whispercpp 1.3.4 → 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 (630) 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 +47 -23
  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/addon.node/vad-example.js +2 -2
  23. data/ext/sources/examples/cli/cli.cpp +121 -112
  24. data/ext/sources/examples/lsp/CMakeLists.txt +2 -1
  25. data/ext/sources/examples/quantize/CMakeLists.txt +2 -1
  26. data/ext/sources/examples/server/server.cpp +10 -11
  27. data/ext/sources/examples/talk-llama/CMakeLists.txt +5 -1
  28. data/ext/sources/examples/talk-llama/llama-adapter.cpp +12 -3
  29. data/ext/sources/examples/talk-llama/llama-adapter.h +7 -1
  30. data/ext/sources/examples/talk-llama/llama-arch.cpp +2046 -1974
  31. data/ext/sources/examples/talk-llama/llama-arch.h +67 -2
  32. data/ext/sources/examples/talk-llama/llama-batch.cpp +75 -33
  33. data/ext/sources/examples/talk-llama/llama-batch.h +17 -4
  34. data/ext/sources/examples/talk-llama/llama-chat.cpp +79 -3
  35. data/ext/sources/examples/talk-llama/llama-chat.h +4 -0
  36. data/ext/sources/examples/talk-llama/llama-context.cpp +775 -78
  37. data/ext/sources/examples/talk-llama/llama-context.h +57 -9
  38. data/ext/sources/examples/talk-llama/llama-cparams.h +1 -0
  39. data/ext/sources/examples/talk-llama/llama-grammar.cpp +288 -53
  40. data/ext/sources/examples/talk-llama/llama-grammar.h +22 -1
  41. data/ext/sources/examples/talk-llama/llama-graph.cpp +381 -64
  42. data/ext/sources/examples/talk-llama/llama-graph.h +103 -13
  43. data/ext/sources/examples/talk-llama/llama-hparams.cpp +26 -2
  44. data/ext/sources/examples/talk-llama/llama-hparams.h +41 -10
  45. data/ext/sources/examples/talk-llama/llama-impl.cpp +7 -3
  46. data/ext/sources/examples/talk-llama/llama-impl.h +1 -1
  47. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +5 -3
  48. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +145 -65
  49. data/ext/sources/examples/talk-llama/llama-kv-cache.h +22 -7
  50. data/ext/sources/examples/talk-llama/llama-kv-cells.h +44 -2
  51. data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +12 -10
  52. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +32 -19
  53. data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +2 -2
  54. data/ext/sources/examples/talk-llama/llama-mmap.cpp +172 -37
  55. data/ext/sources/examples/talk-llama/llama-mmap.h +8 -3
  56. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +91 -9
  57. data/ext/sources/examples/talk-llama/llama-model-loader.h +6 -0
  58. data/ext/sources/examples/talk-llama/llama-model-saver.cpp +3 -0
  59. data/ext/sources/examples/talk-llama/llama-model.cpp +1529 -13134
  60. data/ext/sources/examples/talk-llama/llama-model.h +44 -3
  61. data/ext/sources/examples/talk-llama/llama-quant.cpp +8 -23
  62. data/ext/sources/examples/talk-llama/llama-sampling.cpp +1294 -198
  63. data/ext/sources/examples/talk-llama/llama-sampling.h +19 -7
  64. data/ext/sources/examples/talk-llama/llama-vocab.cpp +133 -37
  65. data/ext/sources/examples/talk-llama/llama-vocab.h +45 -40
  66. data/ext/sources/examples/talk-llama/llama.cpp +729 -2
  67. data/ext/sources/examples/talk-llama/llama.h +152 -14
  68. data/ext/sources/examples/talk-llama/models/afmoe.cpp +191 -0
  69. data/ext/sources/examples/talk-llama/models/apertus.cpp +125 -0
  70. data/ext/sources/examples/talk-llama/models/arcee.cpp +135 -0
  71. data/ext/sources/examples/talk-llama/models/arctic.cpp +138 -0
  72. data/ext/sources/examples/talk-llama/models/arwkv7.cpp +86 -0
  73. data/ext/sources/examples/talk-llama/models/baichuan.cpp +122 -0
  74. data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +144 -0
  75. data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +135 -0
  76. data/ext/sources/examples/talk-llama/models/bert.cpp +178 -0
  77. data/ext/sources/examples/talk-llama/models/bitnet.cpp +160 -0
  78. data/ext/sources/examples/talk-llama/models/bloom.cpp +101 -0
  79. data/ext/sources/examples/talk-llama/models/chameleon.cpp +178 -0
  80. data/ext/sources/examples/talk-llama/models/chatglm.cpp +132 -0
  81. data/ext/sources/examples/talk-llama/models/codeshell.cpp +111 -0
  82. data/ext/sources/examples/talk-llama/models/cogvlm.cpp +102 -0
  83. data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +134 -0
  84. data/ext/sources/examples/talk-llama/models/command-r.cpp +122 -0
  85. data/ext/sources/examples/talk-llama/models/dbrx.cpp +123 -0
  86. data/ext/sources/examples/talk-llama/models/deci.cpp +135 -0
  87. data/ext/sources/examples/talk-llama/models/deepseek.cpp +144 -0
  88. data/ext/sources/examples/talk-llama/models/deepseek2.cpp +259 -0
  89. data/ext/sources/examples/talk-llama/models/dots1.cpp +134 -0
  90. data/ext/sources/examples/talk-llama/models/dream.cpp +105 -0
  91. data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +150 -0
  92. data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +110 -0
  93. data/ext/sources/examples/talk-llama/models/exaone.cpp +114 -0
  94. data/ext/sources/examples/talk-llama/models/exaone4.cpp +123 -0
  95. data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +113 -0
  96. data/ext/sources/examples/talk-llama/models/falcon.cpp +120 -0
  97. data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +116 -0
  98. data/ext/sources/examples/talk-llama/models/gemma.cpp +112 -0
  99. data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +128 -0
  100. data/ext/sources/examples/talk-llama/models/gemma3.cpp +155 -0
  101. data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +384 -0
  102. data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +170 -0
  103. data/ext/sources/examples/talk-llama/models/glm4.cpp +150 -0
  104. data/ext/sources/examples/talk-llama/models/gpt2.cpp +105 -0
  105. data/ext/sources/examples/talk-llama/models/gptneox.cpp +144 -0
  106. data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +196 -0
  107. data/ext/sources/examples/talk-llama/models/granite.cpp +211 -0
  108. data/ext/sources/examples/talk-llama/models/graph-context-mamba.cpp +283 -0
  109. data/ext/sources/examples/talk-llama/models/grok.cpp +159 -0
  110. data/ext/sources/examples/talk-llama/models/grovemoe.cpp +141 -0
  111. data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +132 -0
  112. data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +154 -0
  113. data/ext/sources/examples/talk-llama/models/internlm2.cpp +120 -0
  114. data/ext/sources/examples/talk-llama/models/jais.cpp +86 -0
  115. data/ext/sources/examples/talk-llama/models/jamba.cpp +106 -0
  116. data/ext/sources/examples/talk-llama/models/lfm2.cpp +175 -0
  117. data/ext/sources/examples/talk-llama/models/llada-moe.cpp +122 -0
  118. data/ext/sources/examples/talk-llama/models/llada.cpp +99 -0
  119. data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +178 -0
  120. data/ext/sources/examples/talk-llama/models/llama.cpp +168 -0
  121. data/ext/sources/examples/talk-llama/models/maincoder.cpp +117 -0
  122. data/ext/sources/examples/talk-llama/models/mamba.cpp +55 -0
  123. data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +123 -0
  124. data/ext/sources/examples/talk-llama/models/minicpm3.cpp +199 -0
  125. data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +124 -0
  126. data/ext/sources/examples/talk-llama/models/mistral3.cpp +160 -0
  127. data/ext/sources/examples/talk-llama/models/models.h +569 -0
  128. data/ext/sources/examples/talk-llama/models/modern-bert.cpp +116 -0
  129. data/ext/sources/examples/talk-llama/models/mpt.cpp +126 -0
  130. data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +150 -0
  131. data/ext/sources/examples/talk-llama/models/nemotron.cpp +122 -0
  132. data/ext/sources/examples/talk-llama/models/neo-bert.cpp +104 -0
  133. data/ext/sources/examples/talk-llama/models/olmo.cpp +121 -0
  134. data/ext/sources/examples/talk-llama/models/olmo2.cpp +150 -0
  135. data/ext/sources/examples/talk-llama/models/olmoe.cpp +124 -0
  136. data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +127 -0
  137. data/ext/sources/examples/talk-llama/models/openelm.cpp +124 -0
  138. data/ext/sources/examples/talk-llama/models/orion.cpp +123 -0
  139. data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +121 -0
  140. data/ext/sources/examples/talk-llama/models/phi2.cpp +121 -0
  141. data/ext/sources/examples/talk-llama/models/phi3.cpp +152 -0
  142. data/ext/sources/examples/talk-llama/models/plamo.cpp +110 -0
  143. data/ext/sources/examples/talk-llama/models/plamo2.cpp +316 -0
  144. data/ext/sources/examples/talk-llama/models/plamo3.cpp +128 -0
  145. data/ext/sources/examples/talk-llama/models/plm.cpp +168 -0
  146. data/ext/sources/examples/talk-llama/models/qwen.cpp +108 -0
  147. data/ext/sources/examples/talk-llama/models/qwen2.cpp +126 -0
  148. data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +151 -0
  149. data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +117 -0
  150. data/ext/sources/examples/talk-llama/models/qwen3.cpp +117 -0
  151. data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +124 -0
  152. data/ext/sources/examples/talk-llama/models/qwen3next.cpp +873 -0
  153. data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +149 -0
  154. data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +141 -0
  155. data/ext/sources/examples/talk-llama/models/refact.cpp +94 -0
  156. data/ext/sources/examples/talk-llama/models/rnd1.cpp +126 -0
  157. data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +162 -0
  158. data/ext/sources/examples/talk-llama/models/rwkv6.cpp +94 -0
  159. data/ext/sources/examples/talk-llama/models/rwkv6qwen2.cpp +86 -0
  160. data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +135 -0
  161. data/ext/sources/examples/talk-llama/models/rwkv7.cpp +90 -0
  162. data/ext/sources/examples/talk-llama/models/seed-oss.cpp +124 -0
  163. data/ext/sources/examples/talk-llama/models/smallthinker.cpp +126 -0
  164. data/ext/sources/examples/talk-llama/models/smollm3.cpp +128 -0
  165. data/ext/sources/examples/talk-llama/models/stablelm.cpp +146 -0
  166. data/ext/sources/examples/talk-llama/models/starcoder.cpp +100 -0
  167. data/ext/sources/examples/talk-llama/models/starcoder2.cpp +121 -0
  168. data/ext/sources/examples/talk-llama/models/t5-dec.cpp +166 -0
  169. data/ext/sources/examples/talk-llama/models/t5-enc.cpp +96 -0
  170. data/ext/sources/examples/talk-llama/models/wavtokenizer-dec.cpp +149 -0
  171. data/ext/sources/examples/talk-llama/models/xverse.cpp +108 -0
  172. data/ext/sources/examples/talk-llama/unicode.cpp +102 -16
  173. data/ext/sources/examples/vad-speech-segments/CMakeLists.txt +1 -1
  174. data/ext/sources/examples/whisper.wasm/index-tmpl.html +1 -1
  175. data/ext/sources/ggml/CMakeLists.txt +82 -54
  176. data/ext/sources/ggml/include/ggml-alloc.h +9 -0
  177. data/ext/sources/ggml/include/ggml-backend.h +4 -1
  178. data/ext/sources/ggml/include/ggml-cpu.h +1 -0
  179. data/ext/sources/ggml/include/ggml-hexagon.h +19 -0
  180. data/ext/sources/ggml/include/ggml-rpc.h +8 -11
  181. data/ext/sources/ggml/include/ggml-zendnn.h +22 -0
  182. data/ext/sources/ggml/include/ggml.h +190 -12
  183. data/ext/sources/ggml/src/CMakeLists.txt +82 -11
  184. data/ext/sources/ggml/src/ggml-alloc.c +124 -41
  185. data/ext/sources/ggml/src/ggml-backend-impl.h +1 -4
  186. data/ext/sources/ggml/src/ggml-backend-reg.cpp +27 -3
  187. data/ext/sources/ggml/src/ggml-backend.cpp +71 -21
  188. data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +17 -3
  189. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +5 -9
  190. data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +57 -45
  191. data/ext/sources/ggml/src/ggml-cann/acl_tensor.h +138 -47
  192. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +2179 -1696
  193. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +238 -317
  194. data/ext/sources/ggml/src/ggml-cann/common.h +283 -208
  195. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +626 -776
  196. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +156 -86
  197. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +1 -0
  198. data/ext/sources/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +4 -0
  199. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +428 -26
  200. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +1004 -0
  201. data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +4 -5
  202. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/cpu-feats.cpp +38 -0
  203. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +108 -49
  204. data/ext/sources/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
  205. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +6 -6
  206. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +50 -2
  207. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +5 -3
  208. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +195 -71
  209. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +4 -0
  210. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +573 -106
  211. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +33 -44
  212. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +298 -112
  213. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm-ppc.h +333 -0
  214. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +819 -125
  215. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.h +6 -0
  216. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +708 -431
  217. data/ext/sources/ggml/src/ggml-cpu/ops.h +5 -4
  218. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +671 -31
  219. data/ext/sources/ggml/src/ggml-cpu/repack.h +14 -0
  220. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +41 -43
  221. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +3 -2
  222. data/ext/sources/ggml/src/ggml-cpu/unary-ops.cpp +151 -0
  223. data/ext/sources/ggml/src/ggml-cpu/unary-ops.h +7 -0
  224. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +124 -1
  225. data/ext/sources/ggml/src/ggml-cpu/vec.h +261 -146
  226. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +72 -1
  227. data/ext/sources/ggml/src/ggml-cuda/argmax.cu +2 -2
  228. data/ext/sources/ggml/src/ggml-cuda/argsort.cu +123 -6
  229. data/ext/sources/ggml/src/ggml-cuda/argsort.cuh +16 -0
  230. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +1 -1
  231. data/ext/sources/ggml/src/ggml-cuda/common.cuh +353 -80
  232. data/ext/sources/ggml/src/ggml-cuda/convert.cuh +10 -0
  233. data/ext/sources/ggml/src/ggml-cuda/cpy-utils.cuh +1 -1
  234. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +339 -246
  235. data/ext/sources/ggml/src/ggml-cuda/cpy.cuh +1 -5
  236. data/ext/sources/ggml/src/ggml-cuda/cumsum.cu +307 -0
  237. data/ext/sources/ggml/src/ggml-cuda/cumsum.cuh +5 -0
  238. data/ext/sources/ggml/src/ggml-cuda/diag.cu +77 -0
  239. data/ext/sources/ggml/src/ggml-cuda/diag.cuh +5 -0
  240. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +31 -21
  241. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +663 -596
  242. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +35 -741
  243. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +1241 -0
  244. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +30 -37
  245. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +14 -13
  246. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +48 -0
  247. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +83 -37
  248. data/ext/sources/ggml/src/ggml-cuda/fill.cu +37 -0
  249. data/ext/sources/ggml/src/ggml-cuda/fill.cuh +3 -0
  250. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +1155 -164
  251. data/ext/sources/ggml/src/ggml-cuda/mean.cu +5 -4
  252. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +741 -48
  253. data/ext/sources/ggml/src/ggml-cuda/mmf.cu +60 -12
  254. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +381 -42
  255. data/ext/sources/ggml/src/ggml-cuda/mmid.cu +164 -0
  256. data/ext/sources/ggml/src/ggml-cuda/mmid.cuh +5 -0
  257. data/ext/sources/ggml/src/ggml-cuda/mmq.cu +69 -176
  258. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +498 -171
  259. data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +375 -79
  260. data/ext/sources/ggml/src/ggml-cuda/mmvf.cuh +3 -2
  261. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +241 -95
  262. data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +1 -1
  263. data/ext/sources/ggml/src/ggml-cuda/pad.cu +64 -33
  264. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +151 -0
  265. data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +14 -0
  266. data/ext/sources/ggml/src/ggml-cuda/rope.cu +192 -77
  267. data/ext/sources/ggml/src/ggml-cuda/rope.cuh +2 -0
  268. data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +101 -47
  269. data/ext/sources/ggml/src/ggml-cuda/set.cu +39 -0
  270. data/ext/sources/ggml/src/ggml-cuda/set.cuh +7 -0
  271. data/ext/sources/ggml/src/ggml-cuda/softmax.cu +203 -6
  272. data/ext/sources/ggml/src/ggml-cuda/solve_tri.cu +275 -0
  273. data/ext/sources/ggml/src/ggml-cuda/solve_tri.cuh +3 -0
  274. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +14 -20
  275. data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +49 -84
  276. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
  277. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
  278. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
  279. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
  280. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
  281. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
  282. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
  283. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
  284. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
  285. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +19 -1
  286. data/ext/sources/ggml/src/ggml-cuda/top-k.cu +96 -0
  287. data/ext/sources/ggml/src/ggml-cuda/top-k.cuh +3 -0
  288. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +168 -76
  289. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +11 -4
  290. data/ext/sources/ggml/src/ggml-cuda/tri.cu +136 -0
  291. data/ext/sources/ggml/src/ggml-cuda/tri.cuh +5 -0
  292. data/ext/sources/ggml/src/ggml-cuda/unary.cu +105 -11
  293. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +36 -0
  294. data/ext/sources/ggml/src/ggml-cuda/upscale.cu +163 -7
  295. data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +4 -0
  296. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +12 -1
  297. data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +6 -0
  298. data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +80 -0
  299. data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3151 -0
  300. data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +44 -0
  301. data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +682 -0
  302. data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +360 -0
  303. data/ext/sources/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
  304. data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +566 -0
  305. data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +112 -0
  306. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +35 -0
  307. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-dma.c +63 -0
  308. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-dma.h +157 -0
  309. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +165 -0
  310. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +92 -0
  311. data/ext/sources/ggml/src/ggml-hexagon/htp/htp_iface.idl +16 -0
  312. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.c +94 -0
  313. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-inverse.c +72 -0
  314. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +49 -0
  315. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.c +1020 -0
  316. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +1353 -0
  317. data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +1001 -0
  318. data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +2503 -0
  319. data/ext/sources/ggml/src/ggml-hexagon/htp/ops-utils.h +149 -0
  320. data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +487 -0
  321. data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +168 -0
  322. data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +402 -0
  323. data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +287 -0
  324. data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.c +297 -0
  325. data/ext/sources/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
  326. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.c +454 -0
  327. data/ext/sources/ggml/src/ggml-hexagon/htp-utils.h +221 -0
  328. data/ext/sources/ggml/src/ggml-hexagon/op-desc.h +153 -0
  329. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +8 -13
  330. data/ext/sources/ggml/src/ggml-impl.h +67 -6
  331. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +2 -2
  332. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +29 -20
  333. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +652 -285
  334. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +103 -56
  335. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +496 -118
  336. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +231 -9
  337. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +1227 -224
  338. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +12 -0
  339. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +14 -8
  340. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +1972 -704
  341. data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +3 -1
  342. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +11 -0
  343. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +1430 -120
  344. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +63 -0
  345. data/ext/sources/ggml/src/ggml-opencl/kernels/expm1.cl +82 -0
  346. data/ext/sources/ggml/src/ggml-opencl/kernels/fill.cl +17 -0
  347. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +4 -3
  348. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
  349. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
  350. data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +36 -12
  351. data/ext/sources/ggml/src/ggml-opencl/kernels/mean.cl +39 -0
  352. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
  353. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +24 -10
  354. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +24 -10
  355. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
  356. data/ext/sources/ggml/src/ggml-opencl/kernels/pad.cl +29 -20
  357. data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +25 -10
  358. data/ext/sources/ggml/src/ggml-opencl/kernels/rope.cl +50 -24
  359. data/ext/sources/ggml/src/ggml-opencl/kernels/set_rows.cl +35 -16
  360. data/ext/sources/ggml/src/ggml-opencl/kernels/softplus.cl +88 -0
  361. data/ext/sources/ggml/src/ggml-opencl/kernels/sqr.cl +53 -0
  362. data/ext/sources/ggml/src/ggml-opencl/kernels/sqrt.cl +53 -0
  363. data/ext/sources/ggml/src/ggml-opencl/kernels/ssm_conv.cl +77 -0
  364. data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +13 -0
  365. data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +438 -156
  366. data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +48 -3
  367. data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +77 -0
  368. data/ext/sources/ggml/src/ggml-sycl/add-id.hpp +8 -0
  369. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +6 -0
  370. data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +0 -9
  371. data/ext/sources/ggml/src/ggml-sycl/binbcast.hpp +0 -6
  372. data/ext/sources/ggml/src/ggml-sycl/common.hpp +117 -15
  373. data/ext/sources/ggml/src/ggml-sycl/concat.cpp +55 -44
  374. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +34 -0
  375. data/ext/sources/ggml/src/ggml-sycl/count-equal.cpp +79 -0
  376. data/ext/sources/ggml/src/ggml-sycl/count-equal.hpp +9 -0
  377. data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +0 -3
  378. data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +18 -0
  379. data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +76 -3
  380. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +333 -300
  381. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +10 -2
  382. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +335 -110
  383. data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +22 -0
  384. data/ext/sources/ggml/src/ggml-sycl/norm.cpp +156 -0
  385. data/ext/sources/ggml/src/ggml-sycl/norm.hpp +2 -0
  386. data/ext/sources/ggml/src/ggml-sycl/pad.cpp +97 -0
  387. data/ext/sources/ggml/src/ggml-sycl/pad.hpp +24 -0
  388. data/ext/sources/ggml/src/ggml-sycl/pad_reflect_1d.cpp +100 -0
  389. data/ext/sources/ggml/src/ggml-sycl/pad_reflect_1d.hpp +10 -0
  390. data/ext/sources/ggml/src/ggml-sycl/presets.hpp +2 -0
  391. data/ext/sources/ggml/src/ggml-sycl/repeat_back.cpp +76 -0
  392. data/ext/sources/ggml/src/ggml-sycl/repeat_back.hpp +8 -0
  393. data/ext/sources/ggml/src/ggml-sycl/roll.cpp +122 -0
  394. data/ext/sources/ggml/src/ggml-sycl/roll.hpp +20 -0
  395. data/ext/sources/ggml/src/ggml-sycl/rope.cpp +30 -17
  396. data/ext/sources/ggml/src/ggml-sycl/set.cpp +73 -0
  397. data/ext/sources/ggml/src/ggml-sycl/set.hpp +5 -0
  398. data/ext/sources/ggml/src/ggml-sycl/softmax.cpp +327 -162
  399. data/ext/sources/ggml/src/ggml-sycl/softmax.hpp +4 -0
  400. data/ext/sources/ggml/src/ggml-sycl/ssm_conv.cpp +127 -0
  401. data/ext/sources/ggml/src/ggml-sycl/ssm_conv.hpp +5 -0
  402. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +58 -0
  403. data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +38 -18
  404. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +5013 -2859
  405. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/abs.comp +21 -0
  406. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +2 -2
  407. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +2 -2
  408. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add1.comp +28 -0
  409. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +1 -1
  410. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/arange.comp +20 -0
  411. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +2 -2
  412. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +33 -26
  413. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort_large.comp +114 -0
  414. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ceil.comp +22 -0
  415. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +2 -2
  416. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +2 -2
  417. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +2 -2
  418. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +1 -1
  419. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +47 -49
  420. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +1 -1
  421. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +2 -2
  422. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +3 -3
  423. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +4 -4
  424. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_transpose.comp +67 -0
  425. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +2 -2
  426. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +2 -2
  427. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/count_experts.comp +51 -0
  428. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum.comp +83 -0
  429. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass1.comp +60 -0
  430. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cumsum_multipass2.comp +66 -0
  431. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +1 -1
  432. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs.comp → dequant_funcs.glsl} +9 -21
  433. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs_cm2.comp → dequant_funcs_cm2.glsl} +18 -4
  434. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_head.comp → dequant_head.glsl} +1 -1
  435. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
  436. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +1 -1
  437. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +1 -1
  438. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +1 -1
  439. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +1 -1
  440. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +1 -1
  441. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +1 -1
  442. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +1 -1
  443. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +1 -1
  444. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +3 -3
  445. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +3 -3
  446. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
  447. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +1 -1
  448. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +1 -1
  449. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +3 -3
  450. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +1 -1
  451. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +1 -1
  452. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +3 -3
  453. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
  454. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +1 -1
  455. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +29 -0
  456. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +1 -1
  457. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +2 -2
  458. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +3 -3
  459. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/fill.comp +19 -0
  460. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +39 -17
  461. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{flash_attn_base.comp → flash_attn_base.glsl} +19 -1
  462. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +45 -7
  463. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +50 -12
  464. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +1 -1
  465. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/floor.comp +22 -0
  466. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +2 -2
  467. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +2 -2
  468. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +2 -2
  469. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +2 -2
  470. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +2 -2
  471. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +2 -2
  472. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_binary_head.comp → generic_binary_head.glsl} +17 -2
  473. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_head.comp → generic_head.glsl} +2 -0
  474. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{generic_unary_head.comp → generic_unary_head.glsl} +7 -0
  475. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +4 -4
  476. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +3 -3
  477. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{glu_head.comp → glu_head.glsl} +1 -1
  478. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +2 -2
  479. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +2 -2
  480. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +2 -2
  481. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +19 -7
  482. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +2 -3
  483. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +2 -2
  484. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +2 -2
  485. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +18 -0
  486. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +2 -2
  487. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +2 -2
  488. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{mul_mat_vec_base.comp → mul_mat_vec_base.glsl} +70 -25
  489. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +35 -0
  490. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +71 -21
  491. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +41 -25
  492. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +2 -2
  493. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +44 -26
  494. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +2 -2
  495. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +2 -2
  496. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +2 -2
  497. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +9 -7
  498. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +9 -7
  499. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +4 -6
  500. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +2 -2
  501. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +4 -6
  502. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +4 -6
  503. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +2 -2
  504. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +39 -36
  505. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +494 -0
  506. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +78 -103
  507. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +34 -23
  508. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{mul_mm_funcs.comp → mul_mm_funcs.glsl} +69 -59
  509. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +72 -0
  510. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +88 -228
  511. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +454 -0
  512. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +78 -0
  513. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +97 -13
  514. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/neg.comp +20 -0
  515. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +2 -2
  516. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +2 -2
  517. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +1 -1
  518. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +21 -6
  519. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +1 -1
  520. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +10 -10
  521. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +2 -2
  522. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +2 -2
  523. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +2 -2
  524. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +2 -2
  525. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +50 -4
  526. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +2 -2
  527. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +2 -2
  528. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +2 -2
  529. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +234 -0
  530. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +20 -0
  531. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +6 -50
  532. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +6 -33
  533. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +6 -33
  534. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +28 -0
  535. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +6 -39
  536. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/round.comp +29 -0
  537. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +2 -2
  538. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +2 -2
  539. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +2 -2
  540. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +2 -2
  541. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +2 -2
  542. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +1 -1
  543. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +2 -2
  544. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large1.comp +62 -0
  545. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large2.comp +79 -0
  546. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large3.comp +65 -0
  547. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_large_common.glsl +53 -0
  548. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp +23 -0
  549. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/solve_tri.comp +81 -0
  550. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +2 -2
  551. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +2 -2
  552. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +44 -0
  553. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_scan.comp +124 -0
  554. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/step.comp +22 -0
  555. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +2 -2
  556. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +2 -25
  557. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.glsl +25 -0
  558. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +2 -2
  559. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +2 -2
  560. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +2 -2
  561. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +1 -1
  562. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_argsort.comp +118 -0
  563. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp +213 -0
  564. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/topk_nary_search.comp +246 -0
  565. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +43 -0
  566. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/trunc.comp +22 -0
  567. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{types.comp → types.glsl} +345 -26
  568. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +90 -12
  569. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +335 -151
  570. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/xielu.comp +35 -0
  571. data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +28 -2
  572. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +169 -0
  573. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1964 -435
  574. data/ext/sources/ggml/src/ggml-webgpu/pre_wgsl.hpp +778 -0
  575. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +188 -0
  576. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +101 -0
  577. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +33 -10
  578. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +591 -0
  579. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +1 -1
  580. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +323 -0
  581. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +6 -6
  582. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +97 -0
  583. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl +247 -0
  584. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.tmpl.wgsl +302 -0
  585. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +267 -0
  586. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +83 -17
  587. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rope.tmpl.wgsl +295 -0
  588. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl +90 -0
  589. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.tmpl.wgsl +112 -0
  590. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.tmpl.wgsl +345 -0
  591. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary_op.wgsl +483 -0
  592. data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +92 -0
  593. data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +466 -0
  594. data/ext/sources/ggml/src/ggml.c +425 -33
  595. data/ext/sources/include/whisper.h +1 -0
  596. data/ext/sources/src/CMakeLists.txt +3 -1
  597. data/ext/sources/src/whisper.cpp +101 -35
  598. data/ext/sources/tests/CMakeLists.txt +2 -2
  599. data/ext/sources/tests/test-vad-full.cpp +4 -2
  600. data/ext/sources/tests/test-vad.cpp +1 -1
  601. data/extsources.rb +1 -0
  602. data/lib/whisper/model/uri.rb +17 -18
  603. data/sig/whisper.rbs +119 -2
  604. data/test/test_params.rb +16 -8
  605. data/test/test_segment.rb +0 -1
  606. data/test/test_token.rb +70 -0
  607. data/test/test_vad.rb +1 -1
  608. data/test/test_vad_context.rb +50 -0
  609. data/test/test_vad_segment.rb +19 -0
  610. data/test/test_vad_segments.rb +16 -0
  611. data/test/test_whisper.rb +7 -0
  612. data/whispercpp.gemspec +1 -1
  613. metadata +287 -34
  614. data/ext/sources/build-xcframework.sh +0 -571
  615. data/ext/sources/ggml/src/ggml-cann/Doxyfile +0 -2579
  616. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +0 -105
  617. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +0 -55
  618. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add.tmpl.wgsl +0 -44
  619. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_in_place.tmpl.wgsl +0 -41
  620. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +0 -60
  621. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul.tmpl.wgsl +0 -44
  622. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_in_place.tmpl.wgsl +0 -41
  623. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_in_place.wgsl +0 -48
  624. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_bfloat16_support.comp → feature-tests/bfloat16.comp} +0 -0
  625. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat_support.comp → feature-tests/coopmat.comp} +0 -0
  626. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat2_support.comp → feature-tests/coopmat2.comp} +0 -0
  627. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{test_integer_dot_support.comp → feature-tests/integer_dot.comp} +0 -0
  628. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{glu_main.comp → glu_main.glsl} +0 -0
  629. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{rte.comp → rte.glsl} +0 -0
  630. /data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/{utils.comp → utils.glsl} +0 -0
@@ -1,19 +1,23 @@
1
1
  #include <ruby.h>
2
2
  #include "ruby_whisper.h"
3
3
 
4
- #define N_KEY_NAMES 5
4
+ #define N_KEY_NAMES 6
5
5
 
6
+ extern ID id___method__;
7
+ extern ID id_to_enum;
6
8
  static VALUE sym_start_time;
7
9
  static VALUE sym_end_time;
8
10
  static VALUE sym_text;
9
11
  static VALUE sym_no_speech_prob;
10
12
  static VALUE sym_speaker_turn_next;
11
- static VALUE key_names;
13
+ static VALUE sym_n_tokens;
12
14
 
13
15
  extern const rb_data_type_t ruby_whisper_type;
14
16
 
15
17
  extern VALUE cSegment;
16
18
 
19
+ extern VALUE ruby_whisper_token_s_init(struct whisper_context *context, int i_segment, int index);
20
+
17
21
  static void
18
22
  rb_whisper_segment_mark(void *p)
19
23
  {
@@ -29,6 +33,9 @@ ruby_whisper_segment_memsize(const void *p)
29
33
  if (!rws) {
30
34
  return 0;
31
35
  }
36
+ if (rws->index) {
37
+ size += sizeof(rws->index);
38
+ }
32
39
  return size;
33
40
  }
34
41
 
@@ -69,7 +76,7 @@ ruby_whisper_segment_get_start_time(VALUE self)
69
76
  ruby_whisper_segment *rws;
70
77
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
71
78
  ruby_whisper *rw;
72
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
79
+ GetContext(rws->context, rw);
73
80
  const int64_t t0 = whisper_full_get_segment_t0(rw->context, rws->index);
74
81
  // able to multiply 10 without overflow because to_timestamp() in whisper.cpp does it
75
82
  return LONG2NUM(t0 * 10);
@@ -87,7 +94,7 @@ ruby_whisper_segment_get_end_time(VALUE self)
87
94
  ruby_whisper_segment *rws;
88
95
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
89
96
  ruby_whisper *rw;
90
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
97
+ GetContext(rws->context, rw);
91
98
  const int64_t t1 = whisper_full_get_segment_t1(rw->context, rws->index);
92
99
  // able to multiply 10 without overflow because to_timestamp() in whisper.cpp does it
93
100
  return LONG2NUM(t1 * 10);
@@ -105,7 +112,7 @@ ruby_whisper_segment_get_speaker_turn_next(VALUE self)
105
112
  ruby_whisper_segment *rws;
106
113
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
107
114
  ruby_whisper *rw;
108
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
115
+ GetContext(rws->context, rw);
109
116
  return whisper_full_get_segment_speaker_turn_next(rw->context, rws->index) ? Qtrue : Qfalse;
110
117
  }
111
118
 
@@ -119,7 +126,7 @@ ruby_whisper_segment_get_text(VALUE self)
119
126
  ruby_whisper_segment *rws;
120
127
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
121
128
  ruby_whisper *rw;
122
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
129
+ GetContext(rws->context, rw);
123
130
  const char * text = whisper_full_get_segment_text(rw->context, rws->index);
124
131
  return rb_str_new2(text);
125
132
  }
@@ -134,15 +141,67 @@ ruby_whisper_segment_get_no_speech_prob(VALUE self)
134
141
  ruby_whisper_segment *rws;
135
142
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
136
143
  ruby_whisper *rw;
137
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
144
+ GetContext(rws->context, rw);
138
145
  return DBL2NUM(whisper_full_get_segment_no_speech_prob(rw->context, rws->index));
139
146
  }
140
147
 
148
+ /*
149
+ * Get number of tokens in the segment
150
+ *
151
+ * call-seq:
152
+ * n_tokens -> Integer
153
+ */
154
+ static VALUE
155
+ ruby_whisper_segment_get_n_tokens(VALUE self)
156
+ {
157
+ ruby_whisper_segment *rws;
158
+ TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
159
+ ruby_whisper *rw;
160
+ GetContext(rws->context, rw);
161
+ return INT2NUM(whisper_full_n_tokens(rw->context, rws->index));
162
+ }
163
+
164
+ /*
165
+ * Yields each Whisper::Token:
166
+ *
167
+ * whisper.each_segment.first.each_token do |token|
168
+ * p token
169
+ * end
170
+ *
171
+ * Returns an Enumerator if no block is given:
172
+ *
173
+ * whisper.each_segment.first.each_token.to_a # => [#<Whisper::Token>, ...]
174
+ *
175
+ * call-seq:
176
+ * each_token {|token| ... }
177
+ * each_token -> Enumerator
178
+ */
179
+ static VALUE
180
+ ruby_whisper_segment_each_token(VALUE self)
181
+ {
182
+ if (!rb_block_given_p()) {
183
+ const VALUE method_name = rb_funcall(self, id___method__, 0);
184
+ return rb_funcall(self, id_to_enum, 1, method_name);
185
+ }
186
+
187
+ ruby_whisper_segment *rws;
188
+ TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
189
+ ruby_whisper *rw;
190
+ GetContext(rws->context, rw);
191
+
192
+ const int n_tokens = whisper_full_n_tokens(rw->context, rws->index);
193
+ for (int i = 0; i < n_tokens; ++i) {
194
+ rb_yield(ruby_whisper_token_s_init(rw->context, rws->index, i));
195
+ }
196
+
197
+ return self;
198
+ }
199
+
141
200
  /*
142
201
  * call-seq:
143
202
  * deconstruct_keys(keys) -> hash
144
203
  *
145
- * Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next
204
+ * Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next, :n_tokens
146
205
  *
147
206
  * whisper.each_segment do |segment|
148
207
  * segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
@@ -156,12 +215,19 @@ ruby_whisper_segment_deconstruct_keys(VALUE self, VALUE keys)
156
215
  ruby_whisper_segment *rws;
157
216
  TypedData_Get_Struct(self, ruby_whisper_segment, &ruby_whisper_segment_type, rws);
158
217
  ruby_whisper *rw;
159
- TypedData_Get_Struct(rws->context, ruby_whisper, &ruby_whisper_type, rw);
218
+ GetContext(rws->context, rw);
160
219
 
161
220
  VALUE hash = rb_hash_new();
162
221
  long n_keys;
163
222
  if (NIL_P(keys)) {
164
- keys = key_names;
223
+ keys = rb_ary_new3(
224
+ N_KEY_NAMES,
225
+ sym_start_time,
226
+ sym_end_time,
227
+ sym_text,
228
+ sym_no_speech_prob,
229
+ sym_speaker_turn_next
230
+ );
165
231
  n_keys = N_KEY_NAMES;
166
232
  } else {
167
233
  n_keys = RARRAY_LEN(keys);
@@ -186,13 +252,16 @@ ruby_whisper_segment_deconstruct_keys(VALUE self, VALUE keys)
186
252
  if (key == sym_speaker_turn_next) {
187
253
  rb_hash_aset(hash, key, ruby_whisper_segment_get_speaker_turn_next(self));
188
254
  }
255
+ if (key == sym_n_tokens) {
256
+ rb_hash_aset(hash, key, ruby_whisper_segment_get_n_tokens(self));
257
+ }
189
258
  }
190
259
 
191
260
  return hash;
192
261
  }
193
262
 
194
263
  void
195
- init_ruby_whisper_segment(VALUE *mWhisper, VALUE *cContext)
264
+ init_ruby_whisper_segment(VALUE *mWhisper)
196
265
  {
197
266
  cSegment = rb_define_class_under(*mWhisper, "Segment", rb_cObject);
198
267
 
@@ -201,14 +270,7 @@ init_ruby_whisper_segment(VALUE *mWhisper, VALUE *cContext)
201
270
  sym_text = ID2SYM(rb_intern("text"));
202
271
  sym_no_speech_prob = ID2SYM(rb_intern("no_speech_prob"));
203
272
  sym_speaker_turn_next = ID2SYM(rb_intern("speaker_turn_next"));
204
- key_names = rb_ary_new3(
205
- N_KEY_NAMES,
206
- sym_start_time,
207
- sym_end_time,
208
- sym_text,
209
- sym_no_speech_prob,
210
- sym_speaker_turn_next
211
- );
273
+ sym_n_tokens = ID2SYM(rb_intern("n_tokens"));
212
274
 
213
275
  rb_define_alloc_func(cSegment, ruby_whisper_segment_allocate);
214
276
  rb_define_method(cSegment, "start_time", ruby_whisper_segment_get_start_time, 0);
@@ -216,5 +278,8 @@ init_ruby_whisper_segment(VALUE *mWhisper, VALUE *cContext)
216
278
  rb_define_method(cSegment, "speaker_turn_next?", ruby_whisper_segment_get_speaker_turn_next, 0);
217
279
  rb_define_method(cSegment, "text", ruby_whisper_segment_get_text, 0);
218
280
  rb_define_method(cSegment, "no_speech_prob", ruby_whisper_segment_get_no_speech_prob, 0);
281
+ rb_define_method(cSegment, "n_tokens", ruby_whisper_segment_get_n_tokens, 0);
282
+ rb_define_method(cSegment, "each_token", ruby_whisper_segment_each_token, 0);
219
283
  rb_define_method(cSegment, "deconstruct_keys", ruby_whisper_segment_deconstruct_keys, 1);
220
284
  }
285
+ #undef N_KEY_NAMES
@@ -0,0 +1,351 @@
1
+ #include <ruby.h>
2
+ #include "ruby_whisper.h"
3
+
4
+ #define N_KEY_NAMES 11
5
+
6
+ extern VALUE cToken;
7
+ extern const rb_data_type_t ruby_whisper_type;
8
+
9
+ static VALUE sym_id;
10
+ static VALUE sym_tid;
11
+ static VALUE sym_probability;
12
+ static VALUE sym_log_probability;
13
+ static VALUE sym_pt;
14
+ static VALUE sym_ptsum;
15
+ static VALUE sym_t_dtw;
16
+ static VALUE sym_voice_length;
17
+ static VALUE sym_start_time;
18
+ static VALUE sym_end_time;
19
+ static VALUE sym_text;
20
+
21
+ static size_t
22
+ ruby_whisper_token_memsize(const void *p)
23
+ {
24
+ const ruby_whisper_token *rwt = (const ruby_whisper_token *)p;
25
+ if (!rwt) {
26
+ return 0;
27
+ }
28
+ return sizeof(rwt);
29
+ }
30
+
31
+ static const rb_data_type_t ruby_whisper_token_type = {
32
+ "ruby_whisper_token",
33
+ {0, RUBY_DEFAULT_FREE, ruby_whisper_token_memsize,},
34
+ 0, 0,
35
+ 0
36
+ };
37
+
38
+ static VALUE
39
+ ruby_whisper_token_allocate(VALUE klass)
40
+ {
41
+ ruby_whisper_token *rwt;
42
+ VALUE token = TypedData_Make_Struct(klass, ruby_whisper_token, &ruby_whisper_token_type, rwt);
43
+ rwt->token_data = NULL;
44
+ rwt->text = NULL;
45
+ return token;
46
+ }
47
+
48
+ VALUE
49
+ ruby_whisper_token_s_init(struct whisper_context *context, int i_segment, int i_token)
50
+ {
51
+ whisper_token_data token_data = whisper_full_get_token_data(context, i_segment, i_token);
52
+ const VALUE token = ruby_whisper_token_allocate(cToken);
53
+ ruby_whisper_token *rwt;
54
+ TypedData_Get_Struct(token, ruby_whisper_token, &ruby_whisper_token_type, rwt);
55
+ rwt->token_data = &token_data;
56
+ rwt->text = whisper_full_get_token_text(context, i_segment, i_token);
57
+ return token;
58
+ }
59
+
60
+ /*
61
+ * Token ID.
62
+ *
63
+ * call-seq:
64
+ * id -> Integer
65
+ */
66
+ static VALUE
67
+ ruby_whisper_token_get_id(VALUE self)
68
+ {
69
+ ruby_whisper_token *rwt;
70
+ GetToken(self, rwt);
71
+ return INT2NUM(rwt->token_data->id);
72
+ }
73
+
74
+ /*
75
+ * Forced timestamp token ID.
76
+ *
77
+ * call-seq:
78
+ * tid -> Integer
79
+ */
80
+ static VALUE
81
+ ruby_whisper_token_get_tid(VALUE self)
82
+ {
83
+ ruby_whisper_token *rwt;
84
+ GetToken(self, rwt);
85
+ return INT2NUM(rwt->token_data->tid);
86
+ }
87
+
88
+ /*
89
+ * Probability of the token.
90
+ *
91
+ * call-seq:
92
+ * probability -> Float
93
+ */
94
+ static VALUE
95
+ ruby_whisper_token_get_p(VALUE self)
96
+ {
97
+ ruby_whisper_token *rwt;
98
+ GetToken(self, rwt);
99
+ return DBL2NUM(rwt->token_data->p);
100
+ }
101
+
102
+ /*
103
+ * Log probability of the token.
104
+ *
105
+ * call-seq:
106
+ * log_probability -> Float
107
+ */
108
+ static VALUE
109
+ ruby_whisper_token_get_plog(VALUE self)
110
+ {
111
+ ruby_whisper_token *rwt;
112
+ GetToken(self, rwt);
113
+ return DBL2NUM(rwt->token_data->plog);
114
+ }
115
+
116
+ /*
117
+ * Probability of the timestamp token.
118
+ *
119
+ * call-seq:
120
+ * pt -> Float
121
+ */
122
+ static VALUE
123
+ ruby_whisper_token_get_pt(VALUE self)
124
+ {
125
+ ruby_whisper_token *rwt;
126
+ GetToken(self, rwt);
127
+ return DBL2NUM(rwt->token_data->pt);
128
+ }
129
+
130
+ /*
131
+ * Sum of probability of all timestamp tokens.
132
+ *
133
+ * call-seq:
134
+ * ptsum -> Float
135
+ */
136
+ static VALUE
137
+ ruby_whisper_token_get_ptsum(VALUE self)
138
+ {
139
+ ruby_whisper_token *rwt;
140
+ GetToken(self, rwt);
141
+ return DBL2NUM(rwt->token_data->ptsum);
142
+ }
143
+
144
+ /*
145
+ * [EXPERIMENTAL] Token-level timestamps with DTW
146
+ *
147
+ * Do not use if you haven't computed token-level timestamps with dtw.
148
+ * Roughly corresponds to the moment in audio in which the token was output.
149
+ *
150
+ * call-seq:
151
+ * t_dtw -> Integer
152
+ */
153
+ static VALUE
154
+ ruby_whisper_token_get_t_dtw(VALUE self)
155
+ {
156
+ ruby_whisper_token *rwt;
157
+ GetToken(self, rwt);
158
+ return LONG2NUM(rwt->token_data->t_dtw);
159
+ }
160
+
161
+ /*
162
+ * Voice length of the token.
163
+ *
164
+ * call-seq:
165
+ * voice_length -> Float
166
+ */
167
+ static VALUE
168
+ ruby_whisper_token_get_vlen(VALUE self)
169
+ {
170
+ ruby_whisper_token *rwt;
171
+ GetToken(self, rwt);
172
+ return DBL2NUM(rwt->token_data->vlen);
173
+ }
174
+
175
+ /*
176
+ * Get the token text of the token.
177
+ *
178
+ * call-seq:
179
+ * text -> String
180
+ */
181
+ static VALUE
182
+ ruby_whisper_token_get_text(VALUE self)
183
+ {
184
+ ruby_whisper_token *rwt;
185
+ GetToken(self, rwt);
186
+ return rb_str_new2(rwt->text);
187
+ }
188
+
189
+
190
+ /*
191
+ * Start time of the token.
192
+ *
193
+ * Token-level timestamp data.
194
+ * Do not use if you haven't computed token-level timestamps.
195
+ *
196
+ * call-seq:
197
+ * start_time -> Integer
198
+ */
199
+ static VALUE
200
+ ruby_whisper_token_get_start_time(VALUE self)
201
+ {
202
+ ruby_whisper_token *rwt;
203
+ GetToken(self, rwt);
204
+ return LONG2NUM(rwt->token_data->t0 * 10);
205
+ }
206
+
207
+ /*
208
+ * End time of the token.
209
+ *
210
+ * Token-level timestamp data.
211
+ * Do not use if you haven't computed token-level timestamps.
212
+ *
213
+ * call-seq:
214
+ * end_time -> Integer
215
+ */
216
+ static VALUE
217
+ ruby_whisper_token_get_end_time(VALUE self)
218
+ {
219
+ ruby_whisper_token *rwt;
220
+ GetToken(self, rwt);
221
+ return LONG2NUM(rwt->token_data->t1 * 10);
222
+ }
223
+
224
+ /*
225
+ * call-seq:
226
+ * deconstruct_keys(keys) -> hash
227
+ *
228
+ * Possible keys: :id, :tid, :probability, :log_probability, :pt, :ptsum,
229
+ * :t_dtw, :voice_length, :start_time, :end_time, :text
230
+ * segment.each_token do |token|
231
+ * token => {text:, probability:}
232
+ puts "#{text} (#{probability})"
233
+ * end
234
+ */
235
+ static VALUE ruby_whisper_token_deconstruct_keys(VALUE self, VALUE keys)
236
+ {
237
+ ruby_whisper_token *rwt;
238
+ GetToken(self, rwt);
239
+ VALUE hash = rb_hash_new();
240
+ long n_keys = 0;
241
+
242
+ if (NIL_P(keys)) {
243
+ keys = rb_ary_new3(
244
+ N_KEY_NAMES,
245
+ sym_id,
246
+ sym_tid,
247
+ sym_probability,
248
+ sym_log_probability,
249
+ sym_pt,
250
+ sym_ptsum,
251
+ sym_t_dtw,
252
+ sym_voice_length,
253
+ sym_start_time,
254
+ sym_end_time,
255
+ sym_text
256
+ );
257
+ n_keys = N_KEY_NAMES;
258
+ } else {
259
+ n_keys = RARRAY_LEN(keys);
260
+ if (n_keys > N_KEY_NAMES) {
261
+ return hash;
262
+ }
263
+ }
264
+
265
+ for (int i = 0; i < n_keys; i++) {
266
+ VALUE key = rb_ary_entry(keys, i);
267
+ if (key == sym_start_time) {
268
+ rb_hash_aset(hash, key, ruby_whisper_token_get_start_time(self));
269
+ continue;
270
+ }
271
+ if (key == sym_end_time) {
272
+ rb_hash_aset(hash, key, ruby_whisper_token_get_end_time(self));
273
+ continue;
274
+ }
275
+ if (key == sym_text) {
276
+ rb_hash_aset(hash, key, ruby_whisper_token_get_text(self));
277
+ continue;
278
+ }
279
+ if (key == sym_probability) {
280
+ rb_hash_aset(hash, key, ruby_whisper_token_get_p(self));
281
+ continue;
282
+ }
283
+ if (key == sym_id) {
284
+ rb_hash_aset(hash, key, ruby_whisper_token_get_id(self));
285
+ continue;
286
+ }
287
+ if (key == sym_tid) {
288
+ rb_hash_aset(hash, key, ruby_whisper_token_get_tid(self));
289
+ continue;
290
+ }
291
+ if (key == sym_log_probability) {
292
+ rb_hash_aset(hash, key, ruby_whisper_token_get_plog(self));
293
+ continue;
294
+ }
295
+ if (key == sym_pt) {
296
+ rb_hash_aset(hash, key, ruby_whisper_token_get_pt(self));
297
+ continue;
298
+ }
299
+ if (key == sym_ptsum) {
300
+ rb_hash_aset(hash, key, ruby_whisper_token_get_ptsum(self));
301
+ continue;
302
+ }
303
+ if (key == sym_t_dtw) {
304
+ rb_hash_aset(hash, key, ruby_whisper_token_get_t_dtw(self));
305
+ continue;
306
+ }
307
+ if (key == sym_voice_length) {
308
+ rb_hash_aset(hash, key, ruby_whisper_token_get_vlen(self));
309
+ continue;
310
+ }
311
+ }
312
+
313
+ return hash;
314
+ }
315
+
316
+
317
+ void
318
+ init_ruby_whisper_token(VALUE *mWhisper)
319
+ {
320
+ cToken = rb_define_class_under(*mWhisper, "Token", rb_cObject);
321
+
322
+ rb_define_alloc_func(cToken, ruby_whisper_token_allocate);
323
+
324
+ sym_id = ID2SYM(rb_intern("id"));
325
+ sym_tid = ID2SYM(rb_intern("tid"));
326
+ sym_probability = ID2SYM(rb_intern("probability"));
327
+ sym_log_probability = ID2SYM(rb_intern("log_probability"));
328
+ sym_pt = ID2SYM(rb_intern("pt"));
329
+ sym_ptsum = ID2SYM(rb_intern("ptsum"));
330
+ sym_t_dtw = ID2SYM(rb_intern("t_dtw"));
331
+ sym_voice_length = ID2SYM(rb_intern("voice_length"));
332
+ sym_start_time = ID2SYM(rb_intern("start_time"));
333
+ sym_end_time = ID2SYM(rb_intern("end_time"));
334
+ sym_text = ID2SYM(rb_intern("text"));
335
+
336
+ rb_define_method(cToken, "id", ruby_whisper_token_get_id, 0);
337
+ rb_define_method(cToken, "tid", ruby_whisper_token_get_tid, 0);
338
+ rb_define_method(cToken, "probability", ruby_whisper_token_get_p, 0);
339
+ rb_define_method(cToken, "log_probability", ruby_whisper_token_get_plog, 0);
340
+ rb_define_method(cToken, "pt", ruby_whisper_token_get_pt, 0);
341
+ rb_define_method(cToken, "ptsum", ruby_whisper_token_get_ptsum, 0);
342
+ rb_define_method(cToken, "t_dtw", ruby_whisper_token_get_t_dtw, 0);
343
+ rb_define_method(cToken, "voice_length", ruby_whisper_token_get_vlen, 0);
344
+ rb_define_method(cToken, "start_time", ruby_whisper_token_get_start_time, 0);
345
+ rb_define_method(cToken, "end_time", ruby_whisper_token_get_end_time, 0);
346
+ rb_define_method(cToken, "text", ruby_whisper_token_get_text, 0);
347
+
348
+ rb_define_method(cToken, "deconstruct_keys", ruby_whisper_token_deconstruct_keys, 1);
349
+ }
350
+
351
+ #undef N_KEY_NAMES
@@ -43,7 +43,7 @@ ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self) {
43
43
 
44
44
  int n_processors = opts[0] == Qundef ? 1 : NUM2INT(opts[0]);
45
45
 
46
- TypedData_Get_Struct(self, ruby_whisper, &ruby_whisper_type, rw);
46
+ GetContext(self, rw);
47
47
  TypedData_Get_Struct(params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
48
48
 
49
49
  if (!rb_respond_to(wave_file_path, id_to_s)) {
@@ -0,0 +1,75 @@
1
+ #include <ruby.h>
2
+ #include "ruby_whisper.h"
3
+
4
+ extern ID id_to_s;
5
+
6
+ extern VALUE cVADContext;
7
+
8
+ extern VALUE ruby_whisper_vad_detect(VALUE self, VALUE file_path, VALUE params);
9
+ extern VALUE ruby_whisper_normalize_model_path(VALUE model_path);
10
+
11
+ static size_t
12
+ ruby_whisper_vad_context_memsize(const void *p)
13
+ {
14
+ const ruby_whisper_vad_context *rwvc = p;
15
+ size_t size = sizeof(rwvc);
16
+ if (!rwvc) {
17
+ return 0;
18
+ }
19
+ if (rwvc->context) {
20
+ size += sizeof(rwvc->context);
21
+ }
22
+ return size;
23
+ }
24
+
25
+ static void
26
+ ruby_whisper_vad_context_free(void *p)
27
+ {
28
+ ruby_whisper_vad_context *rwvc = (ruby_whisper_vad_context *)p;
29
+ if (rwvc->context) {
30
+ whisper_vad_free(rwvc->context);
31
+ rwvc->context = NULL;
32
+ }
33
+ xfree(rwvc);
34
+ }
35
+
36
+ const rb_data_type_t ruby_whisper_vad_context_type = {
37
+ "ruby_whisper_vad_context",
38
+ {0, ruby_whisper_vad_context_free, ruby_whisper_vad_context_memsize,},
39
+ 0, 0,
40
+ 0
41
+ };
42
+
43
+ static VALUE
44
+ ruby_whisper_vad_context_s_allocate(VALUE klass)
45
+ {
46
+ ruby_whisper_vad_context *rwvc;
47
+ VALUE obj = TypedData_Make_Struct(klass, ruby_whisper_vad_context, &ruby_whisper_vad_context_type, rwvc);
48
+ rwvc->context = NULL;
49
+ return obj;
50
+ }
51
+
52
+ static VALUE
53
+ ruby_whisper_vad_context_initialize(VALUE self, VALUE model_path)
54
+ {
55
+ ruby_whisper_vad_context *rwvc;
56
+ struct whisper_vad_context *context;
57
+
58
+ model_path = ruby_whisper_normalize_model_path(model_path);
59
+ context = whisper_vad_init_from_file_with_params(StringValueCStr(model_path), whisper_vad_default_context_params());
60
+ if (context == NULL) {
61
+ rb_raise(rb_eRuntimeError, "Failed to initialize whisper VAD context");
62
+ }
63
+ TypedData_Get_Struct(self, ruby_whisper_vad_context, &ruby_whisper_vad_context_type, rwvc);
64
+ rwvc->context = context;
65
+
66
+ return Qnil;
67
+ }
68
+
69
+ void init_ruby_whisper_vad_context(VALUE *mVAD)
70
+ {
71
+ cVADContext = rb_define_class_under(*mVAD, "Context", rb_cObject);
72
+ rb_define_alloc_func(cVADContext, ruby_whisper_vad_context_s_allocate);
73
+ rb_define_method(cVADContext, "initialize", ruby_whisper_vad_context_initialize, 1);
74
+ rb_define_method(cVADContext, "detect", ruby_whisper_vad_detect, 2);
75
+ }
@@ -0,0 +1,50 @@
1
+ #include <ruby.h>
2
+ #include "ruby_whisper.h"
3
+ #include "common-whisper.h"
4
+ #include <string>
5
+ #include <vector>
6
+
7
+ #ifdef __cplusplus
8
+ extern "C" {
9
+ #endif
10
+
11
+ extern VALUE cVADSegments;
12
+
13
+ extern const rb_data_type_t ruby_whisper_vad_context_type;
14
+ extern const rb_data_type_t ruby_whisper_vad_params_type;
15
+ extern const rb_data_type_t ruby_whisper_vad_segments_type;
16
+
17
+ extern VALUE ruby_whisper_vad_segments_s_init(struct whisper_vad_segments *segments);
18
+
19
+ VALUE
20
+ ruby_whisper_vad_detect(VALUE self, VALUE file_path, VALUE params) {
21
+ ruby_whisper_vad_context *rwvc;
22
+ ruby_whisper_vad_params *rwvp;
23
+ std::string cpp_file_path;
24
+ std::vector<float> pcmf32;
25
+ std::vector<std::vector<float>> pcmf32s;
26
+ whisper_vad_segments *segments;
27
+
28
+ TypedData_Get_Struct(self, ruby_whisper_vad_context, &ruby_whisper_vad_context_type, rwvc);
29
+ if (rwvc->context == NULL) {
30
+ rb_raise(rb_eRuntimeError, "Doesn't have referenxe to context internally");
31
+ }
32
+ TypedData_Get_Struct(params, ruby_whisper_vad_params, &ruby_whisper_vad_params_type, rwvp);
33
+
34
+ cpp_file_path = StringValueCStr(file_path);
35
+
36
+ if (!read_audio_data(cpp_file_path, pcmf32, pcmf32s, false)) {
37
+ rb_raise(rb_eRuntimeError, "Failed to open '%s' as WAV file\n", cpp_file_path.c_str());
38
+ }
39
+
40
+ segments = whisper_vad_segments_from_samples(rwvc->context, rwvp->params, pcmf32.data(), pcmf32.size());
41
+ if (segments == nullptr) {
42
+ rb_raise(rb_eRuntimeError, "Failed to process audio\n");
43
+ }
44
+
45
+ return ruby_whisper_vad_segments_s_init(segments);
46
+ }
47
+
48
+ #ifdef __cplusplus
49
+ }
50
+ #endif