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
data/ext/sources/ggml/src/ggml-sycl/gla.cpp CHANGED
@@ -11,13 +11,13 @@ static void gated_linear_attn_f32_kernel(const dpct::queue_ptr stream, u_int B,
11
11
  const u_int n_seq_tokens = T / B;
12
12
  sycl::range<1> block_dims((C / H));
13
13
  sycl::range<1> grid_dims((B * H));
14
- sycl_launch(stream, [&](sycl::handler & cgh) {
14
+ stream->submit([&](sycl::handler & cgh) {
15
15
  /* local memory accessors*/
16
16
  auto _k = sycl::local_accessor<float, 1>(sycl::range<1>(head_size), cgh);
17
17
  auto _r = sycl::local_accessor<float, 1>(sycl::range<1>(head_size), cgh);
18
18
  auto _td = sycl::local_accessor<float, 1>(sycl::range<1>(head_size), cgh);
19
19
 
20
- sycl_parallel_for<1>(cgh, sycl::nd_range<1>(grid_dims * block_dims, block_dims), [=](sycl::nd_item<1> item) {
20
+ cgh.parallel_for(sycl::nd_range<1>(grid_dims * block_dims, block_dims), [=](sycl::nd_item<1> item) {
21
21
  u_int tid = item.get_local_id(0);
22
22
  u_int bid = item.get_group(0);
23
23
 
data/ext/sources/ggml/src/ggml-sycl/im2col.cpp CHANGED
@@ -26,7 +26,7 @@ static void im2col_kernel(const float * x, T * dst, int64_t batch_offset, int64_
26
26
 
27
27
  // make each work-item deal with more elements since sycl global range can not exceed max int
28
28
  for (int64_t i = global_id; i < pelements; i += (work_group_size * item_ct1.get_group_range(2))) {
29
- const int64_t ksize = OW * (KH > 1 ? KW : 1);
29
+ const int64_t ksize = OW * KH;
30
30
  const int64_t kx = i / ksize;
31
31
  const int64_t kd = kx * ksize;
32
32
  const int64_t ky = (i - kd) / OW;
@@ -70,7 +70,7 @@ static void im2col_sycl_internal(const float * x, T * dst, int64_t IW, int64_t I
70
70
 
71
71
  const int64_t CHW = IC * KH * KW;
72
72
 
73
- sycl_parallel_for(stream, sycl::nd_range<3>(block_nums * local_range, local_range), [=](sycl::nd_item<3> item_ct1) {
73
+ stream->parallel_for(sycl::nd_range<3>(block_nums * local_range, local_range), [=](sycl::nd_item<3> item_ct1) {
74
74
  im2col_kernel<T>(x, dst, batch_offset, offset_delta, IC, IW, IH, OH, OW, KW, KH, parallel_elements, CHW, s0, s1,
75
75
  p0, p1, d0, d1, item_ct1);
76
76
  });
data/ext/sources/ggml/src/ggml-sycl/mmq.cpp CHANGED
@@ -1818,7 +1818,7 @@ static void ggml_mul_mat_q4_0_q8_1_sycl(const void *vx, const void *vy,
1818
1818
  dpct::has_capability_or_fail(stream->get_device(),
1819
1819
  {sycl::aspect::fp16});
1820
1820
 
1821
- sycl_launch(stream, [&](sycl::handler & cgh) {
1821
+ stream->submit([&](sycl::handler &cgh) {
1822
1822
  sycl::local_accessor<int, 1> tile_x_qs_q4_0_acc_ct1(
1823
1823
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
1824
1824
  sycl::local_accessor<float, 1> tile_x_d_q4_0_acc_ct1(
@@ -1829,8 +1829,9 @@ static void ggml_mul_mat_q4_0_q8_1_sycl(const void *vx, const void *vy,
1829
1829
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
1830
1830
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
1831
1831
 
1832
- sycl_parallel_for(
1833
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
1832
+ cgh.parallel_for(
1833
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
1834
+ [=](sycl::nd_item<3> item_ct1) {
1834
1835
  mul_mat_q4_0<need_check>(
1835
1836
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
1836
1837
  nrows_dst, item_ct1,
@@ -1852,7 +1853,7 @@ static void ggml_mul_mat_q4_0_q8_1_sycl(const void *vx, const void *vy,
1852
1853
  dpct::has_capability_or_fail(stream->get_device(),
1853
1854
  {sycl::aspect::fp16});
1854
1855
 
1855
- sycl_launch(stream, [&](sycl::handler & cgh) {
1856
+ stream->submit([&](sycl::handler &cgh) {
1856
1857
  sycl::local_accessor<int, 1> tile_x_qs_q4_0_acc_ct1(
1857
1858
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
1858
1859
  sycl::local_accessor<float, 1> tile_x_d_q4_0_acc_ct1(
@@ -1863,8 +1864,9 @@ static void ggml_mul_mat_q4_0_q8_1_sycl(const void *vx, const void *vy,
1863
1864
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
1864
1865
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
1865
1866
 
1866
- sycl_parallel_for(
1867
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
1867
+ cgh.parallel_for(
1868
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
1869
+ [=](sycl::nd_item<3> item_ct1) {
1868
1870
  mul_mat_q4_0<need_check>(
1869
1871
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
1870
1872
  nrows_dst, item_ct1,
@@ -1931,7 +1933,7 @@ static void ggml_mul_mat_q4_1_q8_1_sycl(const void *vx, const void *vy,
1931
1933
  dpct::has_capability_or_fail(stream->get_device(),
1932
1934
  {sycl::aspect::fp16});
1933
1935
 
1934
- sycl_launch(stream, [&](sycl::handler & cgh) {
1936
+ stream->submit([&](sycl::handler &cgh) {
1935
1937
  sycl::local_accessor<int, 1> tile_x_qs_q4_1_acc_ct1(
1936
1938
  sycl::range<1>(mmq_y * (WARP_SIZE) + +mmq_y), cgh);
1937
1939
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q4_1_acc_ct1(
@@ -1942,8 +1944,9 @@ static void ggml_mul_mat_q4_1_q8_1_sycl(const void *vx, const void *vy,
1942
1944
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
1943
1945
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
1944
1946
 
1945
- sycl_parallel_for(
1946
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
1947
+ cgh.parallel_for(
1948
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
1949
+ [=](sycl::nd_item<3> item_ct1) {
1947
1950
  mul_mat_q4_1<need_check>(
1948
1951
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
1949
1952
  nrows_dst, item_ct1,
@@ -1965,7 +1968,7 @@ static void ggml_mul_mat_q4_1_q8_1_sycl(const void *vx, const void *vy,
1965
1968
  dpct::has_capability_or_fail(stream->get_device(),
1966
1969
  {sycl::aspect::fp16});
1967
1970
 
1968
- sycl_launch(stream, [&](sycl::handler & cgh) {
1971
+ stream->submit([&](sycl::handler &cgh) {
1969
1972
  sycl::local_accessor<int, 1> tile_x_qs_q4_1_acc_ct1(
1970
1973
  sycl::range<1>(mmq_y * (WARP_SIZE) + +mmq_y), cgh);
1971
1974
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q4_1_acc_ct1(
@@ -1976,8 +1979,9 @@ static void ggml_mul_mat_q4_1_q8_1_sycl(const void *vx, const void *vy,
1976
1979
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
1977
1980
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
1978
1981
 
1979
- sycl_parallel_for(
1980
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
1982
+ cgh.parallel_for(
1983
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
1984
+ [=](sycl::nd_item<3> item_ct1) {
1981
1985
  mul_mat_q4_1<need_check>(
1982
1986
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
1983
1987
  nrows_dst, item_ct1,
@@ -2044,7 +2048,7 @@ static void ggml_mul_mat_q5_0_q8_1_sycl(const void *vx, const void *vy,
2044
2048
  dpct::has_capability_or_fail(stream->get_device(),
2045
2049
  {sycl::aspect::fp16});
2046
2050
 
2047
- sycl_launch(stream, [&](sycl::handler & cgh) {
2051
+ stream->submit([&](sycl::handler &cgh) {
2048
2052
  sycl::local_accessor<int, 1> tile_x_ql_q5_0_acc_ct1(
2049
2053
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2050
2054
  sycl::local_accessor<float, 1> tile_x_d_q5_0_acc_ct1(
@@ -2055,8 +2059,9 @@ static void ggml_mul_mat_q5_0_q8_1_sycl(const void *vx, const void *vy,
2055
2059
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2056
2060
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2057
2061
 
2058
- sycl_parallel_for(
2059
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2062
+ cgh.parallel_for(
2063
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2064
+ [=](sycl::nd_item<3> item_ct1) {
2060
2065
  mul_mat_q5_0<need_check>(
2061
2066
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2062
2067
  nrows_dst, item_ct1,
@@ -2078,7 +2083,7 @@ static void ggml_mul_mat_q5_0_q8_1_sycl(const void *vx, const void *vy,
2078
2083
  dpct::has_capability_or_fail(stream->get_device(),
2079
2084
  {sycl::aspect::fp16});
2080
2085
 
2081
- sycl_launch(stream, [&](sycl::handler & cgh) {
2086
+ stream->submit([&](sycl::handler &cgh) {
2082
2087
  sycl::local_accessor<int, 1> tile_x_ql_q5_0_acc_ct1(
2083
2088
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2084
2089
  sycl::local_accessor<float, 1> tile_x_d_q5_0_acc_ct1(
@@ -2089,8 +2094,9 @@ static void ggml_mul_mat_q5_0_q8_1_sycl(const void *vx, const void *vy,
2089
2094
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2090
2095
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2091
2096
 
2092
- sycl_parallel_for(
2093
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2097
+ cgh.parallel_for(
2098
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2099
+ [=](sycl::nd_item<3> item_ct1) {
2094
2100
  mul_mat_q5_0<need_check>(
2095
2101
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2096
2102
  nrows_dst, item_ct1,
@@ -2157,7 +2163,7 @@ static void ggml_mul_mat_q5_1_q8_1_sycl(const void *vx, const void *vy,
2157
2163
  dpct::has_capability_or_fail(stream->get_device(),
2158
2164
  {sycl::aspect::fp16});
2159
2165
 
2160
- sycl_launch(stream, [&](sycl::handler & cgh) {
2166
+ stream->submit([&](sycl::handler &cgh) {
2161
2167
  sycl::local_accessor<int, 1> tile_x_ql_q5_1_acc_ct1(
2162
2168
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2163
2169
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q5_1_acc_ct1(
@@ -2168,8 +2174,9 @@ static void ggml_mul_mat_q5_1_q8_1_sycl(const void *vx, const void *vy,
2168
2174
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2169
2175
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2170
2176
 
2171
- sycl_parallel_for(
2172
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2177
+ cgh.parallel_for(
2178
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2179
+ [=](sycl::nd_item<3> item_ct1) {
2173
2180
  mul_mat_q5_1<need_check>(
2174
2181
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2175
2182
  nrows_dst, item_ct1,
@@ -2191,7 +2198,7 @@ static void ggml_mul_mat_q5_1_q8_1_sycl(const void *vx, const void *vy,
2191
2198
  dpct::has_capability_or_fail(stream->get_device(),
2192
2199
  {sycl::aspect::fp16});
2193
2200
 
2194
- sycl_launch(stream, [&](sycl::handler & cgh) {
2201
+ stream->submit([&](sycl::handler &cgh) {
2195
2202
  sycl::local_accessor<int, 1> tile_x_ql_q5_1_acc_ct1(
2196
2203
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2197
2204
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q5_1_acc_ct1(
@@ -2202,8 +2209,9 @@ static void ggml_mul_mat_q5_1_q8_1_sycl(const void *vx, const void *vy,
2202
2209
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2203
2210
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2204
2211
 
2205
- sycl_parallel_for(
2206
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2212
+ cgh.parallel_for(
2213
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2214
+ [=](sycl::nd_item<3> item_ct1) {
2207
2215
  mul_mat_q5_1<need_check>(
2208
2216
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2209
2217
  nrows_dst, item_ct1,
@@ -2270,7 +2278,7 @@ static void ggml_mul_mat_q8_0_q8_1_sycl(const void *vx, const void *vy,
2270
2278
  dpct::has_capability_or_fail(stream->get_device(),
2271
2279
  {sycl::aspect::fp16});
2272
2280
 
2273
- sycl_launch(stream, [&](sycl::handler & cgh) {
2281
+ stream->submit([&](sycl::handler &cgh) {
2274
2282
  sycl::local_accessor<int, 1> tile_x_qs_q8_0_acc_ct1(
2275
2283
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2276
2284
  sycl::local_accessor<float, 1> tile_x_d_q8_0_acc_ct1(
@@ -2281,8 +2289,9 @@ static void ggml_mul_mat_q8_0_q8_1_sycl(const void *vx, const void *vy,
2281
2289
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2282
2290
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2283
2291
 
2284
- sycl_parallel_for(
2285
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2292
+ cgh.parallel_for(
2293
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2294
+ [=](sycl::nd_item<3> item_ct1) {
2286
2295
  mul_mat_q8_0<need_check>(
2287
2296
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2288
2297
  nrows_dst, item_ct1,
@@ -2304,7 +2313,7 @@ static void ggml_mul_mat_q8_0_q8_1_sycl(const void *vx, const void *vy,
2304
2313
  dpct::has_capability_or_fail(stream->get_device(),
2305
2314
  {sycl::aspect::fp16});
2306
2315
 
2307
- sycl_launch(stream, [&](sycl::handler & cgh) {
2316
+ stream->submit([&](sycl::handler &cgh) {
2308
2317
  sycl::local_accessor<int, 1> tile_x_qs_q8_0_acc_ct1(
2309
2318
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2310
2319
  sycl::local_accessor<float, 1> tile_x_d_q8_0_acc_ct1(
@@ -2315,8 +2324,9 @@ static void ggml_mul_mat_q8_0_q8_1_sycl(const void *vx, const void *vy,
2315
2324
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2316
2325
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2317
2326
 
2318
- sycl_parallel_for(
2319
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2327
+ cgh.parallel_for(
2328
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2329
+ [=](sycl::nd_item<3> item_ct1) {
2320
2330
  mul_mat_q8_0<need_check>(
2321
2331
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2322
2332
  nrows_dst, item_ct1,
@@ -2383,7 +2393,7 @@ static void ggml_mul_mat_q2_K_q8_1_sycl(const void *vx, const void *vy,
2383
2393
  dpct::has_capability_or_fail(stream->get_device(),
2384
2394
  {sycl::aspect::fp16});
2385
2395
 
2386
- sycl_launch(stream, [&](sycl::handler & cgh) {
2396
+ stream->submit([&](sycl::handler &cgh) {
2387
2397
  sycl::local_accessor<int, 1> tile_x_ql_q2_K_acc_ct1(
2388
2398
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2389
2399
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q2_K_acc_ct1(
@@ -2396,8 +2406,9 @@ static void ggml_mul_mat_q2_K_q8_1_sycl(const void *vx, const void *vy,
2396
2406
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2397
2407
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2398
2408
 
2399
- sycl_parallel_for(
2400
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2409
+ cgh.parallel_for(
2410
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2411
+ [=](sycl::nd_item<3> item_ct1) {
2401
2412
  mul_mat_q2_K<need_check>(
2402
2413
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2403
2414
  nrows_dst, item_ct1,
@@ -2420,7 +2431,7 @@ static void ggml_mul_mat_q2_K_q8_1_sycl(const void *vx, const void *vy,
2420
2431
  dpct::has_capability_or_fail(stream->get_device(),
2421
2432
  {sycl::aspect::fp16});
2422
2433
 
2423
- sycl_launch(stream, [&](sycl::handler & cgh) {
2434
+ stream->submit([&](sycl::handler &cgh) {
2424
2435
  sycl::local_accessor<int, 1> tile_x_ql_q2_K_acc_ct1(
2425
2436
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2426
2437
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q2_K_acc_ct1(
@@ -2433,8 +2444,9 @@ static void ggml_mul_mat_q2_K_q8_1_sycl(const void *vx, const void *vy,
2433
2444
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2434
2445
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2435
2446
 
2436
- sycl_parallel_for(
2437
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2447
+ cgh.parallel_for(
2448
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2449
+ [=](sycl::nd_item<3> item_ct1) {
2438
2450
  mul_mat_q2_K<need_check>(
2439
2451
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2440
2452
  nrows_dst, item_ct1,
@@ -2504,7 +2516,7 @@ static void ggml_mul_mat_q3_K_q8_1_sycl(const void *vx, const void *vy,
2504
2516
  dpct::has_capability_or_fail(stream->get_device(),
2505
2517
  {sycl::aspect::fp16});
2506
2518
 
2507
- sycl_launch(stream, [&](sycl::handler & cgh) {
2519
+ stream->submit([&](sycl::handler &cgh) {
2508
2520
  sycl::local_accessor<int, 1> tile_x_ql_q3_K_acc_ct1(
2509
2521
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2510
2522
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q3_K_acc_ct1(
@@ -2519,8 +2531,9 @@ static void ggml_mul_mat_q3_K_q8_1_sycl(const void *vx, const void *vy,
2519
2531
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2520
2532
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2521
2533
 
2522
- sycl_parallel_for(
2523
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2534
+ cgh.parallel_for(
2535
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2536
+ [=](sycl::nd_item<3> item_ct1) {
2524
2537
  mul_mat_q3_K<need_check>(
2525
2538
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2526
2539
  nrows_dst, item_ct1,
@@ -2544,7 +2557,7 @@ static void ggml_mul_mat_q3_K_q8_1_sycl(const void *vx, const void *vy,
2544
2557
  dpct::has_capability_or_fail(stream->get_device(),
2545
2558
  {sycl::aspect::fp16});
2546
2559
 
2547
- sycl_launch(stream, [&](sycl::handler & cgh) {
2560
+ stream->submit([&](sycl::handler &cgh) {
2548
2561
  sycl::local_accessor<int, 1> tile_x_ql_q3_K_acc_ct1(
2549
2562
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2550
2563
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q3_K_acc_ct1(
@@ -2559,8 +2572,9 @@ static void ggml_mul_mat_q3_K_q8_1_sycl(const void *vx, const void *vy,
2559
2572
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2560
2573
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2561
2574
 
2562
- sycl_parallel_for(
2563
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2575
+ cgh.parallel_for(
2576
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2577
+ [=](sycl::nd_item<3> item_ct1) {
2564
2578
  mul_mat_q3_K<need_check>(
2565
2579
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2566
2580
  nrows_dst, item_ct1,
@@ -2630,7 +2644,7 @@ static void ggml_mul_mat_q4_K_q8_1_sycl(const void *vx, const void *vy,
2630
2644
  dpct::has_capability_or_fail(stream->get_device(),
2631
2645
  {sycl::aspect::fp16});
2632
2646
 
2633
- sycl_launch(stream, [&](sycl::handler & cgh) {
2647
+ stream->submit([&](sycl::handler &cgh) {
2634
2648
  sycl::local_accessor<int, 1> tile_x_ql_q4_K_acc_ct1(
2635
2649
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2636
2650
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q4_K_acc_ct1(
@@ -2643,8 +2657,9 @@ static void ggml_mul_mat_q4_K_q8_1_sycl(const void *vx, const void *vy,
2643
2657
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2644
2658
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2645
2659
 
2646
- sycl_parallel_for(
2647
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2660
+ cgh.parallel_for(
2661
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2662
+ [=](sycl::nd_item<3> item_ct1) {
2648
2663
  mul_mat_q4_K<need_check>(
2649
2664
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2650
2665
  nrows_dst, item_ct1,
@@ -2667,7 +2682,7 @@ static void ggml_mul_mat_q4_K_q8_1_sycl(const void *vx, const void *vy,
2667
2682
  dpct::has_capability_or_fail(stream->get_device(),
2668
2683
  {sycl::aspect::fp16});
2669
2684
 
2670
- sycl_launch(stream, [&](sycl::handler & cgh) {
2685
+ stream->submit([&](sycl::handler &cgh) {
2671
2686
  sycl::local_accessor<int, 1> tile_x_ql_q4_K_acc_ct1(
2672
2687
  sycl::range<1>(mmq_y * (WARP_SIZE) + mmq_y), cgh);
2673
2688
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q4_K_acc_ct1(
@@ -2680,8 +2695,9 @@ static void ggml_mul_mat_q4_K_q8_1_sycl(const void *vx, const void *vy,
2680
2695
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2681
2696
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2682
2697
 
2683
- sycl_parallel_for(
2684
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2698
+ cgh.parallel_for(
2699
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2700
+ [=](sycl::nd_item<3> item_ct1) {
2685
2701
  mul_mat_q4_K<need_check>(
2686
2702
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2687
2703
  nrows_dst, item_ct1,
@@ -2749,7 +2765,7 @@ static void ggml_mul_mat_q5_K_q8_1_sycl(const void *vx, const void *vy,
2749
2765
  dpct::has_capability_or_fail(stream->get_device(),
2750
2766
  {sycl::aspect::fp16});
2751
2767
 
2752
- sycl_launch(stream, [&](sycl::handler & cgh) {
2768
+ stream->submit([&](sycl::handler &cgh) {
2753
2769
  sycl::local_accessor<int, 1> tile_x_ql_q5_K_acc_ct1(
2754
2770
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2755
2771
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q5_K_acc_ct1(
@@ -2762,8 +2778,9 @@ static void ggml_mul_mat_q5_K_q8_1_sycl(const void *vx, const void *vy,
2762
2778
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2763
2779
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2764
2780
 
2765
- sycl_parallel_for(
2766
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2781
+ cgh.parallel_for(
2782
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2783
+ [=](sycl::nd_item<3> item_ct1) {
2767
2784
  mul_mat_q5_K<need_check>(
2768
2785
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2769
2786
  nrows_dst, item_ct1,
@@ -2786,7 +2803,7 @@ static void ggml_mul_mat_q5_K_q8_1_sycl(const void *vx, const void *vy,
2786
2803
  dpct::has_capability_or_fail(stream->get_device(),
2787
2804
  {sycl::aspect::fp16});
2788
2805
 
2789
- sycl_launch(stream, [&](sycl::handler & cgh) {
2806
+ stream->submit([&](sycl::handler &cgh) {
2790
2807
  sycl::local_accessor<int, 1> tile_x_ql_q5_K_acc_ct1(
2791
2808
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2792
2809
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_q5_K_acc_ct1(
@@ -2799,8 +2816,9 @@ static void ggml_mul_mat_q5_K_q8_1_sycl(const void *vx, const void *vy,
2799
2816
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2800
2817
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2801
2818
 
2802
- sycl_parallel_for(
2803
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2819
+ cgh.parallel_for(
2820
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2821
+ [=](sycl::nd_item<3> item_ct1) {
2804
2822
  mul_mat_q5_K<need_check>(
2805
2823
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2806
2824
  nrows_dst, item_ct1,
@@ -2868,7 +2886,7 @@ static void ggml_mul_mat_q6_K_q8_1_sycl(const void *vx, const void *vy,
2868
2886
  dpct::has_capability_or_fail(stream->get_device(),
2869
2887
  {sycl::aspect::fp16});
2870
2888
 
2871
- sycl_launch(stream, [&](sycl::handler & cgh) {
2889
+ stream->submit([&](sycl::handler &cgh) {
2872
2890
  sycl::local_accessor<int, 1> tile_x_ql_acc_ct1(
2873
2891
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2874
2892
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_acc_ct1(
@@ -2881,8 +2899,9 @@ static void ggml_mul_mat_q6_K_q8_1_sycl(const void *vx, const void *vy,
2881
2899
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2882
2900
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2883
2901
 
2884
- sycl_parallel_for(
2885
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2902
+ cgh.parallel_for(
2903
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2904
+ [=](sycl::nd_item<3> item_ct1) {
2886
2905
  mul_mat_q6_K<need_check>(
2887
2906
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2888
2907
  nrows_dst, item_ct1,
@@ -2905,7 +2924,7 @@ static void ggml_mul_mat_q6_K_q8_1_sycl(const void *vx, const void *vy,
2905
2924
  dpct::has_capability_or_fail(stream->get_device(),
2906
2925
  {sycl::aspect::fp16});
2907
2926
 
2908
- sycl_launch(stream, [&](sycl::handler & cgh) {
2927
+ stream->submit([&](sycl::handler &cgh) {
2909
2928
  sycl::local_accessor<int, 1> tile_x_ql_acc_ct1(
2910
2929
  sycl::range<1>(mmq_y * (2 * WARP_SIZE) + mmq_y), cgh);
2911
2930
  sycl::local_accessor<sycl::half2, 1> tile_x_dm_acc_ct1(
@@ -2918,8 +2937,9 @@ static void ggml_mul_mat_q6_K_q8_1_sycl(const void *vx, const void *vy,
2918
2937
  sycl::local_accessor<sycl::half2, 1> tile_y_ds_acc_ct1(
2919
2938
  sycl::range<1>(mmq_x * WARP_SIZE / QI8_1), cgh);
2920
2939
 
2921
- sycl_parallel_for(
2922
- cgh, sycl::nd_range<3>(block_nums * block_dims, block_dims), [=](sycl::nd_item<3> item_ct1) {
2940
+ cgh.parallel_for(
2941
+ sycl::nd_range<3>(block_nums * block_dims, block_dims),
2942
+ [=](sycl::nd_item<3> item_ct1) {
2923
2943
  mul_mat_q6_K<need_check>(
2924
2944
  vx, vy, dst, ncols_x, nrows_x, ncols_y, nrows_y,
2925
2945
  nrows_dst, item_ct1,