whispercpp 1.3.6 → 1.3.7

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 (828) hide show
  1. checksums.yaml +4 -4
  2. data/.document +3 -0
  3. data/.rdoc_options +2 -0
  4. data/README.md +38 -5
  5. data/Rakefile +18 -3
  6. data/ext/dependencies.rb +10 -4
  7. data/ext/dependencies_for_windows.rb +17 -0
  8. data/ext/extconf.rb +20 -8
  9. data/ext/options.rb +54 -14
  10. data/ext/options_for_windows.rb +51 -0
  11. data/ext/ruby_whisper.c +36 -42
  12. data/ext/ruby_whisper.h +135 -0
  13. data/ext/ruby_whisper_context.c +107 -28
  14. data/ext/ruby_whisper_log_queue.c +180 -0
  15. data/ext/ruby_whisper_log_settable.h +47 -0
  16. data/ext/ruby_whisper_parakeet.c +49 -0
  17. data/ext/ruby_whisper_parakeet_context.c +304 -0
  18. data/ext/ruby_whisper_parakeet_context_params.c +117 -0
  19. data/ext/ruby_whisper_parakeet_model.c +84 -0
  20. data/ext/ruby_whisper_parakeet_params.c +548 -0
  21. data/ext/ruby_whisper_parakeet_segment.c +157 -0
  22. data/ext/ruby_whisper_parakeet_token.c +188 -0
  23. data/ext/ruby_whisper_parakeet_transcribe.cpp +58 -0
  24. data/ext/ruby_whisper_params.c +256 -65
  25. data/ext/ruby_whisper_segment.c +6 -6
  26. data/ext/ruby_whisper_transcribe.cpp +42 -15
  27. data/ext/sources/CMakeLists.txt +41 -3
  28. data/ext/sources/CMakePresets.json +95 -0
  29. data/ext/sources/cmake/parakeet-config.cmake.in +30 -0
  30. data/ext/sources/cmake/parakeet.pc.in +10 -0
  31. data/ext/sources/cmake/whisper.pc.in +1 -1
  32. data/ext/sources/examples/CMakeLists.txt +4 -2
  33. data/ext/sources/examples/bench/bench.cpp +1 -1
  34. data/ext/sources/examples/cli/cli.cpp +43 -9
  35. data/ext/sources/examples/common-ggml.cpp +2 -0
  36. data/ext/sources/examples/common-whisper.cpp +139 -67
  37. data/ext/sources/examples/common-whisper.h +11 -0
  38. data/ext/sources/examples/ffmpeg-transcode.cpp +211 -341
  39. data/ext/sources/examples/parakeet-cli/CMakeLists.txt +8 -0
  40. data/ext/sources/examples/parakeet-cli/parakeet-cli.cpp +243 -0
  41. data/ext/sources/examples/parakeet-quantize/CMakeLists.txt +7 -0
  42. data/ext/sources/examples/parakeet-quantize/parakeet-quantize.cpp +230 -0
  43. data/ext/sources/examples/server/server.cpp +199 -163
  44. data/ext/sources/ggml/CMakeLists.txt +21 -13
  45. data/ext/sources/ggml/cmake/FindNCCL.cmake +36 -0
  46. data/ext/sources/ggml/cmake/ggml-config.cmake.in +12 -2
  47. data/ext/sources/ggml/include/ggml-alloc.h +1 -0
  48. data/ext/sources/ggml/include/ggml-backend.h +72 -10
  49. data/ext/sources/ggml/include/ggml-cuda.h +3 -0
  50. data/ext/sources/ggml/include/ggml-rpc.h +3 -3
  51. data/ext/sources/ggml/include/ggml.h +101 -9
  52. data/ext/sources/ggml/include/gguf.h +10 -2
  53. data/ext/sources/ggml/src/CMakeLists.txt +22 -5
  54. data/ext/sources/ggml/src/ggml-alloc.c +5 -1
  55. data/ext/sources/ggml/src/ggml-backend-impl.h +22 -2
  56. data/ext/sources/ggml/src/ggml-backend-meta.cpp +2263 -0
  57. data/ext/sources/ggml/src/ggml-backend-reg.cpp +12 -0
  58. data/ext/sources/ggml/src/ggml-backend.cpp +110 -9
  59. data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +4 -0
  60. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +672 -257
  61. data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +71 -0
  62. data/ext/sources/ggml/src/ggml-cann/common.h +20 -10
  63. data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +211 -30
  64. data/ext/sources/ggml/src/ggml-common.h +11 -0
  65. data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +58 -29
  66. data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +2 -0
  67. data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +16 -16
  68. data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +116 -7
  69. data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +65 -0
  70. data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +151 -1
  71. data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +0 -1
  72. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +4279 -1292
  73. data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +5 -35
  74. data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +0 -1
  75. data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +72 -1
  76. data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +177 -27
  77. data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +1 -1
  78. data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +5 -0
  79. data/ext/sources/ggml/src/ggml-cpu/cmake/FindSMTIME.cmake +32 -0
  80. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +10 -0
  81. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +95 -5
  82. data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +2 -0
  83. data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +146 -134
  84. data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +88 -70
  85. data/ext/sources/ggml/src/ggml-cpu/ops.cpp +372 -73
  86. data/ext/sources/ggml/src/ggml-cpu/ops.h +3 -0
  87. data/ext/sources/ggml/src/ggml-cpu/quants.c +55 -0
  88. data/ext/sources/ggml/src/ggml-cpu/quants.h +3 -0
  89. data/ext/sources/ggml/src/ggml-cpu/repack.cpp +3 -0
  90. data/ext/sources/ggml/src/ggml-cpu/simd-gemm.h +90 -0
  91. data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +3 -16
  92. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1402 -687
  93. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +8 -0
  94. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +597 -2766
  95. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime2_kernels.cpp +5768 -0
  96. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.cpp +320 -0
  97. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.h +55 -0
  98. data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +182 -19
  99. data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.cpp +1795 -0
  100. data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.h +14 -0
  101. data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.cpp +3178 -0
  102. data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.h +95 -0
  103. data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_barrier.h +34 -0
  104. data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.cpp +760 -0
  105. data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.h +32 -0
  106. data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_tcm.h +409 -0
  107. data/ext/sources/ggml/src/ggml-cpu/vec.cpp +37 -53
  108. data/ext/sources/ggml/src/ggml-cpu/vec.h +225 -240
  109. data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +17 -7
  110. data/ext/sources/ggml/src/ggml-cuda/allreduce.cu +971 -0
  111. data/ext/sources/ggml/src/ggml-cuda/allreduce.cuh +29 -0
  112. data/ext/sources/ggml/src/ggml-cuda/argsort.cu +62 -26
  113. data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +44 -18
  114. data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +1 -0
  115. data/ext/sources/ggml/src/ggml-cuda/common.cuh +242 -28
  116. data/ext/sources/ggml/src/ggml-cuda/concat.cu +120 -114
  117. data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cu +45 -21
  118. data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cuh +1 -0
  119. data/ext/sources/ggml/src/ggml-cuda/convert.cu +53 -0
  120. data/ext/sources/ggml/src/ggml-cuda/convert.cuh +10 -0
  121. data/ext/sources/ggml/src/ggml-cuda/cpy.cu +14 -6
  122. data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +22 -0
  123. data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +278 -44
  124. data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +331 -130
  125. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +12 -0
  126. data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +126 -27
  127. data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +40 -15
  128. data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +18 -9
  129. data/ext/sources/ggml/src/ggml-cuda/fattn.cu +152 -49
  130. data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
  131. data/ext/sources/ggml/src/ggml-cuda/fwht.cu +101 -0
  132. data/ext/sources/ggml/src/ggml-cuda/fwht.cuh +4 -0
  133. data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cu +84 -35
  134. data/ext/sources/ggml/src/ggml-cuda/getrows.cu +34 -12
  135. data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +1069 -609
  136. data/ext/sources/ggml/src/ggml-cuda/im2col.cu +32 -29
  137. data/ext/sources/ggml/src/ggml-cuda/mean.cu +4 -2
  138. data/ext/sources/ggml/src/ggml-cuda/mma.cuh +242 -195
  139. data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +3 -3
  140. data/ext/sources/ggml/src/ggml-cuda/mmq.cu +18 -12
  141. data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +502 -423
  142. data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +19 -12
  143. data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +485 -57
  144. data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +6 -1
  145. data/ext/sources/ggml/src/ggml-cuda/norm.cu +36 -10
  146. data/ext/sources/ggml/src/ggml-cuda/out-prod.cu +23 -7
  147. data/ext/sources/ggml/src/ggml-cuda/quantize.cu +133 -26
  148. data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +1 -1
  149. data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +5 -1
  150. data/ext/sources/ggml/src/ggml-cuda/rope.cu +11 -4
  151. data/ext/sources/ggml/src/ggml-cuda/scale.cu +4 -1
  152. data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +14 -6
  153. data/ext/sources/ggml/src/ggml-cuda/snake.cu +72 -0
  154. data/ext/sources/ggml/src/ggml-cuda/snake.cuh +8 -0
  155. data/ext/sources/ggml/src/ggml-cuda/softcap.cu +4 -1
  156. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +45 -13
  157. data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cuh +1 -1
  158. data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +40 -18
  159. data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +8 -4
  160. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +1 -0
  161. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +1 -0
  162. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +2 -0
  163. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +1 -0
  164. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +1 -0
  165. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +1 -0
  166. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +1 -0
  167. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +2 -0
  168. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +1 -0
  169. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +1 -0
  170. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +2 -0
  171. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +1 -0
  172. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +2 -0
  173. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq192-dv128.cu +5 -0
  174. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq320-dv256.cu +5 -0
  175. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq512-dv512.cu +5 -0
  176. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu +7 -0
  177. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-f16.cu +7 -0
  178. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_0.cu +7 -0
  179. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_1.cu +7 -0
  180. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_0.cu +7 -0
  181. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_1.cu +7 -0
  182. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q8_0.cu +7 -0
  183. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-bf16.cu +7 -0
  184. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-bf16.cu +7 -0
  185. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-bf16.cu +7 -0
  186. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-bf16.cu +7 -0
  187. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-bf16.cu +7 -0
  188. data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-bf16.cu +7 -0
  189. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-nvfp4.cu +5 -0
  190. data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q1_0.cu +5 -0
  191. data/ext/sources/ggml/src/ggml-cuda/top-k.cu +5 -4
  192. data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +26 -23
  193. data/ext/sources/ggml/src/ggml-cuda/unary.cu +31 -2
  194. data/ext/sources/ggml/src/ggml-cuda/unary.cuh +2 -0
  195. data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +80 -0
  196. data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +7 -2
  197. data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +22 -4
  198. data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +3 -0
  199. data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +2 -1
  200. data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +1428 -743
  201. data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +45 -7
  202. data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +53 -84
  203. data/ext/sources/ggml/src/ggml-hexagon/htp/argsort-ops.c +25 -12
  204. data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +165 -184
  205. data/ext/sources/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +5 -5
  206. data/ext/sources/ggml/src/ggml-hexagon/htp/concat-ops.c +277 -0
  207. data/ext/sources/ggml/src/ggml-hexagon/htp/cpy-ops.c +170 -127
  208. data/ext/sources/ggml/src/ggml-hexagon/htp/cumsum-ops.c +270 -0
  209. data/ext/sources/ggml/src/ggml-hexagon/htp/diag-ops.c +216 -0
  210. data/ext/sources/ggml/src/ggml-hexagon/htp/fill-ops.c +123 -0
  211. data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +125 -97
  212. data/ext/sources/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c +1148 -0
  213. data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +148 -42
  214. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dma.c +2 -2
  215. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dma.h +252 -62
  216. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dump.h +9 -0
  217. data/ext/sources/ggml/src/ggml-hexagon/htp/hex-utils.h +87 -1
  218. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-flash-attn-ops.c +1878 -0
  219. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-matmul-ops.c +2066 -0
  220. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.c +6 -0
  221. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.h +88 -0
  222. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-profile.h +34 -0
  223. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.c +158 -0
  224. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.h +134 -0
  225. data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-utils.h +200 -0
  226. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +96 -13
  227. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +182 -57
  228. data/ext/sources/ggml/src/ggml-hexagon/htp/htp_iface.idl +9 -3
  229. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-base.h +71 -3
  230. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-copy.h +27 -10
  231. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-div.h +63 -23
  232. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.h +9 -8
  233. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-flash-attn.h +47 -0
  234. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-log.h +65 -0
  235. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-pow.h +42 -0
  236. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-repl.h +74 -0
  237. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +1 -0
  238. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sin-cos.h +90 -0
  239. data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +5 -8
  240. data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +529 -815
  241. data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +2522 -234
  242. data/ext/sources/ggml/src/ggml-hexagon/htp/pad-ops.c +547 -0
  243. data/ext/sources/ggml/src/ggml-hexagon/htp/repeat-ops.c +148 -0
  244. data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +291 -95
  245. data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +59 -37
  246. data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +121 -133
  247. data/ext/sources/ggml/src/ggml-hexagon/htp/solve-tri-ops.c +267 -0
  248. data/ext/sources/ggml/src/ggml-hexagon/htp/ssm-conv.c +244 -151
  249. data/ext/sources/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +6 -6
  250. data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +719 -45
  251. data/ext/sources/ggml/src/ggml-hexagon/htp/vtcm-utils.h +16 -0
  252. data/ext/sources/ggml/src/ggml-hexagon/htp-opnode.h +272 -0
  253. data/ext/sources/ggml/src/ggml-hexagon/libggml-htp.inf +3 -1
  254. data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +22 -9
  255. data/ext/sources/ggml/src/ggml-impl.h +6 -1
  256. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +138 -13
  257. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +32 -1
  258. data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +164 -28
  259. data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +80 -0
  260. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +190 -19
  261. data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +2 -0
  262. data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +39 -26
  263. data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +823 -322
  264. data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +5 -6
  265. data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +54 -5
  266. data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +12248 -5907
  267. data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +67 -0
  268. data/ext/sources/ggml/src/ggml-opencl/kernels/cpy.cl +59 -0
  269. data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +1819 -112
  270. data/ext/sources/ggml/src/ggml-opencl/kernels/gated_delta_net.cl +249 -0
  271. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl +306 -0
  272. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_0_f32_ns.cl +256 -0
  273. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_1_f32_ns.cl +258 -0
  274. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_k_f32_ns.cl +283 -0
  275. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_0_f32_ns.cl +260 -0
  276. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_1_f32_ns.cl +262 -0
  277. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_k_f32_ns.cl +288 -0
  278. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q6_k_f32_ns.cl +267 -0
  279. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_iq4_nl_f32.cl +150 -0
  280. data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mat_Ab_Bi_8x4.cl → gemm_noshuffle_q4_0_f32.cl} +1 -1
  281. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_k_f32.cl +172 -0
  282. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_0_f32.cl +131 -0
  283. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_1_f32.cl +134 -0
  284. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_k_f32.cl +176 -0
  285. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q6_k_f32.cl +140 -0
  286. data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mm_q8_0_f32_8x4.cl → gemm_noshuffle_q8_0_f32.cl} +1 -1
  287. data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_xmem_f16_f32_os8.cl +233 -0
  288. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl +165 -0
  289. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_0_f32_ns.cl +120 -0
  290. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_1_f32_ns.cl +123 -0
  291. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_k_f32_ns.cl +155 -0
  292. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_0_f32_ns.cl +123 -0
  293. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_1_f32_ns.cl +125 -0
  294. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_k_f32_ns.cl +160 -0
  295. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q6_k_f32_ns.cl +141 -0
  296. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_iq4_nl_f32.cl +302 -0
  297. data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle_general.cl → gemv_noshuffle_q4_0_f32.cl} +5 -5
  298. data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle.cl → gemv_noshuffle_q4_0_f32_spec.cl} +5 -5
  299. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_k_f32.cl +318 -0
  300. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_0_f32.cl +291 -0
  301. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_1_f32.cl +294 -0
  302. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_k_f32.cl +326 -0
  303. data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q6_k_f32.cl +293 -0
  304. data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +15 -9
  305. data/ext/sources/ggml/src/ggml-opencl/kernels/moe_reorder_b.cl +30 -0
  306. data/ext/sources/ggml/src/ggml-opencl/kernels/moe_sort_by_expert.cl +82 -0
  307. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_iq4_nl_f32_l4_lm.cl +171 -0
  308. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_k_f32_l4_lm.cl +179 -0
  309. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_0_f32_l4_lm.cl +173 -0
  310. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_1_f32_l4_lm.cl +175 -0
  311. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_k_f32_l4_lm.cl +192 -0
  312. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32.cl +164 -0
  313. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32_flat.cl +202 -0
  314. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32_flat.cl +196 -0
  315. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32.cl +241 -0
  316. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32_flat.cl +243 -0
  317. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32.cl +243 -0
  318. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32_flat.cl +247 -0
  319. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32.cl +187 -0
  320. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32_flat.cl +203 -0
  321. data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +48 -64
  322. data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.cpp +15 -5
  323. data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.cpp +18 -11
  324. data/ext/sources/ggml/src/ggml-openvino/ggml-openvino.cpp +35 -13
  325. data/ext/sources/ggml/src/ggml-openvino/ggml-quants.cpp +264 -192
  326. data/ext/sources/ggml/src/ggml-openvino/openvino/op/rope.cpp +33 -7
  327. data/ext/sources/ggml/src/ggml-openvino/openvino/op/unary_gelu.cpp +25 -0
  328. data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.cpp +1 -0
  329. data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.h +1 -0
  330. data/ext/sources/ggml/src/ggml-openvino/openvino/rt_info/weightless_caching_attributes.hpp +41 -0
  331. data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.cpp +27 -3
  332. data/ext/sources/ggml/src/ggml-openvino/openvino/utils.cpp +67 -36
  333. data/ext/sources/ggml/src/ggml-openvino/openvino/utils.h +1 -0
  334. data/ext/sources/ggml/src/ggml-openvino/utils.cpp +101 -44
  335. data/ext/sources/ggml/src/ggml-openvino/utils.h +23 -3
  336. data/ext/sources/ggml/src/ggml-opt.cpp +1 -0
  337. data/ext/sources/ggml/src/ggml-quants.c +289 -114
  338. data/ext/sources/ggml/src/ggml-quants.h +3 -0
  339. data/ext/sources/ggml/src/ggml-rpc/CMakeLists.txt +24 -0
  340. data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +167 -311
  341. data/ext/sources/ggml/src/ggml-rpc/transport.cpp +683 -0
  342. data/ext/sources/ggml/src/ggml-rpc/transport.h +34 -0
  343. data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +50 -4
  344. data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +1 -1
  345. data/ext/sources/ggml/src/ggml-sycl/backend.hpp +3 -1
  346. data/ext/sources/ggml/src/ggml-sycl/common.cpp +74 -2
  347. data/ext/sources/ggml/src/ggml-sycl/common.hpp +41 -1
  348. data/ext/sources/ggml/src/ggml-sycl/convert.cpp +115 -13
  349. data/ext/sources/ggml/src/ggml-sycl/convert.hpp +9 -0
  350. data/ext/sources/ggml/src/ggml-sycl/cumsum.cpp +148 -0
  351. data/ext/sources/ggml/src/ggml-sycl/cumsum.hpp +5 -0
  352. data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +663 -0
  353. data/ext/sources/ggml/src/ggml-sycl/diag.cpp +67 -0
  354. data/ext/sources/ggml/src/ggml-sycl/diag.hpp +5 -0
  355. data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +586 -6
  356. data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +1 -90
  357. data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +0 -2
  358. data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.cpp +56 -0
  359. data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.hpp +63 -0
  360. data/ext/sources/ggml/src/ggml-sycl/fattn-common.hpp +7 -5
  361. data/ext/sources/ggml/src/ggml-sycl/fattn-tile.cpp +4 -0
  362. data/ext/sources/ggml/src/ggml-sycl/fattn-tile.hpp +76 -168
  363. data/ext/sources/ggml/src/ggml-sycl/fattn-vec.hpp +7 -0
  364. data/ext/sources/ggml/src/ggml-sycl/fattn.cpp +3 -1
  365. data/ext/sources/ggml/src/ggml-sycl/fill.cpp +55 -0
  366. data/ext/sources/ggml/src/ggml-sycl/fill.hpp +5 -0
  367. data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.cpp +69 -31
  368. data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.hpp +1 -0
  369. data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +3 -0
  370. data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +79 -3
  371. data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +823 -190
  372. data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +353 -89
  373. data/ext/sources/ggml/src/ggml-sycl/im2col.hpp +5 -3
  374. data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +1344 -26
  375. data/ext/sources/ggml/src/ggml-sycl/mmvq.hpp +16 -0
  376. data/ext/sources/ggml/src/ggml-sycl/pad.cpp +27 -27
  377. data/ext/sources/ggml/src/ggml-sycl/quants.hpp +71 -0
  378. data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +7 -1
  379. data/ext/sources/ggml/src/ggml-sycl/solve_tri.cpp +172 -0
  380. data/ext/sources/ggml/src/ggml-sycl/solve_tri.hpp +8 -0
  381. data/ext/sources/ggml/src/ggml-sycl/ssm_conv.cpp +6 -1
  382. data/ext/sources/ggml/src/ggml-sycl/ssm_scan.cpp +156 -0
  383. data/ext/sources/ggml/src/ggml-sycl/ssm_scan.hpp +5 -0
  384. data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +62 -10
  385. data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +18 -6
  386. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq512-dv512.cpp +6 -0
  387. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-f16.cpp +1 -0
  388. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_0.cpp +1 -0
  389. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_1.cpp +1 -0
  390. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_0.cpp +1 -0
  391. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_1.cpp +1 -0
  392. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q8_0.cpp +1 -0
  393. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-f16.cpp +1 -0
  394. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_0.cpp +1 -0
  395. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_1.cpp +1 -0
  396. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_0.cpp +1 -0
  397. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_1.cpp +1 -0
  398. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q8_0.cpp +1 -0
  399. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-f16.cpp +1 -0
  400. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_0.cpp +1 -0
  401. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_1.cpp +1 -0
  402. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_0.cpp +1 -0
  403. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_1.cpp +1 -0
  404. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q8_0.cpp +1 -0
  405. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-f16.cpp +1 -0
  406. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_0.cpp +1 -0
  407. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_1.cpp +1 -0
  408. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_0.cpp +1 -0
  409. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_1.cpp +1 -0
  410. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q8_0.cpp +1 -0
  411. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-f16.cpp +1 -0
  412. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_0.cpp +1 -0
  413. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_1.cpp +1 -0
  414. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_0.cpp +1 -0
  415. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_1.cpp +1 -0
  416. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q8_0.cpp +1 -0
  417. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-f16.cpp +1 -0
  418. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_0.cpp +1 -0
  419. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_1.cpp +1 -0
  420. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_0.cpp +1 -0
  421. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_1.cpp +1 -0
  422. data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q8_0.cpp +1 -0
  423. data/ext/sources/ggml/src/ggml-sycl/type.hpp +112 -0
  424. data/ext/sources/ggml/src/ggml-sycl/upscale.cpp +410 -0
  425. data/ext/sources/ggml/src/ggml-sycl/upscale.hpp +9 -0
  426. data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +215 -53
  427. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +4 -0
  428. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +2 -0
  429. data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend.cpp +2 -0
  430. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +1 -0
  431. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.cpp +1 -0
  432. data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.h +0 -2
  433. data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +11 -0
  434. data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +2060 -535
  435. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +4 -0
  436. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +6 -2
  437. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +146 -13
  438. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +3 -1
  439. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +1 -1
  440. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +25 -1
  441. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl +88 -0
  442. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl +643 -1
  443. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_nvfp4.comp +32 -0
  444. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q1_0.comp +29 -0
  445. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +0 -1
  446. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dot_product_funcs.glsl +27 -0
  447. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +0 -1
  448. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/coopmat2_decode_vector.comp +7 -0
  449. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +197 -48
  450. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +60 -59
  451. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +115 -113
  452. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +122 -31
  453. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_dequant.glsl +131 -0
  454. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mmq_funcs.glsl +203 -0
  455. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/fwht.comp +115 -0
  456. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gated_delta_net.comp +125 -64
  457. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.glsl +0 -1
  458. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl +10 -1
  459. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl +16 -6
  460. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +76 -54
  461. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +0 -1
  462. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +0 -1
  463. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +122 -27
  464. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +6 -6
  465. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +1 -1
  466. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +1 -1
  467. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +1 -1
  468. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +1 -0
  469. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +88 -55
  470. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +11 -17
  471. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +43 -10
  472. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +159 -125
  473. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +8 -8
  474. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +24 -9
  475. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +0 -1
  476. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +5 -2
  477. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +0 -1
  478. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +3 -2
  479. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/snake.comp +49 -0
  480. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +11 -1
  481. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +0 -1
  482. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl +79 -2
  483. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +171 -147
  484. data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +5 -2
  485. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +2202 -283
  486. data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +2610 -1403
  487. data/ext/sources/ggml/src/ggml-webgpu/pre_wgsl.hpp +37 -7
  488. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_id.wgsl +64 -0
  489. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +8 -7
  490. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +76 -95
  491. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/concat.wgsl +19 -1
  492. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/conv2d.wgsl +165 -0
  493. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{cpy.tmpl.wgsl → cpy.wgsl} +25 -50
  494. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +107 -184
  495. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_quant_staging.tmpl +124 -0
  496. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_tile.wgsl +397 -0
  497. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_blk.wgsl +101 -0
  498. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_reduce.wgsl +84 -0
  499. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_split.wgsl +619 -0
  500. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/gated_delta_net.wgsl +149 -0
  501. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.wgsl +183 -78
  502. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.wgsl +155 -0
  503. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/im2col.wgsl +101 -0
  504. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +655 -495
  505. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id.wgsl +195 -0
  506. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_gather.wgsl +52 -0
  507. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_vec.wgsl +154 -0
  508. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl +8 -6
  509. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.wgsl +5 -1
  510. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +80 -409
  511. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_acc.tmpl +1432 -0
  512. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_q_acc.tmpl +303 -0
  513. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quant_inner_loops.tmpl +21 -0
  514. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quantize_q8.wgsl +173 -0
  515. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_mul.wgsl +152 -0
  516. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{rope.tmpl.wgsl → rope.wgsl} +71 -142
  517. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/row_norm.wgsl +153 -0
  518. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/scale.wgsl +6 -4
  519. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set.wgsl +109 -0
  520. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +2 -3
  521. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows_quant.wgsl +224 -0
  522. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{soft_max.tmpl.wgsl → soft_max.wgsl} +106 -206
  523. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/solve_tri.wgsl +121 -0
  524. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_conv.wgsl +65 -0
  525. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_scan.wgsl +193 -0
  526. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +68 -48
  527. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/upscale.wgsl +240 -0
  528. data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +18 -14
  529. data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +1 -1
  530. data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +244 -10
  531. data/ext/sources/ggml/src/ggml.c +110 -28
  532. data/ext/sources/ggml/src/gguf.cpp +173 -28
  533. data/ext/sources/include/parakeet.h +342 -0
  534. data/ext/sources/include/whisper.h +10 -0
  535. data/ext/sources/media/matmul.png +0 -0
  536. data/ext/sources/src/CMakeLists.txt +23 -0
  537. data/ext/sources/src/parakeet-arch.h +188 -0
  538. data/ext/sources/src/parakeet.cpp +3838 -0
  539. data/ext/sources/src/whisper.cpp +56 -12
  540. data/extsources.rb +26 -10
  541. data/lib/whisper/log_settable.rb +36 -0
  542. data/lib/whisper/model/uri.rb +13 -1
  543. data/lib/whisper/output.rb +74 -0
  544. data/sig/whisper.rbs +411 -62
  545. data/test/helper.rb +2 -0
  546. data/test/jfk_reader/jfk_reader.c +50 -7
  547. data/test/test_callback.rb +1 -0
  548. data/test/test_package.rb +6 -5
  549. data/test/test_parakeet.rb +28 -0
  550. data/test/test_parakeet_callback.rb +107 -0
  551. data/test/test_parakeet_context.rb +116 -0
  552. data/test/test_parakeet_context_params.rb +24 -0
  553. data/test/test_parakeet_model.rb +21 -0
  554. data/test/test_parakeet_params.rb +78 -0
  555. data/test/test_parakeet_segment.rb +42 -0
  556. data/test/test_parakeet_token.rb +73 -0
  557. data/test/test_params.rb +2 -0
  558. data/test/test_vad_segment.rb +1 -1
  559. data/test/test_whisper.rb +24 -6
  560. data/whispercpp.gemspec +2 -2
  561. metadata +215 -281
  562. data/ext/sources/bindings/javascript/CMakeLists.txt +0 -41
  563. data/ext/sources/bindings/javascript/emscripten.cpp +0 -93
  564. data/ext/sources/bindings/javascript/libwhisper.worker.js +0 -1
  565. data/ext/sources/bindings/javascript/package.json +0 -26
  566. data/ext/sources/bindings/javascript/whisper.js +0 -19
  567. data/ext/sources/examples/addon.node/CMakeLists.txt +0 -31
  568. data/ext/sources/examples/addon.node/__test__/whisper.spec.js +0 -133
  569. data/ext/sources/examples/addon.node/addon.cpp +0 -557
  570. data/ext/sources/examples/addon.node/index.js +0 -59
  571. data/ext/sources/examples/addon.node/package.json +0 -16
  572. data/ext/sources/examples/addon.node/vad-example.js +0 -132
  573. data/ext/sources/examples/bench.wasm/CMakeLists.txt +0 -49
  574. data/ext/sources/examples/bench.wasm/emscripten.cpp +0 -87
  575. data/ext/sources/examples/bench.wasm/index-tmpl.html +0 -285
  576. data/ext/sources/examples/coi-serviceworker.js +0 -146
  577. data/ext/sources/examples/command/CMakeLists.txt +0 -10
  578. data/ext/sources/examples/command/command.cpp +0 -802
  579. data/ext/sources/examples/command/commands.txt +0 -9
  580. data/ext/sources/examples/command.wasm/CMakeLists.txt +0 -50
  581. data/ext/sources/examples/command.wasm/emscripten.cpp +0 -327
  582. data/ext/sources/examples/command.wasm/index-tmpl.html +0 -415
  583. data/ext/sources/examples/generate-karaoke.sh +0 -57
  584. data/ext/sources/examples/helpers.js +0 -191
  585. data/ext/sources/examples/livestream.sh +0 -112
  586. data/ext/sources/examples/lsp/CMakeLists.txt +0 -10
  587. data/ext/sources/examples/lsp/lsp.cpp +0 -471
  588. data/ext/sources/examples/lsp/whisper.vim +0 -362
  589. data/ext/sources/examples/python/test_whisper_processor.py +0 -7
  590. data/ext/sources/examples/python/whisper_processor.py +0 -54
  591. data/ext/sources/examples/server/bench.js +0 -29
  592. data/ext/sources/examples/server.py +0 -120
  593. data/ext/sources/examples/stream/CMakeLists.txt +0 -10
  594. data/ext/sources/examples/stream/stream.cpp +0 -437
  595. data/ext/sources/examples/stream.wasm/CMakeLists.txt +0 -49
  596. data/ext/sources/examples/stream.wasm/emscripten.cpp +0 -216
  597. data/ext/sources/examples/stream.wasm/index-tmpl.html +0 -491
  598. data/ext/sources/examples/sycl/CMakeLists.txt +0 -9
  599. data/ext/sources/examples/sycl/build.sh +0 -22
  600. data/ext/sources/examples/sycl/ls-sycl-device.cpp +0 -11
  601. data/ext/sources/examples/sycl/run-whisper.sh +0 -17
  602. data/ext/sources/examples/talk-llama/CMakeLists.txt +0 -48
  603. data/ext/sources/examples/talk-llama/eleven-labs.py +0 -80
  604. data/ext/sources/examples/talk-llama/llama-adapter.cpp +0 -488
  605. data/ext/sources/examples/talk-llama/llama-adapter.h +0 -89
  606. data/ext/sources/examples/talk-llama/llama-arch.cpp +0 -2877
  607. data/ext/sources/examples/talk-llama/llama-arch.h +0 -628
  608. data/ext/sources/examples/talk-llama/llama-batch.cpp +0 -919
  609. data/ext/sources/examples/talk-llama/llama-batch.h +0 -173
  610. data/ext/sources/examples/talk-llama/llama-chat.cpp +0 -896
  611. data/ext/sources/examples/talk-llama/llama-chat.h +0 -71
  612. data/ext/sources/examples/talk-llama/llama-context.cpp +0 -3633
  613. data/ext/sources/examples/talk-llama/llama-context.h +0 -359
  614. data/ext/sources/examples/talk-llama/llama-cparams.cpp +0 -5
  615. data/ext/sources/examples/talk-llama/llama-cparams.h +0 -47
  616. data/ext/sources/examples/talk-llama/llama-ext.h +0 -12
  617. data/ext/sources/examples/talk-llama/llama-grammar.cpp +0 -1464
  618. data/ext/sources/examples/talk-llama/llama-grammar.h +0 -194
  619. data/ext/sources/examples/talk-llama/llama-graph.cpp +0 -2735
  620. data/ext/sources/examples/talk-llama/llama-graph.h +0 -1031
  621. data/ext/sources/examples/talk-llama/llama-hparams.cpp +0 -258
  622. data/ext/sources/examples/talk-llama/llama-hparams.h +0 -353
  623. data/ext/sources/examples/talk-llama/llama-impl.cpp +0 -171
  624. data/ext/sources/examples/talk-llama/llama-impl.h +0 -75
  625. data/ext/sources/examples/talk-llama/llama-io.cpp +0 -15
  626. data/ext/sources/examples/talk-llama/llama-io.h +0 -35
  627. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +0 -330
  628. data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.h +0 -137
  629. data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +0 -2285
  630. data/ext/sources/examples/talk-llama/llama-kv-cache.h +0 -389
  631. data/ext/sources/examples/talk-llama/llama-kv-cells.h +0 -533
  632. data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.cpp +0 -275
  633. data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.h +0 -140
  634. data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +0 -268
  635. data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +0 -139
  636. data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +0 -1165
  637. data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +0 -182
  638. data/ext/sources/examples/talk-llama/llama-memory.cpp +0 -59
  639. data/ext/sources/examples/talk-llama/llama-memory.h +0 -122
  640. data/ext/sources/examples/talk-llama/llama-mmap.cpp +0 -752
  641. data/ext/sources/examples/talk-llama/llama-mmap.h +0 -73
  642. data/ext/sources/examples/talk-llama/llama-model-loader.cpp +0 -1655
  643. data/ext/sources/examples/talk-llama/llama-model-loader.h +0 -206
  644. data/ext/sources/examples/talk-llama/llama-model-saver.cpp +0 -299
  645. data/ext/sources/examples/talk-llama/llama-model-saver.h +0 -40
  646. data/ext/sources/examples/talk-llama/llama-model.cpp +0 -9056
  647. data/ext/sources/examples/talk-llama/llama-model.h +0 -597
  648. data/ext/sources/examples/talk-llama/llama-quant.cpp +0 -1304
  649. data/ext/sources/examples/talk-llama/llama-quant.h +0 -1
  650. data/ext/sources/examples/talk-llama/llama-sampler.cpp +0 -3885
  651. data/ext/sources/examples/talk-llama/llama-sampler.h +0 -42
  652. data/ext/sources/examples/talk-llama/llama-vocab.cpp +0 -3970
  653. data/ext/sources/examples/talk-llama/llama-vocab.h +0 -187
  654. data/ext/sources/examples/talk-llama/llama.cpp +0 -1194
  655. data/ext/sources/examples/talk-llama/llama.h +0 -1573
  656. data/ext/sources/examples/talk-llama/models/afmoe.cpp +0 -190
  657. data/ext/sources/examples/talk-llama/models/apertus.cpp +0 -125
  658. data/ext/sources/examples/talk-llama/models/arcee.cpp +0 -135
  659. data/ext/sources/examples/talk-llama/models/arctic.cpp +0 -137
  660. data/ext/sources/examples/talk-llama/models/arwkv7.cpp +0 -86
  661. data/ext/sources/examples/talk-llama/models/baichuan.cpp +0 -123
  662. data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +0 -143
  663. data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +0 -133
  664. data/ext/sources/examples/talk-llama/models/bert.cpp +0 -184
  665. data/ext/sources/examples/talk-llama/models/bitnet.cpp +0 -145
  666. data/ext/sources/examples/talk-llama/models/bloom.cpp +0 -101
  667. data/ext/sources/examples/talk-llama/models/chameleon.cpp +0 -178
  668. data/ext/sources/examples/talk-llama/models/chatglm.cpp +0 -132
  669. data/ext/sources/examples/talk-llama/models/codeshell.cpp +0 -111
  670. data/ext/sources/examples/talk-llama/models/cogvlm.cpp +0 -102
  671. data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +0 -134
  672. data/ext/sources/examples/talk-llama/models/command-r.cpp +0 -122
  673. data/ext/sources/examples/talk-llama/models/dbrx.cpp +0 -122
  674. data/ext/sources/examples/talk-llama/models/deci.cpp +0 -135
  675. data/ext/sources/examples/talk-llama/models/deepseek.cpp +0 -142
  676. data/ext/sources/examples/talk-llama/models/deepseek2.cpp +0 -262
  677. data/ext/sources/examples/talk-llama/models/delta-net-base.cpp +0 -445
  678. data/ext/sources/examples/talk-llama/models/dots1.cpp +0 -132
  679. data/ext/sources/examples/talk-llama/models/dream.cpp +0 -105
  680. data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +0 -148
  681. data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +0 -110
  682. data/ext/sources/examples/talk-llama/models/eurobert.cpp +0 -97
  683. data/ext/sources/examples/talk-llama/models/exaone-moe.cpp +0 -145
  684. data/ext/sources/examples/talk-llama/models/exaone.cpp +0 -114
  685. data/ext/sources/examples/talk-llama/models/exaone4.cpp +0 -123
  686. data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +0 -111
  687. data/ext/sources/examples/talk-llama/models/falcon.cpp +0 -120
  688. data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +0 -116
  689. data/ext/sources/examples/talk-llama/models/gemma.cpp +0 -112
  690. data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +0 -128
  691. data/ext/sources/examples/talk-llama/models/gemma3.cpp +0 -155
  692. data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +0 -384
  693. data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +0 -170
  694. data/ext/sources/examples/talk-llama/models/glm4.cpp +0 -157
  695. data/ext/sources/examples/talk-llama/models/gpt2.cpp +0 -105
  696. data/ext/sources/examples/talk-llama/models/gptneox.cpp +0 -144
  697. data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +0 -195
  698. data/ext/sources/examples/talk-llama/models/granite.cpp +0 -210
  699. data/ext/sources/examples/talk-llama/models/grok.cpp +0 -159
  700. data/ext/sources/examples/talk-llama/models/grovemoe.cpp +0 -139
  701. data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +0 -132
  702. data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +0 -153
  703. data/ext/sources/examples/talk-llama/models/internlm2.cpp +0 -120
  704. data/ext/sources/examples/talk-llama/models/jais.cpp +0 -86
  705. data/ext/sources/examples/talk-llama/models/jais2.cpp +0 -123
  706. data/ext/sources/examples/talk-llama/models/jamba.cpp +0 -106
  707. data/ext/sources/examples/talk-llama/models/kimi-linear.cpp +0 -381
  708. data/ext/sources/examples/talk-llama/models/lfm2.cpp +0 -196
  709. data/ext/sources/examples/talk-llama/models/llada-moe.cpp +0 -122
  710. data/ext/sources/examples/talk-llama/models/llada.cpp +0 -99
  711. data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +0 -178
  712. data/ext/sources/examples/talk-llama/models/llama.cpp +0 -175
  713. data/ext/sources/examples/talk-llama/models/maincoder.cpp +0 -117
  714. data/ext/sources/examples/talk-llama/models/mamba-base.cpp +0 -289
  715. data/ext/sources/examples/talk-llama/models/mamba.cpp +0 -54
  716. data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +0 -129
  717. data/ext/sources/examples/talk-llama/models/minicpm3.cpp +0 -200
  718. data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +0 -123
  719. data/ext/sources/examples/talk-llama/models/mistral3.cpp +0 -160
  720. data/ext/sources/examples/talk-llama/models/models.h +0 -704
  721. data/ext/sources/examples/talk-llama/models/modern-bert.cpp +0 -109
  722. data/ext/sources/examples/talk-llama/models/mpt.cpp +0 -126
  723. data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +0 -162
  724. data/ext/sources/examples/talk-llama/models/nemotron.cpp +0 -122
  725. data/ext/sources/examples/talk-llama/models/neo-bert.cpp +0 -104
  726. data/ext/sources/examples/talk-llama/models/olmo.cpp +0 -121
  727. data/ext/sources/examples/talk-llama/models/olmo2.cpp +0 -150
  728. data/ext/sources/examples/talk-llama/models/olmoe.cpp +0 -124
  729. data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +0 -127
  730. data/ext/sources/examples/talk-llama/models/openelm.cpp +0 -124
  731. data/ext/sources/examples/talk-llama/models/orion.cpp +0 -123
  732. data/ext/sources/examples/talk-llama/models/paddleocr.cpp +0 -122
  733. data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +0 -121
  734. data/ext/sources/examples/talk-llama/models/phi2.cpp +0 -121
  735. data/ext/sources/examples/talk-llama/models/phi3.cpp +0 -152
  736. data/ext/sources/examples/talk-llama/models/plamo.cpp +0 -110
  737. data/ext/sources/examples/talk-llama/models/plamo2.cpp +0 -320
  738. data/ext/sources/examples/talk-llama/models/plamo3.cpp +0 -128
  739. data/ext/sources/examples/talk-llama/models/plm.cpp +0 -169
  740. data/ext/sources/examples/talk-llama/models/qwen.cpp +0 -108
  741. data/ext/sources/examples/talk-llama/models/qwen2.cpp +0 -126
  742. data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +0 -151
  743. data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +0 -117
  744. data/ext/sources/examples/talk-llama/models/qwen3.cpp +0 -120
  745. data/ext/sources/examples/talk-llama/models/qwen35.cpp +0 -381
  746. data/ext/sources/examples/talk-llama/models/qwen35moe.cpp +0 -422
  747. data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +0 -131
  748. data/ext/sources/examples/talk-llama/models/qwen3next.cpp +0 -525
  749. data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +0 -140
  750. data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +0 -132
  751. data/ext/sources/examples/talk-llama/models/refact.cpp +0 -94
  752. data/ext/sources/examples/talk-llama/models/rnd1.cpp +0 -126
  753. data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +0 -164
  754. data/ext/sources/examples/talk-llama/models/rwkv6.cpp +0 -94
  755. data/ext/sources/examples/talk-llama/models/rwkv6qwen2.cpp +0 -86
  756. data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +0 -137
  757. data/ext/sources/examples/talk-llama/models/rwkv7.cpp +0 -90
  758. data/ext/sources/examples/talk-llama/models/seed-oss.cpp +0 -124
  759. data/ext/sources/examples/talk-llama/models/smallthinker.cpp +0 -126
  760. data/ext/sources/examples/talk-llama/models/smollm3.cpp +0 -128
  761. data/ext/sources/examples/talk-llama/models/stablelm.cpp +0 -146
  762. data/ext/sources/examples/talk-llama/models/starcoder.cpp +0 -100
  763. data/ext/sources/examples/talk-llama/models/starcoder2.cpp +0 -121
  764. data/ext/sources/examples/talk-llama/models/step35-iswa.cpp +0 -165
  765. data/ext/sources/examples/talk-llama/models/t5-dec.cpp +0 -166
  766. data/ext/sources/examples/talk-llama/models/t5-enc.cpp +0 -96
  767. data/ext/sources/examples/talk-llama/models/wavtokenizer-dec.cpp +0 -149
  768. data/ext/sources/examples/talk-llama/models/xverse.cpp +0 -108
  769. data/ext/sources/examples/talk-llama/prompts/talk-alpaca.txt +0 -23
  770. data/ext/sources/examples/talk-llama/speak +0 -40
  771. data/ext/sources/examples/talk-llama/speak.bat +0 -1
  772. data/ext/sources/examples/talk-llama/speak.ps1 +0 -14
  773. data/ext/sources/examples/talk-llama/talk-llama.cpp +0 -813
  774. data/ext/sources/examples/talk-llama/unicode-data.cpp +0 -7034
  775. data/ext/sources/examples/talk-llama/unicode-data.h +0 -20
  776. data/ext/sources/examples/talk-llama/unicode.cpp +0 -1103
  777. data/ext/sources/examples/talk-llama/unicode.h +0 -111
  778. data/ext/sources/examples/wchess/CMakeLists.txt +0 -10
  779. data/ext/sources/examples/wchess/libwchess/CMakeLists.txt +0 -19
  780. data/ext/sources/examples/wchess/libwchess/Chessboard.cpp +0 -803
  781. data/ext/sources/examples/wchess/libwchess/Chessboard.h +0 -33
  782. data/ext/sources/examples/wchess/libwchess/WChess.cpp +0 -193
  783. data/ext/sources/examples/wchess/libwchess/WChess.h +0 -63
  784. data/ext/sources/examples/wchess/libwchess/test-chessboard.cpp +0 -117
  785. data/ext/sources/examples/wchess/wchess.cmd/CMakeLists.txt +0 -8
  786. data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +0 -253
  787. data/ext/sources/examples/whisper.wasm/CMakeLists.txt +0 -50
  788. data/ext/sources/examples/whisper.wasm/emscripten.cpp +0 -118
  789. data/ext/sources/examples/whisper.wasm/index-tmpl.html +0 -659
  790. data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +0 -99
  791. data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +0 -155
  792. data/ext/sources/ggml/src/ggml-hexagon/op-desc.h +0 -153
  793. data/ext/sources/ggml/src/ggml-opencl/kernels/embed_kernel.py +0 -26
  794. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.cpp +0 -123
  795. data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.h +0 -17
  796. data/ext/sources/ggml/src/ggml-virtgpu/regenerate_remoting.py +0 -333
  797. data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.glsl +0 -5
  798. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +0 -182
  799. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +0 -323
  800. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.wgsl +0 -718
  801. data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +0 -123
  802. data/ext/sources/tests/CMakeLists.txt +0 -112
  803. data/ext/sources/tests/earnings21/eval.mk +0 -58
  804. data/ext/sources/tests/earnings21/eval.py +0 -68
  805. data/ext/sources/tests/earnings21/normalizers/__init__.py +0 -2
  806. data/ext/sources/tests/earnings21/normalizers/basic.py +0 -80
  807. data/ext/sources/tests/earnings21/normalizers/english.json +0 -1741
  808. data/ext/sources/tests/earnings21/normalizers/english.py +0 -550
  809. data/ext/sources/tests/earnings21/requirements.txt +0 -6
  810. data/ext/sources/tests/en-0-ref.txt +0 -1
  811. data/ext/sources/tests/en-1-ref.txt +0 -1
  812. data/ext/sources/tests/en-2-ref.txt +0 -1
  813. data/ext/sources/tests/es-0-ref.txt +0 -1
  814. data/ext/sources/tests/librispeech/eval.mk +0 -39
  815. data/ext/sources/tests/librispeech/eval.py +0 -47
  816. data/ext/sources/tests/librispeech/normalizers/__init__.py +0 -2
  817. data/ext/sources/tests/librispeech/normalizers/basic.py +0 -80
  818. data/ext/sources/tests/librispeech/normalizers/english.json +0 -1741
  819. data/ext/sources/tests/librispeech/normalizers/english.py +0 -550
  820. data/ext/sources/tests/librispeech/requirements.txt +0 -6
  821. data/ext/sources/tests/run-tests.sh +0 -130
  822. data/ext/sources/tests/test-c.c +0 -3
  823. data/ext/sources/tests/test-vad-full.cpp +0 -56
  824. data/ext/sources/tests/test-vad.cpp +0 -83
  825. data/ext/sources/tests/test-whisper.js +0 -58
  826. data/lib/whisper/context.rb +0 -15
  827. data/lib/whisper/segment.rb +0 -58
  828. /data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle_general_q8_0_f32.cl → gemv_noshuffle_q8_0_f32.cl} +0 -0
data/ext/ruby_whisper.h CHANGED
@@ -2,8 +2,20 @@
2
2
  #define RUBY_WHISPER_H
3
3
 
4
4
  #include <ruby.h>
5
+ #include <ruby/version.h>
6
+ #include <ruby/util.h>
7
+ #include <ruby/thread.h>
8
+ #include <ruby/thread_native.h>
9
+ #include <ruby/atomic.h>
5
10
  #include <ruby/memory_view.h>
6
11
  #include "whisper.h"
12
+ #include "parakeet.h"
13
+ #include "ruby_whisper_log_settable.h"
14
+
15
+ #if RUBY_API_VERSION_MAJOR < 4
16
+ // Exists but not declared as public API
17
+ int ruby_thread_has_gvl_p(void);
18
+ #endif
7
19
 
8
20
  typedef struct {
9
21
  VALUE *context;
@@ -12,6 +24,29 @@ typedef struct {
12
24
  VALUE callbacks;
13
25
  } ruby_whisper_callback_container;
14
26
 
27
+ typedef struct ruby_whisper_abort_callback_user_data {
28
+ volatile rb_atomic_t is_interrupted;
29
+ ruby_whisper_callback_container *callback_container;
30
+ } ruby_whisper_abort_callback_user_data;
31
+
32
+ typedef struct ruby_whisper_log {
33
+ enum ggml_log_level level;
34
+ char *text;
35
+ size_t length;
36
+ size_t capacity;
37
+ } ruby_whisper_log;
38
+
39
+ typedef struct ruby_whisper_log_queue {
40
+ rb_nativethread_lock_t lock;
41
+ rb_nativethread_cond_t cond;
42
+ bool is_open;
43
+
44
+ size_t head;
45
+ size_t tail;
46
+ size_t size;
47
+ ruby_whisper_log *logs;
48
+ } ruby_whisper_log_queue;
49
+
15
50
  typedef struct {
16
51
  struct whisper_context *context;
17
52
  } ruby_whisper;
@@ -68,6 +103,63 @@ typedef struct parsed_samples_t {
68
103
  bool memview_exported;
69
104
  } parsed_samples_t;
70
105
 
106
+ typedef struct {
107
+ VALUE *context;
108
+ VALUE *params;
109
+ float *samples;
110
+ int n_samples;
111
+ } ruby_whisper_full_args;
112
+
113
+ typedef struct ruby_whisper_full_parallel_args {
114
+ VALUE *context;
115
+ VALUE *params;
116
+ float *samples;
117
+ int n_samples;
118
+ int n_processors;
119
+ } ruby_whisper_full_parallel_args;
120
+
121
+ typedef struct {
122
+ struct parakeet_full_params params;
123
+ ruby_whisper_callback_container *new_segment_callback_container;
124
+ ruby_whisper_callback_container *new_token_callback_container;
125
+ ruby_whisper_callback_container *progress_callback_container;
126
+ ruby_whisper_callback_container *encoder_begin_callback_container;
127
+ ruby_whisper_callback_container *abort_callback_container;
128
+ } ruby_whisper_parakeet_params;
129
+
130
+ typedef struct {
131
+ struct parakeet_context_params params;
132
+ } ruby_whisper_parakeet_context_params;
133
+
134
+ typedef struct {
135
+ struct parakeet_context *context;
136
+ } ruby_whisper_parakeet_context;
137
+
138
+ typedef struct {
139
+ VALUE context;
140
+ int index;
141
+ } ruby_whisper_parakeet_segment;
142
+
143
+ typedef struct {
144
+ parakeet_token_data *token_data;
145
+ VALUE text;
146
+ } ruby_whisper_parakeet_token;
147
+
148
+ typedef struct {
149
+ VALUE context;
150
+ } ruby_whisper_parakeet_model;
151
+
152
+ extern ID id_extended;
153
+ extern ID id_log_callback_thread;
154
+ extern ID id_start_log_callback_thread;
155
+ extern ID id_alive_p;
156
+ extern ID id_join;
157
+ extern void ruby_whisper_log_queue_initialize(ruby_whisper_log_queue *log_queue);
158
+ extern void ruby_whisper_log_queue_open(ruby_whisper_log_queue *log_queue);
159
+ extern void ruby_whisper_log_queue_close(ruby_whisper_log_queue *log_queue);
160
+ extern void ruby_whisper_log_queue_enqueue(ruby_whisper_log_queue *log_queue, enum ggml_log_level level, const char *text);
161
+ extern VALUE ruby_whisper_log_queue_drain(ruby_whisper_log_queue *log_queue);
162
+
71
163
  #define GetContext(obj, rw) do { \
72
164
  TypedData_Get_Struct((obj), ruby_whisper, &ruby_whisper_type, (rw)); \
73
165
  if ((rw)->context == NULL) { \
@@ -104,4 +196,47 @@ typedef struct parsed_samples_t {
104
196
  } \
105
197
  } while (0)
106
198
 
199
+ #define GetParakeetContextParams(obj, rwpcp) do { \
200
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_context_params, &ruby_whisper_parakeet_context_params_type, (rwpcp)); \
201
+ } while (0)
202
+
203
+ #define GetParakeetContext(obj, rwpc) do { \
204
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_context, &ruby_whisper_parakeet_context_type, (rwpc)); \
205
+ if ((rwpc)->context == NULL) { \
206
+ rb_raise(rb_eRuntimeError, "Not initialized"); \
207
+ } \
208
+ } while (0)
209
+
210
+ #define GetParakeetParams(obj, rwpp) do { \
211
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_params, &ruby_whisper_parakeet_params_type, (rwpp)); \
212
+ if (!(rwpp)->new_segment_callback_container || \
213
+ !(rwpp)->new_token_callback_container || \
214
+ !(rwpp)->progress_callback_container || \
215
+ !(rwpp)->encoder_begin_callback_container || \
216
+ !(rwpp)->abort_callback_container) { \
217
+ rb_raise(rb_eRuntimeError, "Not initialized"); \
218
+ } \
219
+ } while (0)
220
+
221
+ #define GetParakeetSegment(obj, rwps) do { \
222
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_segment, &ruby_whisper_parakeet_segment_type, (rwps)); \
223
+ if (!(rwps)->context) { \
224
+ rb_raise(rb_eRuntimeError, "Not initialized"); \
225
+ } \
226
+ } while (0)
227
+
228
+ #define GetParakeetToken(obj, rwpt) do { \
229
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_token, &ruby_whisper_parakeet_token_type, (rwpt)); \
230
+ if (!(rwpt)->token_data) { \
231
+ rb_raise(rb_eRuntimeError, "Not initialized"); \
232
+ } \
233
+ } while (0)
234
+
235
+ #define GetParakeetModel(obj, rwpm) do { \
236
+ TypedData_Get_Struct((obj), ruby_whisper_parakeet_model, &ruby_whisper_parakeet_model_type, (rwpm)); \
237
+ if (NIL_P((rwpm)->context)) { \
238
+ rb_raise(rb_eRuntimeError, "Not initialized"); \
239
+ } \
240
+ } while (0)
241
+
107
242
  #endif
@@ -1,5 +1,11 @@
1
1
  #include "ruby_whisper.h"
2
2
 
3
+ #ifdef WORDS_BIGENDIAN
4
+ #define IS_BIGENDIAN true
5
+ #else
6
+ #define IS_BIGENDIAN false
7
+ #endif
8
+
3
9
  extern ID id_to_s;
4
10
  extern ID id___method__;
5
11
  extern ID id_to_enum;
@@ -22,7 +28,7 @@ extern const rb_data_type_t ruby_whisper_context_params_type;
22
28
  extern VALUE ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self);
23
29
  extern VALUE rb_whisper_model_s_new(VALUE context);
24
30
  extern VALUE rb_whisper_segment_s_new(VALUE context, int index);
25
- extern void prepare_transcription(ruby_whisper_params *rwp, VALUE *context);
31
+ extern void prepare_transcription(ruby_whisper_params *rwp, VALUE *context, int n_processors, ruby_whisper_abort_callback_user_data *abort_callback_user_data);
26
32
 
27
33
  ID transcribe_option_names[1];
28
34
 
@@ -32,20 +38,26 @@ typedef struct fill_samples_args {
32
38
  int n_samples;
33
39
  } fill_samples_args;
34
40
 
35
- typedef struct full_args {
36
- VALUE *context;
37
- VALUE *params;
41
+ typedef struct full_without_gvl_args {
42
+ struct whisper_context *context;
43
+ struct whisper_full_params *params;
38
44
  float *samples;
39
45
  int n_samples;
40
- } full_args;
46
+ int result;
47
+ } full_without_gvl_args;
41
48
 
42
- typedef struct full_parallel_args {
43
- VALUE *context;
44
- VALUE *params;
49
+ typedef struct full_parallel_without_gvl_args {
50
+ struct whisper_context *context;
51
+ struct whisper_full_params *params;
45
52
  float *samples;
46
53
  int n_samples;
47
54
  int n_processors;
48
- } full_parallel_args;
55
+ int result;
56
+ } full_parallel_without_gvl_args;
57
+
58
+ typedef struct full_ubf_args {
59
+ ruby_whisper_abort_callback_user_data *abort_callback_user_data;
60
+ } full_ubf_args;
49
61
 
50
62
  static void
51
63
  ruby_whisper_free(ruby_whisper *rw)
@@ -74,7 +86,7 @@ static size_t
74
86
  ruby_whisper_memsize(const void *p)
75
87
  {
76
88
  const ruby_whisper *rw = (const ruby_whisper *)p;
77
- size_t size = sizeof(rw);
89
+ size_t size = sizeof(*rw);
78
90
  if (!rw) {
79
91
  return 0;
80
92
  }
@@ -304,11 +316,25 @@ VALUE ruby_whisper_model_type(VALUE self)
304
316
  static bool
305
317
  check_memory_view(rb_memory_view_t *memview)
306
318
  {
307
- if (memview->format != NULL && strcmp(memview->format, "f") != 0) {
308
- rb_warn("currently only format \"f\" is supported for MemoryView, but given: %s", memview->format);
319
+ if (!memview->format) {
320
+ rb_warn("currently format is required");
321
+ return false;
322
+ }
323
+
324
+ if (strcmp(memview->format, "f") == 0) {
325
+ // accept
326
+ } else if (strcmp(memview->format, "e") == 0) {
327
+ if (IS_BIGENDIAN) {
328
+ rb_warn("currently format \"e\" is only supported on little-endian environment");
329
+ return false;
330
+ }
331
+ } else {
332
+ rb_warn("currently only format \"f\" and \"e\" on little-endian environment is supported for MemoryView, but given: %s", memview->format);
309
333
  return false;
310
334
  }
311
- if (memview->format != NULL && memview->ndim != 1) {
335
+
336
+ if (memview->ndim != 1 && !(memview->ndim == 2 && memview->shape[1] == 1)) {
337
+ // TODO: Accept ndim == 2 with shape [n_samples, channels] and channels > 1 by averaging the samples in different channels or just taking the first channel
312
338
  rb_warn("currently only 1 dimensional MemoryView is supported, but given: %zd", memview->ndim);
313
339
  return false;
314
340
  }
@@ -338,7 +364,7 @@ fill_samples(VALUE rb_args)
338
364
  return Qnil;
339
365
  }
340
366
 
341
- struct parsed_samples_t
367
+ parsed_samples_t
342
368
  parse_samples(VALUE *samples, VALUE *n_samples)
343
369
  {
344
370
  bool memview_available = rb_memory_view_available_p(*samples);
@@ -426,20 +452,50 @@ release_samples(VALUE rb_parsed_args)
426
452
  return Qnil;
427
453
  }
428
454
 
429
- static VALUE
455
+ static void*
456
+ full_without_gvl(void *rb_args)
457
+ {
458
+ full_without_gvl_args *args = (full_without_gvl_args *)rb_args;
459
+ args->result = whisper_full(args->context, *args->params, args->samples, args->n_samples);
460
+ return NULL;
461
+ }
462
+
463
+ static void
464
+ full_ubf(void *rb_args)
465
+ {
466
+ full_ubf_args *args = (full_ubf_args *)rb_args;
467
+
468
+ RUBY_ATOMIC_SET(args->abort_callback_user_data->is_interrupted, 1);
469
+ }
470
+
471
+ VALUE
430
472
  full_body(VALUE rb_args)
431
473
  {
432
- full_args *args = (full_args *)rb_args;
474
+ ruby_whisper_full_args *args = (ruby_whisper_full_args *)rb_args;
433
475
 
434
476
  ruby_whisper *rw;
435
477
  ruby_whisper_params *rwp;
436
478
  GetContext(*args->context, rw);
437
479
  TypedData_Get_Struct(*args->params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
438
480
 
439
- prepare_transcription(rwp, args->context);
440
- int result = whisper_full(rw->context, rwp->params, args->samples, args->n_samples);
441
-
442
- return INT2NUM(result);
481
+ ruby_whisper_abort_callback_user_data abort_callback_user_data = {
482
+ 0,
483
+ NULL,
484
+ };
485
+ prepare_transcription(rwp, args->context, 1, &abort_callback_user_data);
486
+
487
+ struct full_without_gvl_args full_without_gvl_args = {
488
+ rw->context,
489
+ &rwp->params,
490
+ args->samples,
491
+ args->n_samples,
492
+ 0,
493
+ };
494
+ full_ubf_args full_ubf_args = {
495
+ &abort_callback_user_data,
496
+ };
497
+ rb_thread_call_without_gvl(full_without_gvl, (void *)&full_without_gvl_args, full_ubf, (void *)&full_ubf_args);
498
+ return INT2NUM(full_without_gvl_args.result);
443
499
  }
444
500
 
445
501
  /*
@@ -462,7 +518,7 @@ VALUE ruby_whisper_full(int argc, VALUE *argv, VALUE self)
462
518
  VALUE n_samples = argc == 2 ? Qnil : argv[2];
463
519
 
464
520
  struct parsed_samples_t parsed = parse_samples(&argv[1], &n_samples);
465
- full_args args = {
521
+ ruby_whisper_full_args args = {
466
522
  &self,
467
523
  &argv[0],
468
524
  parsed.samples,
@@ -477,20 +533,43 @@ VALUE ruby_whisper_full(int argc, VALUE *argv, VALUE self)
477
533
  }
478
534
  }
479
535
 
480
- static VALUE
536
+ static void*
537
+ full_parallel_without_gvl(void *rb_args)
538
+ {
539
+ full_parallel_without_gvl_args *args = (full_parallel_without_gvl_args *)rb_args;
540
+ args->result = whisper_full_parallel(args->context, *args->params, args->samples, args->n_samples, args->n_processors);
541
+ return NULL;
542
+ }
543
+
544
+ VALUE
481
545
  full_parallel_body(VALUE rb_args)
482
546
  {
483
- full_parallel_args *args = (full_parallel_args *)rb_args;
547
+ ruby_whisper_full_parallel_args *args = (ruby_whisper_full_parallel_args *)rb_args;
484
548
 
485
549
  ruby_whisper *rw;
486
550
  ruby_whisper_params *rwp;
487
551
  GetContext(*args->context, rw);
488
552
  TypedData_Get_Struct(*args->params, ruby_whisper_params, &ruby_whisper_params_type, rwp);
489
553
 
490
- prepare_transcription(rwp, args->context);
491
- int result = whisper_full_parallel(rw->context, rwp->params, args->samples, args->n_samples, args->n_processors);
492
-
493
- return INT2NUM(result);
554
+ ruby_whisper_abort_callback_user_data abort_callback_user_data = {
555
+ 0,
556
+ NULL,
557
+ };
558
+ prepare_transcription(rwp, args->context, args->n_processors, &abort_callback_user_data);
559
+
560
+ struct full_parallel_without_gvl_args full_parallel_without_gvl_args = {
561
+ rw->context,
562
+ &rwp->params,
563
+ args->samples,
564
+ args->n_samples,
565
+ args->n_processors,
566
+ 0,
567
+ };
568
+ full_ubf_args full_ubf_args = {
569
+ &abort_callback_user_data,
570
+ };
571
+ rb_thread_call_without_gvl(full_parallel_without_gvl, (void *)&full_parallel_without_gvl_args, full_ubf, (void *)&full_ubf_args);
572
+ return INT2NUM(full_parallel_without_gvl_args.result);
494
573
  }
495
574
 
496
575
  /*
@@ -527,7 +606,7 @@ ruby_whisper_full_parallel(int argc, VALUE *argv,VALUE self)
527
606
  break;
528
607
  }
529
608
  struct parsed_samples_t parsed = parse_samples(&argv[1], &n_samples);
530
- const full_parallel_args args = {
609
+ const ruby_whisper_full_parallel_args args = {
531
610
  &self,
532
611
  &argv[0],
533
612
  parsed.samples,
@@ -0,0 +1,180 @@
1
+ #include "ruby_whisper.h"
2
+
3
+ #define LOG_QUEUE_CAPACITY 256
4
+ #define LOG_DEFAULT_CAPACITY 1024
5
+
6
+ void
7
+ ruby_whisper_log_queue_initialize(ruby_whisper_log_queue *log_queue)
8
+ {
9
+ rb_nativethread_lock_initialize(&log_queue->lock);
10
+ rb_native_cond_initialize(&log_queue->cond);
11
+ log_queue->head = 0;
12
+ log_queue->tail = 0;
13
+ log_queue->size = 0;
14
+ log_queue->is_open = true;
15
+ log_queue->logs = ALLOC_N(ruby_whisper_log, LOG_QUEUE_CAPACITY);
16
+ for (size_t i = 0; i < LOG_QUEUE_CAPACITY; i++) {
17
+ // we cannot call Ruby API like ALLOC_N because this slot may be realloced without GVL
18
+ // this doesn't be freed because log queue lives until the end of process
19
+ char *slot = malloc(sizeof(char) * LOG_QUEUE_CAPACITY);
20
+ if (!slot) {
21
+ rb_raise(rb_eRuntimeError, "Could not allocate memory for log text");
22
+ }
23
+ ruby_whisper_log log = {
24
+ 0,
25
+ slot,
26
+ 0,
27
+ LOG_QUEUE_CAPACITY,
28
+ };
29
+ log_queue->logs[i] = log;
30
+ }
31
+ }
32
+
33
+ void
34
+ ruby_whisper_log_queue_open(ruby_whisper_log_queue *log_queue)
35
+ {
36
+ rb_nativethread_lock_lock(&log_queue->lock);
37
+
38
+ log_queue->is_open = true;
39
+
40
+ rb_native_cond_signal(&log_queue->cond);
41
+
42
+ rb_nativethread_lock_unlock(&log_queue->lock);
43
+ }
44
+
45
+ void
46
+ ruby_whisper_log_queue_close(ruby_whisper_log_queue *log_queue)
47
+ {
48
+ rb_nativethread_lock_lock(&log_queue->lock);
49
+
50
+ log_queue->is_open = false;
51
+ rb_native_cond_broadcast(&log_queue->cond);
52
+
53
+ rb_nativethread_lock_unlock(&log_queue->lock);
54
+ }
55
+
56
+ static size_t
57
+ calc_enough_cap(size_t len)
58
+ {
59
+ size_t quot = len / LOG_DEFAULT_CAPACITY;
60
+ size_t rem = len % LOG_DEFAULT_CAPACITY;
61
+
62
+ return sizeof(char) * (rem == 0 ? quot : quot + 1) * LOG_DEFAULT_CAPACITY;
63
+ }
64
+
65
+ void
66
+ ruby_whisper_log_queue_enqueue(ruby_whisper_log_queue *log_queue, enum ggml_log_level level, const char *text)
67
+ {
68
+ rb_nativethread_lock_lock(&log_queue->lock);
69
+
70
+ if (!log_queue->is_open) {
71
+ rb_nativethread_lock_unlock(&log_queue->lock);
72
+ return;
73
+ }
74
+
75
+ size_t len = strlen(text);
76
+ ruby_whisper_log *log = &log_queue->logs[log_queue->head];
77
+ if (len > log->capacity) {
78
+ size_t new_cap = calc_enough_cap(len);
79
+ // we cannot call Ruby API like REALLOC_N because this function is called without GVL
80
+ char *slot = realloc(log->text, new_cap);
81
+ if (!slot) {
82
+ rb_nativethread_lock_unlock(&log_queue->lock);
83
+ return;
84
+ }
85
+ log->text = slot;
86
+ log->capacity = new_cap;
87
+ }
88
+ // we cannot call Ruby API like MEMCPY because this function is called without GVL
89
+ memcpy(log->text, text, sizeof(char) * len);
90
+ log->length = len;
91
+ log->level = level;
92
+ log_queue->head = (log_queue->head + 1) % LOG_QUEUE_CAPACITY;
93
+ bool is_full = log_queue->size >= LOG_QUEUE_CAPACITY;
94
+ log_queue->size = is_full ? LOG_QUEUE_CAPACITY : log_queue->size + 1;
95
+ if (is_full) {
96
+ log_queue->tail = log_queue->head;
97
+ }
98
+
99
+ rb_native_cond_signal(&log_queue->cond);
100
+ rb_nativethread_lock_unlock(&log_queue->lock);
101
+ }
102
+
103
+ static void*
104
+ ruby_whisper_log_queue_wait(void *args)
105
+ {
106
+ ruby_whisper_log_queue *log_queue = (ruby_whisper_log_queue *)args;
107
+
108
+ rb_native_cond_wait(&log_queue->cond, &log_queue->lock);
109
+ rb_nativethread_lock_unlock(&log_queue->lock);
110
+
111
+ return NULL;
112
+ }
113
+
114
+ static void
115
+ ruby_whisper_log_queue_wait_ubf(void *args)
116
+ {
117
+ ruby_whisper_log_queue *log_queue = (ruby_whisper_log_queue *)args;
118
+
119
+ rb_native_cond_broadcast(&log_queue->cond);
120
+ }
121
+
122
+ typedef struct {
123
+ enum ggml_log_level level;
124
+ size_t length;
125
+ char *text;
126
+ } log_snapshot;
127
+
128
+ VALUE
129
+ ruby_whisper_log_queue_drain(ruby_whisper_log_queue *log_queue)
130
+ {
131
+ log_snapshot logs[LOG_QUEUE_CAPACITY];
132
+
133
+ rb_nativethread_lock_lock(&log_queue->lock);
134
+
135
+ while (log_queue->size == 0 && log_queue->is_open) {
136
+ rb_thread_call_without_gvl(ruby_whisper_log_queue_wait, (void *)log_queue, ruby_whisper_log_queue_wait_ubf, (void *)log_queue);
137
+ rb_nativethread_lock_lock(&log_queue->lock);
138
+ }
139
+
140
+ if (log_queue->size == 0 && !log_queue->is_open) {
141
+ rb_native_cond_broadcast(&log_queue->cond);
142
+ rb_nativethread_lock_unlock(&log_queue->lock);
143
+ return Qnil;
144
+ }
145
+
146
+ size_t size = log_queue->size;
147
+ ruby_whisper_log *log;
148
+ size_t i;
149
+ for (i = 0; i < size; i++) {
150
+ log = &log_queue->logs[(log_queue->tail + i) % LOG_QUEUE_CAPACITY];
151
+ logs[i].level = log->level;
152
+ logs[i].length = log->length;
153
+ char *text = malloc(log->length);
154
+ if (!text) {
155
+ logs[i].text = NULL;
156
+ continue;
157
+ }
158
+ logs[i].text = text;
159
+ memcpy(logs[i].text, log->text, log->length);
160
+ }
161
+ log_queue->size = 0;
162
+ log_queue->tail = log_queue->head;
163
+
164
+ rb_native_cond_signal(&log_queue->cond);
165
+
166
+ rb_nativethread_lock_unlock(&log_queue->lock);
167
+
168
+ VALUE rb_logs = rb_ary_new2(size);
169
+ VALUE rb_text;
170
+ for (i = 0; i < size; i++) {
171
+ if (!logs[i].text) {
172
+ continue;
173
+ }
174
+ rb_text = rb_str_new(logs[i].text, logs[i].length);
175
+ free(logs[i].text);
176
+ rb_ary_push(rb_logs, rb_ary_new3(2, INT2NUM(logs[i].level), rb_text));
177
+ }
178
+
179
+ return rb_logs;
180
+ }
@@ -0,0 +1,47 @@
1
+ #ifndef RUBY_WHISPER_LOG_SETTABLE_H
2
+ #define RUBY_WHISPER_LOG_SETTABLE_H
3
+
4
+ #define LOG_SETTABLE_SETUP(log_queue, mod, log_set) \
5
+ static VALUE \
6
+ ruby_whisper_##log_queue##_s_drain_logs(VALUE self) \
7
+ { \
8
+ return ruby_whisper_log_queue_drain(&log_queue); \
9
+ } \
10
+ static void \
11
+ ruby_whisper_##log_queue##_log_callback(enum ggml_log_level level, const char *text, void *user_data) \
12
+ { \
13
+ ruby_whisper_log_queue_enqueue(&log_queue, level, text); \
14
+ } \
15
+ static VALUE \
16
+ ruby_whisper_##log_queue##_s_log_set(VALUE self, VALUE log_callback, VALUE user_data) \
17
+ { \
18
+ rb_iv_set(self, "@log_callback", log_callback); \
19
+ rb_iv_set(self, "@log_callback_user_data", user_data); \
20
+ if (NIL_P(log_callback)) { \
21
+ log_set(NULL, NULL); \
22
+ } else { \
23
+ ruby_whisper_log_queue_open(&log_queue); \
24
+ rb_funcall((mod), id_start_log_callback_thread, 0); \
25
+ log_set(ruby_whisper_##log_queue##_log_callback, NULL); \
26
+ } \
27
+ return Qnil; \
28
+ } \
29
+ static void \
30
+ ruby_whisper_##log_queue##_end_proc(VALUE args) \
31
+ { \
32
+ ruby_whisper_log_queue_close(&log_queue); \
33
+ VALUE log_callback_thread = rb_ivar_get(mod, id_log_callback_thread); \
34
+ if (!NIL_P(log_callback_thread) && RTEST(rb_funcall(log_callback_thread, id_alive_p, 0))) { \
35
+ rb_funcall(log_callback_thread, id_join, 0); \
36
+ } \
37
+ }
38
+
39
+ #define LOG_SETTABLE_INIT(log_queue, mod) \
40
+ ruby_whisper_log_queue_initialize(&log_queue); \
41
+ rb_define_singleton_method(mod, "drain_logs", ruby_whisper_##log_queue##_s_drain_logs, 0); \
42
+ rb_define_singleton_method(mod, "log_set", ruby_whisper_##log_queue##_s_log_set, 2); \
43
+ rb_set_end_proc(ruby_whisper_##log_queue##_end_proc, Qnil); \
44
+ rb_extend_object(mod, mLogSettable); \
45
+ rb_funcall(mLogSettable, id_extended, 1, mod);
46
+
47
+ #endif
@@ -0,0 +1,49 @@
1
+ #include "ruby_whisper.h"
2
+ #include <stdio.h>
3
+ #include <unistd.h>
4
+
5
+ extern VALUE mParakeet;
6
+ extern VALUE mLogSettable;
7
+ extern VALUE cParakeetContext;
8
+ extern VALUE cParakeetSegment;
9
+ extern VALUE mOutputContext;
10
+ extern VALUE mOutputSegment;
11
+
12
+ extern void init_ruby_whisper_parakeet_params(VALUE *mParakeet);
13
+ extern void init_ruby_whisper_parakeet_token(VALUE *mParakeet);
14
+ extern void init_ruby_whisper_parakeet_segment(VALUE *mParakeet);
15
+ extern VALUE init_ruby_whisper_parakeet_context(VALUE *mParakeet);
16
+ extern void init_ruby_whisper_parakeet_context_params(VALUE *cParakeetContext);
17
+ extern void init_ruby_whisper_parakeet_model(VALUE *mParakeet);
18
+
19
+ static ruby_whisper_log_queue parakeet_log_queue;
20
+
21
+ LOG_SETTABLE_SETUP(parakeet_log_queue, mParakeet, parakeet_log_set)
22
+
23
+ static VALUE
24
+ ruby_whisper_parakeet_s_system_info_str(VALUE self)
25
+ {
26
+ return rb_str_new2(parakeet_print_system_info());
27
+ }
28
+
29
+ void
30
+ init_ruby_whisper_parakeet(VALUE *mWhisper)
31
+ {
32
+ mParakeet = rb_define_module_under(*mWhisper, "Parakeet");
33
+
34
+ rb_define_const(mParakeet, "VERSION", rb_str_new2(parakeet_version()));
35
+
36
+ LOG_SETTABLE_INIT(parakeet_log_queue, mParakeet)
37
+
38
+ rb_define_singleton_method(mParakeet, "system_info_str", ruby_whisper_parakeet_s_system_info_str, 0);
39
+
40
+ init_ruby_whisper_parakeet_params(&mParakeet);
41
+ init_ruby_whisper_parakeet_token(&mParakeet);
42
+ init_ruby_whisper_parakeet_segment(&mParakeet);
43
+ cParakeetContext = init_ruby_whisper_parakeet_context(&mParakeet);
44
+ init_ruby_whisper_parakeet_context_params(&cParakeetContext);
45
+ init_ruby_whisper_parakeet_model(&mParakeet);
46
+
47
+ rb_include_module(cParakeetContext, mOutputContext);
48
+ rb_include_module(cParakeetSegment, mOutputSegment);
49
+ }