  853. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
  854. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
  855. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
  856. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
  857. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
  858. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
  859. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
  860. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
  861. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
  862. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
  863. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
  864. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
  865. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
  866. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
  867. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
  868. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
  869. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
  870. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
  871. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
  872. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
  873. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
  874. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
  875. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
  876. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
  877. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
  878. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
  879. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
  880. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
  881. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
  882. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
  883. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
  884. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
  885. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
  886. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
  887. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
  888. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
  889. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
  890. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
  891. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
  892. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
  893. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
  894. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
  895. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
  896. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
  897. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
  898. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
  899. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
  900. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
  901. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
  902. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
  903. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
  904. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
  905. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
  906. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
  907. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
  908. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
  909. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
  910. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
  911. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
  912. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
  913. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
  914. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
  915. data/ext/sources/ggml/src/ggml-kompute/CMakeLists.txt +0 -166
  916. data/ext/sources/ggml/src/ggml-kompute/ggml-kompute.cpp +0 -2251
  917. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/common.comp +0 -112
  918. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +0 -58
  919. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +0 -25
  920. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +0 -52
  921. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +0 -52
  922. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +0 -52
  923. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +0 -52
  924. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +0 -30
  925. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +0 -22
  926. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +0 -17
  927. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +0 -31
  928. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +0 -31
  929. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +0 -38
  930. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +0 -39
  931. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +0 -44
  932. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +0 -52
  933. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +0 -69
  934. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +0 -51
  935. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +0 -33
  936. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +0 -35
  937. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +0 -140
  938. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +0 -106
  939. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +0 -73
  940. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +0 -52
  941. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +0 -28
  942. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +0 -84
  943. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +0 -21
  944. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +0 -53
  945. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +0 -52
  946. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +0 -52
  947. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +0 -52
  948. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +0 -52
  949. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +0 -19
  950. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +0 -23
  951. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +0 -22
  952. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +0 -72
  953. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +0 -71
  954. data/ext/sources/ggml/src/ggml-metal/ggml-metal.m +0 -6280
  955. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +0 -162
  956. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +0 -118
  957. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +0 -99
  958. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +0 -58
  959. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_bfloat16_support.comp → feature-tests/bfloat16.comp} +0 -0
  960. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat_support.comp → feature-tests/coopmat.comp} +0 -0
  961. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat2_support.comp → feature-tests/coopmat2.comp} +0 -0
  962. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_integer_dot_support.comp → feature-tests/integer_dot.comp} +0 -0
  963. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{glu_main.comp → glu_main.glsl} +0 -0
@@ -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 +68,25 @@
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_PAD 100
73
+ #define FC_FLASH_ATTN_EXT_BLK 200
74
+ #define FC_FLASH_ATTN_EXT 300
75
+ #define FC_FLASH_ATTN_EXT_VEC 400
76
+ #define FC_FLASH_ATTN_EXT_VEC_REDUCE 500
77
+ #define FC_MUL_MV 600
78
+ #define FC_MUL_MM 700
79
+ #define FC_ROPE 800
80
+ #define FC_SSM_CONV 900
81
+ #define FC_COUNT_EQUAL 1000
82
+
83
+ // op-specific constants
84
+ #define OP_FLASH_ATTN_EXT_NQPTG 8
85
+ #define OP_FLASH_ATTN_EXT_NCPSG 64
86
+
87
+ #define OP_FLASH_ATTN_EXT_VEC_NQPTG 1
88
+ #define OP_FLASH_ATTN_EXT_VEC_NCPSG 32
89
+
68
90
  // kernel argument structs
69
91
  //
70
92
  // - element counters (e.g. ne00) typically use int32_t to reduce register usage
@@ -126,8 +148,18 @@ typedef struct {
126
148
  uint64_t nb2;
127
149
  uint64_t nb3;
128
150
  uint64_t offs;
151
+ uint64_t o1[8];
129
152
  } ggml_metal_kargs_bin;
130
153
 
154
+ typedef struct {
155
+ int64_t ne0;
156
+ int64_t ne1;
157
+ size_t nb01;
158
+ size_t nb02;
159
+ size_t nb11;
160
+ size_t nb21;
161
+ } ggml_metal_kargs_add_id;
162
+
131
163
  typedef struct {
132
164
  int32_t ne00;
133
165
  int32_t ne01;
@@ -148,6 +180,21 @@ typedef struct {
148
180
  } ggml_metal_kargs_repeat;
149
181
 
150
182
  typedef struct {
183
+ float scale;
184
+ float bias;
185
+ } ggml_metal_kargs_scale;
186
+
187
+ typedef struct {
188
+ float val;
189
+ } ggml_metal_kargs_fill;
190
+
191
+ typedef struct {
192
+ float min;
193
+ float max;
194
+ } ggml_metal_kargs_clamp;
195
+
196
+ typedef struct {
197
+ int64_t nk0;
151
198
  int64_t ne00;
152
199
  int64_t ne01;
153
200
  int64_t ne02;
@@ -211,8 +258,38 @@ typedef struct {
211
258
  int32_t sect_1;
212
259
  int32_t sect_2;
213
260
  int32_t sect_3;
261
+ bool src2;
214
262
  } ggml_metal_kargs_rope;
215
263
 
264
+ typedef struct {
265
+ int32_t ne11;
266
+ int32_t ne_12_2; // assume K and V are same shape
267
+ int32_t ne_12_3;
268
+ uint64_t nb11;
269
+ uint64_t nb12;
270
+ uint64_t nb13;
271
+ uint64_t nb21;
272
+ uint64_t nb22;
273
+ uint64_t nb23;
274
+ int32_t ne31;
275
+ int32_t ne32;
276
+ int32_t ne33;
277
+ uint64_t nb31;
278
+ uint64_t nb32;
279
+ uint64_t nb33;
280
+ } ggml_metal_kargs_flash_attn_ext_pad;
281
+
282
+ typedef struct {
283
+ int32_t ne01;
284
+ int32_t ne30;
285
+ int32_t ne31;
286
+ int32_t ne32;
287
+ int32_t ne33;
288
+ uint64_t nb31;
289
+ uint64_t nb32;
290
+ uint64_t nb33;
291
+ } ggml_metal_kargs_flash_attn_ext_blk;
292
+
216
293
  typedef struct {
217
294
  int32_t ne01;
218
295
  int32_t ne02;
@@ -223,23 +300,70 @@ typedef struct {
223
300
  int32_t ne11;
224
301
  int32_t ne_12_2; // assume K and V are same shape
225
302
  int32_t ne_12_3;
303
+ int32_t ns10;
226
304
  uint64_t nb11;
227
305
  uint64_t nb12;
228
306
  uint64_t nb13;
307
+ int32_t ns20;
229
308
  uint64_t nb21;
230
309
  uint64_t nb22;
231
310
  uint64_t nb23;
311
+ int32_t ne31;
312
+ int32_t ne32;
313
+ int32_t ne33;
232
314
  uint64_t nb31;
315
+ uint64_t nb32;
316
+ uint64_t nb33;
233
317
  int32_t ne1;
234
318
  int32_t ne2;
319
+ int32_t ne3;
235
320
  float scale;
236
321
  float max_bias;
237
322
  float m0;
238
323
  float m1;
239
- uint16_t n_head_log2;
324
+ int32_t n_head_log2;
240
325
  float logit_softcap;
241
326
  } ggml_metal_kargs_flash_attn_ext;
242
327
 
328
+ typedef struct {
329
+ int32_t ne01;
330
+ int32_t ne02;
331
+ int32_t ne03;
332
+ uint64_t nb01;
333
+ uint64_t nb02;
334
+ uint64_t nb03;
335
+ int32_t ne11;
336
+ int32_t ne_12_2; // assume K and V are same shape
337
+ int32_t ne_12_3;
338
+ int32_t ns10;
339
+ uint64_t nb11;
340
+ uint64_t nb12;
341
+ uint64_t nb13;
342
+ int32_t ns20;
343
+ uint64_t nb21;
344
+ uint64_t nb22;
345
+ uint64_t nb23;
346
+ int32_t ne31;
347
+ int32_t ne32;
348
+ int32_t ne33;
349
+ uint64_t nb31;
350
+ uint64_t nb32;
351
+ uint64_t nb33;
352
+ int32_t ne1;
353
+ int32_t ne2;
354
+ int32_t ne3;
355
+ float scale;
356
+ float max_bias;
357
+ float m0;
358
+ float m1;
359
+ int32_t n_head_log2;
360
+ float logit_softcap;
361
+ } ggml_metal_kargs_flash_attn_ext_vec;
362
+
363
+ typedef struct {
364
+ int32_t nrows;
365
+ } ggml_metal_kargs_flash_attn_ext_vec_reduce;
366
+
243
367
  typedef struct {
244
368
  int32_t ne00;
245
369
  int32_t ne02;
@@ -274,6 +398,7 @@ typedef struct {
274
398
  uint64_t nb13;
275
399
  int32_t ne0;
276
400
  int32_t ne1;
401
+ int32_t nr0;
277
402
  int16_t r2;
278
403
  int16_t r3;
279
404
  } ggml_metal_kargs_mul_mv;
@@ -297,46 +422,34 @@ typedef struct {
297
422
  int32_t ne1;
298
423
  int16_t r2;
299
424
  int16_t r3;
300
- int16_t nsg;
301
- int16_t nxpsg;
302
- int16_t r1ptg;
303
425
  } ggml_metal_kargs_mul_mv_ext;
304
426
 
305
427
  typedef struct {
428
+ int32_t ne02;
306
429
  int32_t ne10;
307
430
  int32_t ne11; // n_expert_used (bcast)
308
431
  uint64_t nb11;
309
432
  uint64_t nb12;
310
- int32_t neh11; // n_tokens
311
- uint64_t nbh11;
433
+ int32_t ne21; // n_tokens
312
434
  int32_t ne20; // n_expert_used
313
435
  uint64_t nb21;
314
436
  } ggml_metal_kargs_mul_mm_id_map0;
315
437
 
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
438
  typedef struct {
328
439
  int32_t ne00;
329
440
  int32_t ne02;
330
441
  uint64_t nb01;
331
442
  uint64_t nb02;
332
443
  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;
444
+ int32_t ne11;
445
+ uint64_t nb10;
446
+ uint64_t nb11;
447
+ uint64_t nb12;
448
+ uint64_t nb13;
449
+ int32_t ne20;
450
+ int32_t ne21;
451
+ int32_t ne0;
452
+ int32_t ne1;
340
453
  int16_t r2;
341
454
  int16_t r3;
342
455
  } ggml_metal_kargs_mul_mm_id;
@@ -361,22 +474,26 @@ typedef struct {
361
474
  int32_t ne0;
362
475
  int32_t ne1;
363
476
  uint64_t nb1;
477
+ int32_t nr0;
364
478
  } ggml_metal_kargs_mul_mv_id;
365
479
 
480
+ // NORM
481
+ // RMS_NORM
366
482
  typedef struct {
367
483
  int32_t ne00;
368
- int32_t ne00_4;
369
- uint64_t nb01;
484
+ int32_t ne00_t;
485
+ uint64_t nb1;
486
+ uint64_t nb2;
487
+ uint64_t nb3;
370
488
  float eps;
489
+ int32_t nef1[3];
490
+ int32_t nef2[3];
491
+ int32_t nef3[3];
492
+ uint64_t nbf1[3];
493
+ uint64_t nbf2[3];
494
+ uint64_t nbf3[3];
371
495
  } ggml_metal_kargs_norm;
372
496
 
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
497
  typedef struct {
381
498
  int32_t ne00;
382
499
  int32_t ne00_4;
@@ -391,7 +508,7 @@ typedef struct {
391
508
  uint64_t nb00;
392
509
  uint64_t nb01;
393
510
  uint64_t nb02;
394
- int32_t n_groups;
511
+ int32_t ngrp;
395
512
  float eps;
396
513
  } ggml_metal_kargs_group_norm;
397
514
 
@@ -404,6 +521,49 @@ typedef struct {
404
521
  uint64_t nb1;
405
522
  } ggml_metal_kargs_conv_transpose_1d;
406
523
 
524
+ typedef struct {
525
+ int32_t IC;
526
+ int32_t IH;
527
+ int32_t IW;
528
+ int32_t KH;
529
+ int32_t KW;
530
+ int32_t OC;
531
+ int32_t s0;
532
+ uint64_t nb0;
533
+ uint64_t nb1;
534
+ uint64_t nb2;
535
+ } ggml_metal_kargs_conv_transpose_2d;
536
+
537
+ typedef struct {
538
+ uint64_t nb00;
539
+ uint64_t nb01;
540
+ uint64_t nb02;
541
+ uint64_t nb03;
542
+ uint64_t nb10;
543
+ uint64_t nb11;
544
+ uint64_t nb12;
545
+ uint64_t nb13;
546
+ uint64_t nb0;
547
+ uint64_t nb1;
548
+ uint64_t nb2;
549
+ uint64_t nb3;
550
+ int32_t IW;
551
+ int32_t IH;
552
+ int32_t KW;
553
+ int32_t KH;
554
+ int32_t IC;
555
+ int32_t OC;
556
+ int32_t OW;
557
+ int32_t OH;
558
+ int32_t N;
559
+ int32_t s0;
560
+ int32_t s1;
561
+ int32_t p0;
562
+ int32_t p1;
563
+ int32_t d0;
564
+ int32_t d1;
565
+ } ggml_metal_kargs_conv_2d;
566
+
407
567
  typedef struct {
408
568
  uint64_t ofs0;
409
569
  uint64_t ofs1;
@@ -431,8 +591,14 @@ typedef struct{
431
591
  uint64_t nb1;
432
592
  int32_t i00;
433
593
  int32_t i10;
594
+ float alpha;
595
+ float limit;
434
596
  } ggml_metal_kargs_glu;
435
597
 
598
+ typedef struct {
599
+ uint64_t np;
600
+ } ggml_metal_kargs_sum;
601
+
436
602
  typedef struct {
437
603
  int64_t ne00;
438
604
  int64_t ne01;
@@ -442,14 +608,6 @@ typedef struct {
442
608
  uint64_t nb01;
443
609
  uint64_t nb02;
444
610
  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
611
  int64_t ne0;
454
612
  int64_t ne1;
455
613
  int64_t ne2;
@@ -464,19 +622,64 @@ typedef struct {
464
622
  int64_t ne00;
465
623
  int64_t ne01;
466
624
  int64_t ne02;
625
+ int64_t ne03;
626
+ uint64_t nb00;
627
+ uint64_t nb01;
628
+ uint64_t nb02;
629
+ uint64_t nb03;
630
+ int64_t net0;
631
+ int64_t net1;
632
+ int64_t net2;
633
+ int64_t net3;
634
+ uint64_t nbt0;
635
+ uint64_t nbt1;
636
+ uint64_t nbt2;
637
+ uint64_t nbt3;
638
+ bool outb;
639
+ } ggml_metal_kargs_cumsum_blk;
640
+
641
+ typedef struct {
642
+ int64_t ne00;
643
+ int64_t ne01;
644
+ int64_t ne02;
645
+ int64_t ne03;
646
+ uint64_t nb00;
647
+ uint64_t nb01;
648
+ uint64_t nb02;
649
+ uint64_t nb03;
650
+ int64_t net0;
651
+ int64_t net1;
652
+ int64_t net2;
653
+ int64_t net3;
654
+ uint64_t nbt0;
655
+ uint64_t nbt1;
656
+ uint64_t nbt2;
657
+ uint64_t nbt3;
658
+ } ggml_metal_kargs_cumsum_add;
659
+
660
+ typedef struct {
661
+ int32_t ne00;
662
+ int32_t ne01;
663
+ int32_t ne02;
664
+ uint64_t nb01;
665
+ uint64_t nb02;
666
+ uint64_t nb03;
667
+ int32_t ne11;
668
+ int32_t ne12;
669
+ int32_t ne13;
670
+ uint64_t nb11;
671
+ uint64_t nb12;
672
+ uint64_t nb13;
673
+ uint64_t nb1;
674
+ uint64_t nb2;
675
+ uint64_t nb3;
467
676
  float scale;
468
677
  float max_bias;
469
678
  float m0;
470
679
  float m1;
471
- uint32_t n_head_log2;
680
+ int32_t n_head_log2;
472
681
  } ggml_metal_kargs_soft_max;
473
682
 
474
- typedef struct {
475
- int64_t ne00;
476
- int64_t ne01;
477
- int n_past;
478
- } ggml_metal_kargs_diag_mask_inf;
479
-
480
683
  typedef struct {
481
684
  int64_t ne00;
482
685
  int64_t ne01;
@@ -499,37 +702,50 @@ typedef struct {
499
702
  typedef struct {
500
703
  int64_t d_state;
501
704
  int64_t d_inner;
705
+ int64_t n_head;
706
+ int64_t n_group;
502
707
  int64_t n_seq_tokens;
503
708
  int64_t n_seqs;
709
+ uint64_t s_off;
504
710
  uint64_t nb00;
505
711
  uint64_t nb01;
506
712
  uint64_t nb02;
713
+ uint64_t nb03;
507
714
  uint64_t nb10;
508
715
  uint64_t nb11;
509
716
  uint64_t nb12;
717
+ uint64_t ns12;
510
718
  uint64_t nb13;
511
719
  uint64_t nb20;
512
720
  uint64_t nb21;
721
+ uint64_t ns21;
513
722
  uint64_t nb22;
514
- uint64_t nb30;
723
+ int64_t ne30;
515
724
  uint64_t nb31;
516
- uint64_t nb40;
517
725
  uint64_t nb41;
518
726
  uint64_t nb42;
519
- uint64_t nb50;
727
+ uint64_t ns42;
728
+ uint64_t nb43;
520
729
  uint64_t nb51;
521
730
  uint64_t nb52;
731
+ uint64_t ns52;
732
+ uint64_t nb53;
733
+ uint64_t nb0;
522
734
  } ggml_metal_kargs_ssm_scan;
523
735
 
524
736
  typedef struct {
525
- int64_t ne00;
737
+ int32_t ne00t;
738
+ int32_t ne00;
526
739
  uint64_t nb01;
527
740
  uint64_t nb02;
528
- int64_t ne10;
741
+ uint64_t nb03;
742
+ int32_t ne10;
529
743
  uint64_t nb10;
530
744
  uint64_t nb11;
745
+ uint64_t nb12;
531
746
  uint64_t nb1;
532
747
  uint64_t nb2;
748
+ uint64_t nb3;
533
749
  } ggml_metal_kargs_get_rows;
534
750
 
535
751
  typedef struct {
@@ -622,16 +838,82 @@ typedef struct {
622
838
  } ggml_metal_kargs_leaky_relu;
623
839
 
624
840
  typedef struct {
625
- int64_t ncols;
626
- int64_t ncols_pad;
841
+ int32_t ne00;
842
+ int32_t ne01;
843
+ int32_t ne02;
844
+ int32_t ne03;
845
+ uint64_t nb00;
846
+ uint64_t nb01;
847
+ uint64_t nb02;
848
+ uint64_t nb03;
849
+ int32_t ne0;
850
+ int32_t ne1;
851
+ int32_t ne2;
852
+ int32_t ne3;
853
+ uint64_t nb0;
854
+ uint64_t nb1;
855
+ uint64_t nb2;
856
+ uint64_t nb3;
857
+ } ggml_metal_kargs_tri;
858
+
859
+ typedef struct {
860
+ int32_t ne00;
861
+ int32_t ne01;
862
+ int32_t ne02;
863
+ int32_t ne03;
864
+ uint64_t nb00;
865
+ uint64_t nb01;
866
+ uint64_t nb02;
867
+ uint64_t nb03;
868
+ int32_t ne0;
869
+ int32_t ne1;
870
+ int32_t ne2;
871
+ int32_t ne3;
872
+ int32_t top_k;
627
873
  } ggml_metal_kargs_argsort;
628
874
 
875
+ typedef struct {
876
+ int64_t ne00;
877
+ int64_t ne01;
878
+ int64_t ne02;
879
+ int64_t ne03;
880
+ uint64_t nb00;
881
+ uint64_t nb01;
882
+ uint64_t nb02;
883
+ uint64_t nb03;
884
+ int32_t ne0;
885
+ int32_t ne1;
886
+ int32_t ne2;
887
+ int32_t ne3;
888
+ int32_t top_k;
889
+ int32_t len;
890
+ } ggml_metal_kargs_argsort_merge;
891
+
629
892
  typedef struct {
630
893
  int64_t ne0;
631
894
  float start;
632
895
  float step;
633
896
  } ggml_metal_kargs_arange;
634
897
 
898
+ typedef struct {
899
+ int64_t val;
900
+ } ggml_metal_kargs_memset;
901
+
902
+ typedef struct {
903
+ int32_t ne00;
904
+ int32_t ne01;
905
+ int32_t ne02;
906
+ int32_t ne03;
907
+ uint64_t nb00;
908
+ uint64_t nb01;
909
+ uint64_t nb02;
910
+ uint64_t nb03;
911
+ uint64_t nb10;
912
+ uint64_t nb11;
913
+ uint64_t nb12;
914
+ uint64_t nb13;
915
+ } ggml_metal_kargs_count_equal;
916
+
635
917
  typedef struct {
636
918
  int32_t k0;
637
919
  int32_t k1;
@@ -643,7 +925,20 @@ typedef struct {
643
925
  int64_t IW;
644
926
  int64_t OH;
645
927
  int64_t OW;
646
- int64_t parallel_elements;
928
+ int64_t np;
647
929
  } ggml_metal_kargs_pool_2d;
648
930
 
931
+ typedef struct {
932
+ int64_t ne00;
933
+ uint64_t nb01;
934
+ } ggml_metal_kargs_argmax;
935
+
936
+ typedef struct {
937
+ int64_t np;
938
+ } ggml_metal_kargs_opt_step_adamw;
939
+
940
+ typedef struct {
941
+ int64_t np;
942
+ } ggml_metal_kargs_opt_step_sgd;
943
+
649
944
  #endif // GGML_METAL_IMPL