whispercpp 1.3.2 → 1.3.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (664) hide show
  1. checksums.yaml +4 -4
  2. data/.gitignore +6 -3
  3. data/README.md +71 -14
  4. data/Rakefile +20 -7
  5. data/ext/.gitignore +4 -6
  6. data/ext/dependencies.rb +36 -24
  7. data/ext/extconf.rb +1 -1
  8. data/ext/options.rb +48 -184
  9. data/ext/ruby_whisper.c +18 -0
  10. data/ext/ruby_whisper_context.c +43 -12
  11. data/ext/ruby_whisper_model.c +1 -1
  12. data/ext/ruby_whisper_params.c +59 -27
  13. data/ext/ruby_whisper_segment.c +81 -4
  14. data/ext/ruby_whisper_transcribe.cpp +13 -7
  15. data/ext/ruby_whisper_vad_params.c +1 -1
  16. data/ext/sources/CMakeLists.txt +5 -1
  17. data/ext/sources/bindings/javascript/package.json +1 -1
  18. data/ext/sources/build-xcframework.sh +24 -0
  19. data/ext/sources/examples/CMakeLists.txt +1 -0
  20. data/ext/sources/examples/addon.node/__test__/whisper.spec.js +120 -24
  21. data/ext/sources/examples/addon.node/addon.cpp +154 -35
  22. data/ext/sources/examples/addon.node/index.js +10 -5
  23. data/ext/sources/examples/addon.node/vad-example.js +132 -0
  24. data/ext/sources/examples/bench/bench.cpp +29 -18
  25. data/ext/sources/examples/bench.wasm/index-tmpl.html +10 -9
  26. data/ext/sources/examples/cli/cli.cpp +7 -4
  27. data/ext/sources/examples/command/command.cpp +58 -32
  28. data/ext/sources/examples/command.wasm/index-tmpl.html +5 -4
  29. data/ext/sources/examples/common-ggml.cpp +2 -0
  30. data/ext/sources/examples/common-whisper.cpp +14 -7
  31. data/ext/sources/examples/lsp/lsp.cpp +21 -17
  32. data/ext/sources/examples/quantize/quantize.cpp +3 -0
  33. data/ext/sources/examples/server/CMakeLists.txt +3 -0
  34. data/ext/sources/examples/server/server.cpp +193 -35
  35. data/ext/sources/examples/server.py +6 -1
  36. data/ext/sources/examples/stream/stream.cpp +10 -2
  37. data/ext/sources/examples/stream.wasm/emscripten.cpp +6 -6
  38. data/ext/sources/examples/stream.wasm/index-tmpl.html +82 -5
  39. data/ext/sources/examples/talk-llama/CMakeLists.txt +3 -0
  40. data/ext/sources/examples/talk-llama/llama-adapter.cpp +101 -4
  41. data/ext/sources/examples/talk-llama/llama-adapter.h +6 -0
  42. data/ext/sources/examples/talk-llama/llama-arch.cpp +756 -15
  43. data/ext/sources/examples/talk-llama/llama-arch.h +85 -1
  44. data/ext/sources/examples/talk-llama/llama-batch.cpp +773 -272
  45. data/ext/sources/examples/talk-llama/llama-batch.h +126 -55
  46. data/ext/sources/examples/talk-llama/llama-chat.cpp +150 -13
  47. data/ext/sources/examples/talk-llama/llama-chat.h +8 -0
  48. data/ext/sources/examples/talk-llama/llama-context.cpp +814 -542
  49. data/ext/sources/examples/talk-llama/llama-context.h +68 -32
  50. data/ext/sources/examples/talk-llama/llama-cparams.cpp +1 -1
  51. data/ext/sources/examples/talk-llama/llama-cparams.h +4 -4
  52. data/ext/sources/examples/talk-llama/llama-graph.cpp +787 -440
  53. data/ext/sources/examples/talk-llama/llama-graph.h +333 -153
  54. data/ext/sources/examples/talk-llama/llama-hparams.cpp +128 -6
  55. data/ext/sources/examples/talk-llama/llama-hparams.h +80 -17
  56. data/ext/sources/examples/talk-llama/llama-impl.h +2 -0
  57. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +326 -0
  58. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.h +137 -0
  59. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +1248 -1967
  60. data/ext/sources/examples/talk-llama/llama-kv-cache.h +218 -345
  61. data/ext/sources/examples/talk-llama/llama-kv-cells.h +164 -52
  62. data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +266 -0
  63. data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +139 -0
  64. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +1154 -0
  65. data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +182 -0
  66. data/ext/sources/examples/talk-llama/llama-memory.cpp +58 -0
  67. data/ext/sources/examples/talk-llama/llama-memory.h +94 -4
  68. data/ext/sources/examples/talk-llama/llama-mmap.cpp +1 -1
  69. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +44 -17
  70. data/ext/sources/examples/talk-llama/llama-model-loader.h +3 -2
  71. data/ext/sources/examples/talk-llama/llama-model-saver.cpp +1 -0
  72. data/ext/sources/examples/talk-llama/llama-model.cpp +11377 -5248
  73. data/ext/sources/examples/talk-llama/llama-model.h +87 -9
  74. data/ext/sources/examples/talk-llama/llama-quant.cpp +137 -16
  75. data/ext/sources/examples/talk-llama/llama-sampling.cpp +226 -126
  76. data/ext/sources/examples/talk-llama/llama-vocab.cpp +502 -38
  77. data/ext/sources/examples/talk-llama/llama-vocab.h +46 -0
  78. data/ext/sources/examples/talk-llama/llama.cpp +76 -17
  79. data/ext/sources/examples/talk-llama/llama.h +176 -151
  80. data/ext/sources/examples/talk-llama/talk-llama.cpp +11 -6
  81. data/ext/sources/examples/talk-llama/unicode.cpp +212 -0
  82. data/ext/sources/examples/talk-llama/unicode.h +45 -0
  83. data/ext/sources/examples/vad-speech-segments/speech.cpp +6 -0
  84. data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +6 -2
  85. data/ext/sources/examples/whisper.wasm/index-tmpl.html +17 -16
  86. data/ext/sources/ggml/CMakeLists.txt +106 -33
  87. data/ext/sources/ggml/cmake/common.cmake +24 -0
  88. data/ext/sources/ggml/cmake/ggml-config.cmake.in +132 -93
  89. data/ext/sources/ggml/include/ggml-backend.h +18 -2
  90. data/ext/sources/ggml/include/ggml-cpu.h +2 -0
  91. data/ext/sources/ggml/include/ggml-metal.h +1 -6
  92. data/ext/sources/ggml/include/ggml-opt.h +25 -6
  93. data/ext/sources/ggml/include/ggml-webgpu.h +19 -0
  94. data/ext/sources/ggml/include/ggml-zdnn.h +17 -0
  95. data/ext/sources/ggml/include/ggml.h +365 -21
  96. data/ext/sources/ggml/src/CMakeLists.txt +98 -25
  97. data/ext/sources/ggml/src/ggml-alloc.c +265 -141
  98. data/ext/sources/ggml/src/ggml-backend-impl.h +4 -1
  99. data/ext/sources/ggml/src/ggml-backend-reg.cpp +35 -13
  100. data/ext/sources/ggml/src/ggml-backend.cpp +266 -60
  101. data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +4 -4
  102. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +5 -4
  103. data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +15 -0
  104. data/ext/sources/ggml/src/ggml-cann/acl_tensor.cpp +3 -1
  105. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +903 -717
  106. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +143 -25
  107. data/ext/sources/ggml/src/ggml-cann/common.h +149 -2
  108. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +521 -78
  109. data/ext/sources/ggml/src/ggml-common.h +21 -0
  110. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +165 -50
  111. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +5 -3
  112. data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +11 -10
  113. data/ext/sources/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +94 -0
  114. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +3650 -0
  115. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +1891 -0
  116. data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +2160 -0
  117. data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/cpu-feats.cpp +82 -0
  118. data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +2305 -0
  119. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +1897 -0
  120. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +342 -0
  121. data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +1468 -0
  122. data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +1221 -0
  123. data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +3820 -0
  124. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +6307 -0
  125. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +214 -0
  126. data/ext/sources/ggml/src/ggml-cpu/common.h +18 -3
  127. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +23 -7
  128. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +179 -110
  129. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +44 -33
  130. data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-hbm.cpp → hbm.cpp} +1 -1
  131. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +152 -18
  132. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +7 -1
  133. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +228 -98
  134. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +532 -1124
  135. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.h +5 -0
  136. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +3374 -2081
  137. data/ext/sources/ggml/src/ggml-cpu/ops.h +13 -8
  138. data/ext/sources/ggml/src/ggml-cpu/quants.c +1193 -0
  139. data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-quants.h → quants.h} +34 -0
  140. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +1982 -0
  141. data/ext/sources/ggml/src/ggml-cpu/repack.h +120 -0
  142. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +367 -46
  143. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1024 -0
  144. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
  145. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
  146. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
  147. data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-traits.cpp → traits.cpp} +3 -3
  148. data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-traits.h → traits.h} +1 -1
  149. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +272 -35
  150. data/ext/sources/ggml/src/ggml-cpu/vec.h +794 -142
  151. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +20 -16
  152. data/ext/sources/ggml/src/ggml-cuda/add-id.cu +58 -0
  153. data/ext/sources/ggml/src/ggml-cuda/add-id.cuh +3 -0
  154. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +330 -191
  155. data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +2 -0
  156. data/ext/sources/ggml/src/ggml-cuda/common.cuh +291 -81
  157. data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cu +1 -4
  158. data/ext/sources/ggml/src/ggml-cuda/conv2d-dw.cu +161 -0
  159. data/ext/sources/ggml/src/ggml-cuda/conv2d-dw.cuh +5 -0
  160. data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cu +91 -0
  161. data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cuh +4 -0
  162. data/ext/sources/ggml/src/ggml-cuda/conv2d.cu +166 -0
  163. data/ext/sources/ggml/src/ggml-cuda/conv2d.cuh +5 -0
  164. data/ext/sources/ggml/src/ggml-cuda/convert.cu +117 -22
  165. data/ext/sources/ggml/src/ggml-cuda/convert.cuh +20 -0
  166. data/ext/sources/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
  167. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +64 -307
  168. data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cu +2 -14
  169. data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +14 -40
  170. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +499 -368
  171. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +142 -93
  172. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +755 -0
  173. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +3 -0
  174. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +593 -0
  175. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +90 -50
  176. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +185 -198
  177. data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
  178. data/ext/sources/ggml/src/ggml-cuda/getrows.cu +50 -39
  179. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +636 -222
  180. data/ext/sources/ggml/src/ggml-cuda/im2col.cu +196 -35
  181. data/ext/sources/ggml/src/ggml-cuda/im2col.cuh +1 -0
  182. data/ext/sources/ggml/src/ggml-cuda/mean.cu +73 -0
  183. data/ext/sources/ggml/src/ggml-cuda/mean.cuh +3 -0
  184. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +198 -45
  185. data/ext/sources/ggml/src/ggml-cuda/mmf.cu +123 -0
  186. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +496 -0
  187. data/ext/sources/ggml/src/ggml-cuda/mmq.cu +206 -57
  188. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +1262 -721
  189. data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +506 -0
  190. data/ext/sources/ggml/src/ggml-cuda/{mmv.cuh → mmvf.cuh} +4 -5
  191. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +64 -73
  192. data/ext/sources/ggml/src/ggml-cuda/norm.cu +284 -12
  193. data/ext/sources/ggml/src/ggml-cuda/norm.cuh +7 -0
  194. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
  195. data/ext/sources/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
  196. data/ext/sources/ggml/src/ggml-cuda/pad.cu +46 -23
  197. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
  198. data/ext/sources/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
  199. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +12 -10
  200. data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +53 -0
  201. data/ext/sources/ggml/src/ggml-cuda/roll.cu +67 -0
  202. data/ext/sources/ggml/src/ggml-cuda/roll.cuh +5 -0
  203. data/ext/sources/ggml/src/ggml-cuda/rope.cu +21 -27
  204. data/ext/sources/ggml/src/ggml-cuda/scale.cu +14 -11
  205. data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +276 -0
  206. data/ext/sources/ggml/src/ggml-cuda/set-rows.cuh +7 -0
  207. data/ext/sources/ggml/src/ggml-cuda/softcap.cu +34 -0
  208. data/ext/sources/ggml/src/ggml-cuda/softcap.cuh +5 -0
  209. data/ext/sources/ggml/src/ggml-cuda/softmax.cu +126 -59
  210. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +10 -2
  211. data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +322 -98
  212. data/ext/sources/ggml/src/ggml-cuda/sum.cu +6 -10
  213. data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +23 -19
  214. data/ext/sources/ggml/src/ggml-cuda/sumrows.cuh +0 -1
  215. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  216. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  217. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  218. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  219. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  220. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  221. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  222. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  223. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  224. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  225. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  226. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  227. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  228. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  229. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  230. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  231. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  232. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  233. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  234. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  235. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  236. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  237. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  238. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  239. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  240. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  241. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  242. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  243. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  244. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  245. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  246. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  247. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  248. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  249. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  250. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  251. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +21 -18
  252. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  253. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  254. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  255. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  256. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  257. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  258. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  259. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  260. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  261. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  262. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  263. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  264. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  265. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  266. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  267. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  268. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
  269. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +259 -0
  270. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cuh +14 -0
  271. data/ext/sources/ggml/src/ggml-cuda/tsembd.cu +3 -3
  272. data/ext/sources/ggml/src/ggml-cuda/unary.cu +179 -0
  273. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +15 -0
  274. data/ext/sources/ggml/src/ggml-cuda/upscale.cu +92 -6
  275. data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +110 -22
  276. data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +4 -0
  277. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +58 -36
  278. data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +4 -3
  279. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +14 -2
  280. data/ext/sources/ggml/src/ggml-impl.h +229 -175
  281. data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +21 -17
  282. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
  283. data/ext/sources/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  284. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
  285. data/ext/sources/ggml/src/ggml-metal/ggml-metal-context.m +600 -0
  286. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +1376 -0
  287. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +226 -0
  288. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +1308 -0
  289. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +163 -63
  290. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +3158 -0
  291. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +82 -0
  292. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +718 -0
  293. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +3208 -1575
  294. data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +18 -8
  295. data/ext/sources/ggml/src/ggml-musa/mudnn.cuh +2 -2
  296. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +32 -0
  297. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +4430 -792
  298. data/ext/sources/ggml/src/ggml-opencl/kernels/add.cl +107 -0
  299. data/ext/sources/ggml/src/ggml-opencl/kernels/add_id.cl +42 -0
  300. data/ext/sources/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
  301. data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +109 -0
  302. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d.cl +185 -0
  303. data/ext/sources/ggml/src/ggml-opencl/kernels/conv2d_f16_f32.cl +176 -0
  304. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +84 -0
  305. data/ext/sources/ggml/src/ggml-opencl/kernels/div.cl +138 -0
  306. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +370 -0
  307. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +370 -0
  308. data/ext/sources/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +373 -0
  309. data/ext/sources/ggml/src/ggml-opencl/kernels/gelu.cl +27 -0
  310. data/ext/sources/ggml/src/ggml-opencl/kernels/glu.cl +378 -0
  311. data/ext/sources/ggml/src/ggml-opencl/kernels/group_norm.cl +121 -0
  312. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f16.cl +1 -1
  313. data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f32.cl +1 -1
  314. data/ext/sources/ggml/src/ggml-opencl/kernels/mul.cl +73 -0
  315. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
  316. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +132 -0
  317. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +133 -0
  318. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32.cl +189 -0
  319. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
  320. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q4_0_f32_8x_flat.cl +283 -0
  321. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
  322. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
  323. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32.cl +144 -0
  324. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
  325. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
  326. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
  327. data/ext/sources/ggml/src/ggml-opencl/kernels/norm.cl +80 -0
  328. data/ext/sources/ggml/src/ggml-opencl/kernels/pad.cl +30 -0
  329. data/ext/sources/ggml/src/ggml-opencl/kernels/repeat.cl +39 -0
  330. data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +79 -0
  331. data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +3 -2
  332. data/ext/sources/ggml/src/ggml-opencl/kernels/set_rows.cl +189 -0
  333. data/ext/sources/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
  334. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +34 -13
  335. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +34 -13
  336. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f16.cl +34 -13
  337. data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f32.cl +34 -13
  338. data/ext/sources/ggml/src/ggml-opencl/kernels/sub.cl +138 -0
  339. data/ext/sources/ggml/src/ggml-opencl/kernels/sum_rows.cl +39 -0
  340. data/ext/sources/ggml/src/ggml-opencl/kernels/tanh.cl +63 -0
  341. data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +20 -0
  342. data/ext/sources/ggml/src/ggml-opencl/kernels/tsembd.cl +48 -0
  343. data/ext/sources/ggml/src/ggml-opencl/kernels/upscale.cl +120 -0
  344. data/ext/sources/ggml/src/ggml-opt.cpp +97 -41
  345. data/ext/sources/ggml/src/ggml-quants.c +117 -24
  346. data/ext/sources/ggml/src/ggml-quants.h +6 -0
  347. data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +85 -62
  348. data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +3 -3
  349. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +2 -0
  350. data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +9 -0
  351. data/ext/sources/ggml/src/ggml-sycl/binbcast.hpp +6 -0
  352. data/ext/sources/ggml/src/ggml-sycl/common.hpp +20 -48
  353. data/ext/sources/ggml/src/ggml-sycl/concat.cpp +13 -17
  354. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +21 -2
  355. data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +116 -211
  356. data/ext/sources/ggml/src/ggml-sycl/cpy.hpp +213 -1
  357. data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +32 -0
  358. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +700 -1041
  359. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +20 -9
  360. data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +17 -26
  361. data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +2 -96
  362. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +393 -250
  363. data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +1 -1
  364. data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +32 -8
  365. data/ext/sources/ggml/src/ggml-sycl/quantize.hpp +133 -0
  366. data/ext/sources/ggml/src/ggml-sycl/quants.hpp +38 -11
  367. data/ext/sources/ggml/src/ggml-sycl/rope.cpp +125 -21
  368. data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +234 -0
  369. data/ext/sources/ggml/src/ggml-sycl/set_rows.hpp +8 -0
  370. data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +3 -1
  371. data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +3 -0
  372. data/ext/sources/ggml/src/ggml-sycl/tsembd.cpp +4 -3
  373. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +105 -17
  374. data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +36 -32
  375. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +4198 -1145
  376. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +4 -12
  377. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +41 -1
  378. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +42 -0
  379. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +13 -4
  380. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +39 -29
  381. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +349 -0
  382. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +98 -0
  383. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +2 -2
  384. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +66 -12
  385. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp +154 -0
  386. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp +21 -0
  387. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +1 -1
  388. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +2 -1
  389. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +6 -5
  390. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +4 -2
  391. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +32 -0
  392. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +1 -1
  393. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
  394. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +1 -1
  395. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +1 -1
  396. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
  397. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +21 -0
  398. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +69 -24
  399. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +60 -20
  400. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +98 -42
  401. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +64 -27
  402. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +74 -13
  403. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +13 -0
  404. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
  405. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
  406. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
  407. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp +4 -17
  408. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +19 -10
  409. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +25 -15
  410. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.comp +19 -0
  411. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.comp +29 -0
  412. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
  413. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
  414. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +18 -14
  415. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +126 -0
  416. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +65 -1
  417. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +11 -7
  418. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +140 -0
  419. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +144 -531
  420. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +206 -38
  421. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.comp +556 -0
  422. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +12 -5
  423. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +15 -9
  424. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +111 -0
  425. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +22 -0
  426. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +24 -3
  427. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +53 -3
  428. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +9 -0
  429. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +64 -11
  430. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +65 -0
  431. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
  432. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +1 -4
  433. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -9
  434. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -9
  435. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -9
  436. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.comp +5 -0
  437. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +1 -1
  438. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +29 -7
  439. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +4 -0
  440. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +17 -0
  441. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +38 -5
  442. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +9 -0
  443. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +14 -0
  444. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +4 -3
  445. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.comp +101 -9
  446. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +69 -5
  447. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/utils.comp +25 -0
  448. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +338 -71
  449. data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +54 -0
  450. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1558 -0
  451. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add.tmpl.wgsl +44 -0
  452. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_in_place.tmpl.wgsl +41 -0
  453. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
  454. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
  455. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +60 -0
  456. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +124 -0
  457. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
  458. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/memset.wgsl +40 -0
  459. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul.tmpl.wgsl +44 -0
  460. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_in_place.tmpl.wgsl +41 -0
  461. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +907 -0
  462. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +57 -0
  463. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_in_place.wgsl +48 -0
  464. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +81 -0
  465. data/ext/sources/ggml/src/ggml-zdnn/CMakeLists.txt +36 -0
  466. data/ext/sources/ggml/src/ggml-zdnn/common.hpp +59 -0
  467. data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +628 -0
  468. data/ext/sources/ggml/src/ggml-zdnn/mmf.cpp +80 -0
  469. data/ext/sources/ggml/src/ggml-zdnn/mmf.hpp +12 -0
  470. data/ext/sources/ggml/src/ggml-zdnn/utils.cpp +79 -0
  471. data/ext/sources/ggml/src/ggml-zdnn/utils.hpp +19 -0
  472. data/ext/sources/ggml/src/ggml.c +802 -142
  473. data/ext/sources/ggml/src/ggml.cpp +26 -0
  474. data/ext/sources/ggml/src/gguf.cpp +32 -4
  475. data/ext/sources/include/whisper.h +2 -0
  476. data/ext/sources/src/CMakeLists.txt +2 -0
  477. data/ext/sources/src/coreml/whisper-compat.h +10 -0
  478. data/ext/sources/src/coreml/whisper-compat.m +35 -0
  479. data/ext/sources/src/coreml/whisper-decoder-impl.m +1 -0
  480. data/ext/sources/src/coreml/whisper-encoder-impl.m +1 -0
  481. data/ext/sources/src/whisper.cpp +241 -215
  482. data/ext/sources/tests/CMakeLists.txt +8 -1
  483. data/ext/sources/tests/test-vad-full.cpp +3 -3
  484. data/ext/sources/tests/test-vad.cpp +2 -2
  485. data/extsources.rb +15 -9
  486. data/lib/whisper/context.rb +15 -0
  487. data/lib/whisper/model/uri.rb +57 -2
  488. data/lib/whisper/segment.rb +58 -0
  489. data/sig/whisper.rbs +75 -38
  490. data/{tests → test}/helper.rb +1 -12
  491. data/{tests → test}/test_model.rb +9 -0
  492. data/test/test_package.rb +51 -0
  493. data/{tests → test}/test_params.rb +8 -0
  494. data/test/test_segment.rb +146 -0
  495. data/{tests → test}/test_whisper.rb +70 -0
  496. data/whispercpp.gemspec +2 -3
  497. metadata +246 -191
  498. data/ext/sources/.dockerignore +0 -3
  499. data/ext/sources/.github/workflows/bindings-ruby.yml +0 -21
  500. data/ext/sources/ci/run.sh +0 -336
  501. data/ext/sources/close-issue.yml +0 -28
  502. data/ext/sources/ggml/include/ggml-kompute.h +0 -50
  503. data/ext/sources/ggml/src/ggml-amx/CMakeLists.txt +0 -107
  504. data/ext/sources/ggml/src/ggml-amx/common.h +0 -94
  505. data/ext/sources/ggml/src/ggml-amx/ggml-amx.cpp +0 -446
  506. data/ext/sources/ggml/src/ggml-amx/mmq.cpp +0 -2510
  507. data/ext/sources/ggml/src/ggml-amx/mmq.h +0 -17
  508. data/ext/sources/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
  509. data/ext/sources/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
  510. data/ext/sources/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
  511. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
  512. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
  513. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
  514. data/ext/sources/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
  515. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
  516. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
  517. data/ext/sources/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
  518. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +0 -6431
  519. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +0 -8
  520. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-quants.c +0 -13747
  521. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -357
  522. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
  523. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -365
  524. data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
  525. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -482
  526. data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -472
  527. data/ext/sources/ggml/src/ggml-cuda/mmv.cu +0 -336
  528. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
  529. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
  530. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
  531. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
  532. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
  533. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
  534. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
  535. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
  536. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
  537. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
  538. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
  539. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
  540. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
  541. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
  542. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
  543. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
  544. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
  545. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
  546. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
  547. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
  548. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
  549. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
  550. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
  551. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
  552. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
  553. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
  554. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
  555. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
  556. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
  557. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
  558. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
  559. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
  560. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
  561. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
  562. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
  563. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
  564. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
  565. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
  566. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
  567. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
  568. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
  569. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
  570. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
  571. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
  572. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
  573. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
  574. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
  575. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
  576. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
  577. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
  578. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
  579. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
  580. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
  581. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
  582. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
  583. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
  584. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
  585. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
  586. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
  587. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
  588. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
  589. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
  590. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
  591. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
  592. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
  593. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
  594. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
  595. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
  596. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
  597. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
  598. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
  599. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
  600. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
  601. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
  602. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
  603. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
  604. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
  605. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
  606. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
  607. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
  608. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
  609. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
  610. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
  611. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
  612. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
  613. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
  614. data/ext/sources/ggml/src/ggml-kompute/CMakeLists.txt +0 -166
  615. data/ext/sources/ggml/src/ggml-kompute/ggml-kompute.cpp +0 -2251
  616. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/common.comp +0 -112
  617. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +0 -58
  618. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +0 -25
  619. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +0 -52
  620. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +0 -52
  621. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +0 -52
  622. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +0 -52
  623. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +0 -30
  624. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +0 -22
  625. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +0 -17
  626. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +0 -31
  627. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +0 -31
  628. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +0 -38
  629. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +0 -39
  630. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +0 -44
  631. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +0 -52
  632. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +0 -69
  633. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +0 -51
  634. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +0 -33
  635. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +0 -35
  636. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +0 -140
  637. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +0 -106
  638. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +0 -73
  639. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +0 -52
  640. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +0 -28
  641. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +0 -84
  642. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +0 -21
  643. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +0 -53
  644. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +0 -52
  645. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +0 -52
  646. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +0 -52
  647. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +0 -52
  648. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +0 -19
  649. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +0 -23
  650. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +0 -22
  651. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +0 -72
  652. data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +0 -71
  653. data/ext/sources/ggml/src/ggml-metal/ggml-metal.m +0 -5998
  654. data/tests/test_package.rb +0 -46
  655. data/tests/test_segment.rb +0 -74
  656. /data/ext/sources/ggml/src/ggml-cpu/{cpu-feats-x86.cpp → arch/x86/cpu-feats.cpp} +0 -0
  657. /data/ext/sources/ggml/src/ggml-cpu/{ggml-cpu-hbm.h → hbm.h} +0 -0
  658. /data/{tests → test}/jfk_reader/.gitignore +0 -0
  659. /data/{tests → test}/jfk_reader/extconf.rb +0 -0
  660. /data/{tests → test}/jfk_reader/jfk_reader.c +0 -0
  661. /data/{tests → test}/test_callback.rb +0 -0
  662. /data/{tests → test}/test_error.rb +0 -0
  663. /data/{tests → test}/test_vad.rb +0 -0
  664. /data/{tests → test}/test_vad_params.rb +0 -0
@@ -93,6 +93,9 @@ set(VAD_TEST test-vad)
93
93
  add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
94
94
  target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
95
95
  target_link_libraries(${VAD_TEST} PRIVATE common)
96
+ target_compile_definitions(${VAD_TEST} PRIVATE
97
+ VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v5.1.2-ggml.bin"
98
+ SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
96
99
  add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
97
100
  set_tests_properties(${VAD_TEST} PROPERTIES LABELS "unit")
98
101
 
@@ -101,5 +104,9 @@ set(VAD_TEST test-vad-full)
101
104
  add_executable(${VAD_TEST} ${VAD_TEST}.cpp)
102
105
  target_include_directories(${VAD_TEST} PRIVATE ../include ../ggml/include ../examples)
103
106
  target_link_libraries(${VAD_TEST} PRIVATE common)
107
+ target_compile_definitions(${VAD_TEST} PRIVATE
108
+ WHISPER_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/ggml-base.en.bin"
109
+ VAD_MODEL_PATH="${PROJECT_SOURCE_DIR}/models/for-tests-silero-v5.1.2-ggml.bin"
110
+ SAMPLE_PATH="${PROJECT_SOURCE_DIR}/samples/jfk.wav")
104
111
  add_test(NAME ${VAD_TEST} COMMAND ${VAD_TEST})
105
- set_tests_properties(${VAD_TARGET} PROPERTIES LABELS "base;en")
112
+ set_tests_properties(${VAD_TEST} PROPERTIES LABELS "base;en")
@@ -13,9 +13,9 @@
13
13
  #include <cassert>
14
14
 
15
15
  int main() {
16
- std::string whisper_model_path = "../../models/ggml-base.en.bin";
17
- std::string vad_model_path = "../../models/for-tests-silero-v5.1.2-ggml.bin";
18
- std::string sample_path = "../../samples/jfk.wav";
16
+ std::string whisper_model_path = WHISPER_MODEL_PATH;
17
+ std::string vad_model_path = VAD_MODEL_PATH;
18
+ std::string sample_path = SAMPLE_PATH;
19
19
 
20
20
  // Load the sample audio file
21
21
  std::vector<float> pcmf32;
@@ -48,8 +48,8 @@ struct whisper_vad_segments * test_detect_timestamps(
48
48
  }
49
49
 
50
50
  int main() {
51
- std::string vad_model_path = "../../models/for-tests-silero-v5.1.2-ggml.bin";
52
- std::string sample_path = "../../samples/jfk.wav";
51
+ std::string vad_model_path = VAD_MODEL_PATH;
52
+ std::string sample_path = SAMPLE_PATH;
53
53
 
54
54
  // Load the sample audio file
55
55
  std::vector<float> pcmf32;
data/extsources.rb CHANGED
@@ -1,5 +1,10 @@
1
+ require "pathname"
2
+
3
+ root = Pathname("..")/".."
1
4
  ignored_dirs = %w[
2
5
  .devops
6
+ .github
7
+ ci
3
8
  examples/wchess/wchess.wasm
4
9
  examples/whisper.android
5
10
  examples/whisper.android.java
@@ -9,7 +14,7 @@ ignored_dirs = %w[
9
14
  models
10
15
  samples
11
16
  scripts
12
- ]
17
+ ].collect {|dir| root/dir}
13
18
  ignored_files = %w[
14
19
  AUTHORS
15
20
  Makefile
@@ -17,18 +22,19 @@ ignored_files = %w[
17
22
  README_sycl.md
18
23
  .gitignore
19
24
  .gitmodules
25
+ .dockerignore
20
26
  whisper.nvim
21
27
  twitch.sh
22
28
  yt-wsp.sh
29
+ close-issue.yml
23
30
  ]
24
31
 
25
32
  EXTSOURCES =
26
- `git ls-files -z ../..`.split("\x0")
27
- .select {|file|
28
- basename = File.basename(file)
29
-
30
- ignored_dirs.all? {|dir| !file.start_with?("../../#{dir}")} &&
31
- !ignored_files.include?(basename) &&
32
- (file.start_with?("../..") || file.start_with?("../javascript")) &&
33
- (!file.start_with?("../../.github/") || basename == "bindings-ruby.yml")
33
+ `git ls-files -z #{root}`.split("\x0")
34
+ .collect {|file| Pathname(file)}
35
+ .reject {|file|
36
+ ignored_dirs.any? {|dir| file.descend.any? {|desc| desc == dir}} ||
37
+ ignored_files.include?(file.basename.to_path) ||
38
+ (file.descend.to_a[1] != root && file.descend.to_a[1] != Pathname("..")/"javascript")
34
39
  }
40
+ .collect(&:to_path)
@@ -0,0 +1,15 @@
1
+ module Whisper
2
+ class Context
3
+ def to_srt
4
+ each_segment.with_index.reduce("") {|srt, (segment, index)|
5
+ srt << "#{index + 1}\n#{segment.to_srt_cue}\n"
6
+ }
7
+ end
8
+
9
+ def to_webvtt
10
+ each_segment.with_index.reduce("WEBVTT\n\n") {|webvtt, (segment, index)|
11
+ webvtt << "#{index + 1}\n#{segment.to_webvtt_cue}\n"
12
+ }
13
+ end
14
+ end
15
+ end
@@ -94,7 +94,7 @@ module Whisper
94
94
  end
95
95
 
96
96
  def show_progress(current, size)
97
- progress_rate_available = size && $stderr.tty?
97
+ progress_rate_available = size && $stderr.tty? && $stderr.winsize[1] >= line.size
98
98
 
99
99
  unless @prev
100
100
  @prev = Time.now
@@ -130,6 +130,44 @@ module Whisper
130
130
  end
131
131
  end
132
132
 
133
+ class ZipURI < URI
134
+ def cache
135
+ zip_path = super
136
+ dest = unzipped_path
137
+ return if dest.exist? && dest.mtime >= zip_path.mtime
138
+ escaping dest do
139
+ system "unzip", "-q", "-d", zip_path.dirname.to_path, zip_path.to_path, exception: true
140
+ end
141
+ zip_path
142
+ end
143
+
144
+ def clear_cache
145
+ super
146
+ unzipped_path.rmtree if unzipped_path.exist?
147
+ end
148
+
149
+ private
150
+
151
+ def unzipped_path
152
+ cache_path.sub_ext("")
153
+ end
154
+
155
+ def escaping(path)
156
+ escaped = Pathname("#{path}.removing")
157
+ if path.exist?
158
+ escaped.rmtree if escaped.exist?
159
+ path.rename escaped
160
+ end
161
+ yield
162
+ ensure
163
+ if path.exist?
164
+ escaped.rmtree if escaped.exist?
165
+ else
166
+ escaped.rename path if escaped.exist?
167
+ end
168
+ end
169
+ end
170
+
133
171
  @pre_converted_models = %w[
134
172
  tiny
135
173
  tiny.en
@@ -171,8 +209,25 @@ module Whisper
171
209
  @pre_converted_models[name] = URI.new("https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-#{name}.bin")
172
210
  end
173
211
 
212
+ @coreml_compiled_models = %w[
213
+ tiny
214
+ tiny.en
215
+ base
216
+ base.en
217
+ small
218
+ small.en
219
+ medium
220
+ medium.en
221
+ large-v1
222
+ large-v2
223
+ large-v3
224
+ large-v3-turbo
225
+ ].each_with_object({}) do |name, models|
226
+ models[@pre_converted_models[name]] = ZipURI.new("https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-#{name}-encoder.mlmodelc.zip")
227
+ end
228
+
174
229
  class << self
175
- attr_reader :pre_converted_models
230
+ attr_reader :pre_converted_models, :coreml_compiled_models
176
231
  end
177
232
  end
178
233
  end
@@ -0,0 +1,58 @@
1
+ module Whisper
2
+ class Segment
3
+ SRT_ESCAPES = {
4
+ "&" => "&amp;",
5
+ "<" => "&lt;",
6
+ ">" => "&gt;",
7
+ }
8
+ SRT_ESCAPES_RE = Regexp.union(SRT_ESCAPES.keys)
9
+ private_constant :SRT_ESCAPES, :SRT_ESCAPES_RE
10
+
11
+ def to_srt_cue
12
+ "#{srt_start_time} --> #{srt_end_time}\n#{srt_text}\n"
13
+ end
14
+
15
+ def to_webvtt_cue
16
+ "#{webvtt_start_time} --> #{webvtt_end_time}\n#{webvtt_text}\n"
17
+ end
18
+
19
+ private
20
+
21
+ def time_to_a(time)
22
+ sec, decimal_part = time.divmod(1000)
23
+ min, sec = sec.divmod(60)
24
+ hour, min = min.divmod(60)
25
+ [hour, min, sec, decimal_part]
26
+ end
27
+
28
+ def srt_time(time)
29
+ "%02d:%02d:%02d,%03d" % time_to_a(time)
30
+ end
31
+
32
+ def srt_start_time
33
+ srt_time(start_time)
34
+ end
35
+
36
+ def srt_end_time
37
+ srt_time(end_time)
38
+ end
39
+
40
+ def srt_text
41
+ text.gsub(SRT_ESCAPES_RE, SRT_ESCAPES)
42
+ end
43
+
44
+ def webvtt_time(time)
45
+ "%02d:%02d:%02d.%03d" % time_to_a(time)
46
+ end
47
+
48
+ def webvtt_start_time
49
+ webvtt_time(start_time)
50
+ end
51
+
52
+ def webvtt_end_time
53
+ webvtt_time(end_time)
54
+ end
55
+
56
+ alias webvtt_text srt_text
57
+ end
58
+ end
data/sig/whisper.rbs CHANGED
@@ -10,6 +10,7 @@ module Whisper
10
10
  type encoder_begin_callback = ^(Whisper::Context, void, Object user_data) -> void
11
11
  type abort_callback = ^(Whisper::Context, void, Object user_data) -> boolish
12
12
 
13
+ VERSION: String
13
14
  LOG_LEVEL_NONE: Integer
14
15
  LOG_LEVEL_INFO: Integer
15
16
  LOG_LEVEL_WARN: Integer
@@ -22,21 +23,22 @@ module Whisper
22
23
  def self.lang_str: (Integer id) -> String
23
24
  def self.lang_str_full: (Integer id) -> String
24
25
  def self.log_set: (log_callback, Object? user_data) -> log_callback
26
+ def self.system_info_str: () -> String
25
27
 
26
28
  class Context
27
- def self.new: (path | ::URI::HTTP) -> instance
29
+ def self.new: (String | path | ::URI::HTTP) -> instance
28
30
 
29
31
  # transcribe a single file
30
32
  # can emit to a block results
31
33
  #
32
- # params = Whisper::Params.new
33
- # params.duration = 60_000
34
- # whisper.transcribe "path/to/audio.wav", params do |text|
35
- # puts text
36
- # end
34
+ # params = Whisper::Params.new
35
+ # params.duration = 60_000
36
+ # whisper.transcribe "path/to/audio.wav", params do |text|
37
+ # puts text
38
+ # end
37
39
  #
38
- def transcribe: (string, Params) -> self
39
- | (string, Params) { (String) -> void } -> self
40
+ def transcribe: (string, Params, ?n_processors: Integer) -> self
41
+ | (string, Params, ?n_processors: Integer) { (String) -> void } -> self
40
42
 
41
43
  def model_n_vocab: () -> Integer
42
44
  def model_n_audio_ctx: () -> Integer
@@ -49,16 +51,16 @@ module Whisper
49
51
 
50
52
  # Yields each Whisper::Segment:
51
53
  #
52
- # whisper.transcribe("path/to/audio.wav", params)
53
- # whisper.each_segment do |segment|
54
- # puts segment.text
55
- # end
54
+ # whisper.transcribe("path/to/audio.wav", params)
55
+ # whisper.each_segment do |segment|
56
+ # puts segment.text
57
+ # end
56
58
  #
57
59
  # Returns an Enumerator if no block given:
58
60
  #
59
- # whisper.transcribe("path/to/audio.wav", params)
60
- # enum = whisper.each_segment
61
- # enum.to_a # => [#<Whisper::Segment>, ...]
61
+ # whisper.transcribe("path/to/audio.wav", params)
62
+ # enum = whisper.each_segment
63
+ # enum.to_a # => [#<Whisper::Segment>, ...]
62
64
  #
63
65
  def each_segment: { (Segment) -> void } -> void
64
66
  | () -> Enumerator[Segment]
@@ -73,25 +75,25 @@ module Whisper
73
75
 
74
76
  # Start time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
75
77
  #
76
- # full_get_segment_t0(3) # => 1668 (16680 ms)
78
+ # full_get_segment_t0(3) # => 1668 (16680 ms)
77
79
  #
78
80
  def full_get_segment_t0: (Integer) -> Integer
79
81
 
80
82
  # End time of a segment indexed by +segment_index+ in centiseconds (10 times milliseconds).
81
83
  #
82
- # full_get_segment_t1(3) # => 1668 (16680 ms)
84
+ # full_get_segment_t1(3) # => 1668 (16680 ms)
83
85
  #
84
86
  def full_get_segment_t1: (Integer) -> Integer
85
87
 
86
88
  # Whether the next segment indexed by +segment_index+ is predicated as a speaker turn.
87
89
  #
88
- # full_get_segment_speacker_turn_next(3) # => true
90
+ # full_get_segment_speacker_turn_next(3) # => true
89
91
  #
90
92
  def full_get_segment_speaker_turn_next: (Integer) -> (true | false)
91
93
 
92
94
  # Text of a segment indexed by +segment_index+.
93
95
  #
94
- # full_get_segment_text(3) # => "ask not what your country can do for you, ..."
96
+ # full_get_segment_text(3) # => "ask not what your country can do for you, ..."
95
97
  #
96
98
  def full_get_segment_text: (Integer) -> String
97
99
 
@@ -115,6 +117,9 @@ module Whisper
115
117
  def full_parallel: (Params, Array[Float], ?Integer n_samples) -> self
116
118
  | (Params, _Samples, ?Integer n_samples) -> self
117
119
  | (Params, _Samples, ?Integer? n_samples, Integer n_processors) -> self
120
+
121
+ def to_srt: () -> String
122
+ def to_webvtt: () -> String
118
123
  end
119
124
 
120
125
  class Params
@@ -130,6 +135,7 @@ module Whisper
130
135
  ?suppress_blank: boolish,
131
136
  ?suppress_nst: boolish,
132
137
  ?token_timestamps: boolish,
138
+ ?max_len: Integer,
133
139
  ?split_on_word: boolish,
134
140
  ?initial_prompt: string | nil,
135
141
  ?diarize: boolish,
@@ -217,6 +223,12 @@ module Whisper
217
223
  #
218
224
  def token_timestamps: () -> (true | false)
219
225
 
226
+ def max_len=: (Integer) -> Integer
227
+
228
+ # max segment length in characters.
229
+ #
230
+ def max_len: () -> Integer
231
+
220
232
  def split_on_word=: (boolish) -> boolish
221
233
 
222
234
  # If true, split on word rather than on token (when used with max_len).
@@ -281,9 +293,9 @@ module Whisper
281
293
 
282
294
  # Sets new segment callback, called for every newly generated text segment.
283
295
  #
284
- # params.new_segment_callback = ->(context, _, n_new, user_data) {
285
- # # ...
286
- # }
296
+ # params.new_segment_callback = ->(context, _, n_new, user_data) {
297
+ # # ...
298
+ # }
287
299
  #
288
300
  def new_segment_callback=: (new_segment_callback) -> new_segment_callback
289
301
  def new_segment_callback: () -> (new_segment_callback | nil)
@@ -296,9 +308,9 @@ module Whisper
296
308
 
297
309
  # Sets progress callback, called on each progress update.
298
310
  #
299
- # params.new_segment_callback = ->(context, _, progress, user_data) {
300
- # # ...
301
- # }
311
+ # params.new_segment_callback = ->(context, _, progress, user_data) {
312
+ # # ...
313
+ # }
302
314
  #
303
315
  # +progress+ is an Integer between 0 and 100.
304
316
  #
@@ -326,9 +338,9 @@ module Whisper
326
338
 
327
339
  # Sets abort callback, called to check if the process should be aborted.
328
340
  #
329
- # params.abort_callback = ->(user_data) {
330
- # # ...
331
- # }
341
+ # params.abort_callback = ->(user_data) {
342
+ # # ...
343
+ # }
332
344
  #
333
345
  #
334
346
  def abort_callback=: (abort_callback) -> abort_callback
@@ -357,9 +369,9 @@ module Whisper
357
369
 
358
370
  # Hook called on new segment. Yields each Whisper::Segment.
359
371
  #
360
- # whisper.on_new_segment do |segment|
361
- # # ...
362
- # end
372
+ # whisper.on_new_segment do |segment|
373
+ # # ...
374
+ # end
363
375
  #
364
376
  def on_new_segment: { (Segment) -> void } -> void
365
377
 
@@ -373,19 +385,20 @@ module Whisper
373
385
 
374
386
  # Call block to determine whether abort or not. Return +true+ when you want to abort.
375
387
  #
376
- # params.abort_on do
377
- # if some_condition
378
- # true # abort
379
- # else
380
- # false # continue
388
+ # params.abort_on do
389
+ # if some_condition
390
+ # true # abort
391
+ # else
392
+ # false # continue
393
+ # end
381
394
  # end
382
- # end
383
395
  #
384
396
  def abort_on: { (Object user_data) -> boolish } -> void
385
397
  end
386
398
 
387
399
  class Model
388
400
  def self.pre_converted_models: () -> Hash[String, Model::URI]
401
+ def self.coreml_compiled_models: () -> Hash[Model::URI, Model::ZipURI]
389
402
  def self.new: () -> instance
390
403
  def n_vocab: () -> Integer
391
404
  def n_audio_ctx: () -> Integer
@@ -405,9 +418,22 @@ module Whisper
405
418
  def to_path: -> String
406
419
  def clear_cache: -> void
407
420
  end
421
+
422
+ class ZipURI < URI
423
+ def cache: () -> Pathname
424
+ def clear_cache: () -> void
425
+ end
408
426
  end
409
427
 
410
428
  class Segment
429
+ type deconstructed_keys = {
430
+ start_time: (Integer | nil),
431
+ end_time: (Integer | nil),
432
+ text: (String | nil),
433
+ no_speech_prob: (Float | nil),
434
+ speaker_turn_next: (true | false | nil)
435
+ }
436
+
411
437
  # Start time in milliseconds.
412
438
  #
413
439
  def start_time: () -> Integer
@@ -417,10 +443,21 @@ module Whisper
417
443
  def end_time: () -> Integer
418
444
 
419
445
  # Whether the next segment is predicted as a speaker turn.
420
- def speaker_next_turn?: () -> (true | false)
446
+ def speaker_turn_next?: () -> (true | false)
421
447
 
422
448
  def text: () -> String
423
449
  def no_speech_prob: () -> Float
450
+ def to_srt_cue: () -> String
451
+ def to_webvtt_cue: () -> String
452
+
453
+ # Possible keys: :start_time, :end_time, :text, :no_speech_prob, :speaker_turn_next
454
+ #
455
+ # whisper.each_segment do |segment|
456
+ # segment => {start_time:, end_time:, text:, no_speech_prob:, speaker_turn_next:}
457
+ #
458
+ # puts "[#{start_time} --> #{end_time}] #{text} (no speech prob: #{no_speech_prob}#{speaker_turn_next ? ', speaker turns next' : ''})"
459
+ # end
460
+ def deconstruct_keys: (Array[:start_time | :end_time | :text | :no_speech_prob | :speaker_turn_next] | nil) -> deconstructed_keys
424
461
  end
425
462
 
426
463
  module VAD
@@ -3,7 +3,7 @@ require "whisper"
3
3
  require_relative "jfk_reader/jfk_reader"
4
4
 
5
5
  class TestBase < Test::Unit::TestCase
6
- AUDIO = File.join(__dir__, "..", "..", "..", "samples", "jfk.wav")
6
+ AUDIO = File.join(__dir__, "fixtures", "jfk.wav")
7
7
 
8
8
  class << self
9
9
  def whisper
@@ -21,15 +21,4 @@ class TestBase < Test::Unit::TestCase
21
21
  def whisper
22
22
  self.class.whisper
23
23
  end
24
-
25
- module BuildOptions
26
- load "ext/options.rb", self
27
- Options.include self
28
-
29
- def enable_config(name)
30
- end
31
-
32
- def arg_config(name)
33
- end
34
- end
35
24
  end
@@ -106,4 +106,13 @@ class TestModel < TestBase
106
106
  assert_equal 1, model.ftype
107
107
  assert_equal "base", model.type
108
108
  end
109
+
110
+ def test_coreml_model_auto_download
111
+ uri = Whisper::Model.coreml_compiled_models[Whisper::Model.pre_converted_models["tiny"]]
112
+ model_path = Pathname(uri.to_path).sub_ext("")
113
+ model_path.rmtree if model_path.exist?
114
+
115
+ uri.cache
116
+ assert_path_exist model_path
117
+ end
109
118
  end
@@ -0,0 +1,51 @@
1
+ require_relative "helper"
2
+ require 'tempfile'
3
+ require 'tmpdir'
4
+ require 'shellwords'
5
+
6
+ class TestPackage < TestBase
7
+ def test_build
8
+ Tempfile.create do |file|
9
+ assert system("gem", "build", "whispercpp.gemspec", "--output", file.to_path.shellescape, exception: true)
10
+ assert file.size > 0
11
+ assert_path_exist file.to_path
12
+ end
13
+ end
14
+
15
+ sub_test_case "Building binary on installation" do
16
+ def setup
17
+ system "rake", "build", exception: true
18
+ end
19
+
20
+ def test_install
21
+ gemspec = Gem::Specification.load("whispercpp.gemspec")
22
+ Dir.mktmpdir do |dir|
23
+ system "gem", "install", "--install-dir", dir.shellescape, "--no-document", "pkg/#{gemspec.file_name.shellescape}", exception: true
24
+ assert_installed dir, gemspec.version
25
+ end
26
+ end
27
+
28
+ def test_install_with_coreml
29
+ omit_unless RUBY_PLATFORM.match?(/darwin/) do
30
+ gemspec = Gem::Specification.load("whispercpp.gemspec")
31
+ Dir.mktmpdir do |dir|
32
+ system "gem", "install", "--install-dir", dir.shellescape, "--no-document", "pkg/#{gemspec.file_name.shellescape}", "--", "--enable-whisper-coreml", exception: true
33
+ assert_installed dir, gemspec.version
34
+ libdir = File.join(dir, "gems", "#{gemspec.name}-#{gemspec.version}", "lib")
35
+ assert_nothing_raised do
36
+ system "ruby", "-I", libdir, "-r", "whisper", "-e", "Whisper::Context.new('tiny')", exception: true
37
+ end
38
+ assert_match(/COREML = 1/, `ruby -I #{libdir.shellescape} -r whisper -e 'puts Whisper.system_info_str'`)
39
+ end
40
+ end
41
+ end
42
+
43
+ private
44
+
45
+ def assert_installed(dir, version)
46
+ assert_path_exist File.join(dir, "gems/whispercpp-#{version}/lib", "whisper.#{RbConfig::CONFIG["DLEXT"]}")
47
+ assert_path_exist File.join(dir, "gems/whispercpp-#{version}/LICENSE")
48
+ assert_path_not_exist File.join(dir, "gems/whispercpp-#{version}/ext/build")
49
+ end
50
+ end
51
+ end
@@ -13,6 +13,7 @@ class TestParams < TestBase
13
13
  :suppress_blank,
14
14
  :suppress_nst,
15
15
  :token_timestamps,
16
+ :max_len,
16
17
  :split_on_word,
17
18
  :initial_prompt,
18
19
  :diarize,
@@ -139,6 +140,13 @@ class TestParams < TestBase
139
140
  assert !@params.token_timestamps
140
141
  end
141
142
 
143
+ def test_max_len
144
+ @params.max_len = 42
145
+ assert_equal @params.max_len, 42
146
+ @params.max_len = 0
147
+ assert_equal @params.max_len, 0
148
+ end
149
+
142
150
  def test_split_on_word
143
151
  @params.split_on_word = true
144
152
  assert @params.split_on_word