cohere-transcribe 0.1.0

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 (480) hide show
  1. checksums.yaml +7 -0
  2. data/CHANGELOG.md +21 -0
  3. data/LICENSE.txt +201 -0
  4. data/NOTICE +5 -0
  5. data/README.md +265 -0
  6. data/THIRD_PARTY_NOTICES.md +67 -0
  7. data/exe/cohere-transcribe +8 -0
  8. data/exe/cohere-transcribe-doctor +8 -0
  9. data/ext/cohere_transcribe_native/CMakeLists.txt +262 -0
  10. data/ext/cohere_transcribe_native/README.md +26 -0
  11. data/ext/cohere_transcribe_native/audio_abi.cpp +1416 -0
  12. data/ext/cohere_transcribe_native/audio_exports.macos +5 -0
  13. data/ext/cohere_transcribe_native/audio_exports.map +10 -0
  14. data/ext/cohere_transcribe_native/cohere_abi.cpp +1209 -0
  15. data/ext/cohere_transcribe_native/exports.macos +40 -0
  16. data/ext/cohere_transcribe_native/exports.map +45 -0
  17. data/ext/cohere_transcribe_native/extconf.rb +89 -0
  18. data/ext/cohere_transcribe_native/test/abi_smoke.rb +232 -0
  19. data/ext/cohere_transcribe_native/test/audio_matrix_smoke.cpp +121 -0
  20. data/ext/cohere_transcribe_native/test/audio_reliability_smoke.cpp +198 -0
  21. data/ext/cohere_transcribe_native/test/native_abi_reliability_smoke.cpp +186 -0
  22. data/ext/cohere_transcribe_native/test/native_batch_projection_probe.rb +81 -0
  23. data/ext/cohere_transcribe_native/test/native_cancellation_smoke.cpp +194 -0
  24. data/lib/cohere/transcribe/alignment/ATTRIBUTION.md +23 -0
  25. data/lib/cohere/transcribe/alignment/LICENSE.ctc-forced-aligner +407 -0
  26. data/lib/cohere/transcribe/alignment/LICENSE.torchaudio +25 -0
  27. data/lib/cohere/transcribe/alignment/LICENSE.uroman +26 -0
  28. data/lib/cohere/transcribe/alignment/aligner.rb +476 -0
  29. data/lib/cohere/transcribe/alignment/ctc.rb +224 -0
  30. data/lib/cohere/transcribe/alignment/text.rb +237 -0
  31. data/lib/cohere/transcribe/alignment/uroman_data.rb +4950 -0
  32. data/lib/cohere/transcribe/api.rb +173 -0
  33. data/lib/cohere/transcribe/asr/batching.rb +472 -0
  34. data/lib/cohere/transcribe/asr/failure_policy.rb +112 -0
  35. data/lib/cohere/transcribe/asr/native.rb +676 -0
  36. data/lib/cohere/transcribe/audio/ATTRIBUTION.md +8 -0
  37. data/lib/cohere/transcribe/audio/LICENSE.auditok +21 -0
  38. data/lib/cohere/transcribe/audio/decoder.rb +315 -0
  39. data/lib/cohere/transcribe/audio/ffmpeg_native.rb +248 -0
  40. data/lib/cohere/transcribe/audio/segmentation.rb +260 -0
  41. data/lib/cohere/transcribe/cli.rb +727 -0
  42. data/lib/cohere/transcribe/configuration.rb +282 -0
  43. data/lib/cohere/transcribe/constants.rb +14 -0
  44. data/lib/cohere/transcribe/dense_converter.rb +548 -0
  45. data/lib/cohere/transcribe/doctor.rb +576 -0
  46. data/lib/cohere/transcribe/errors.rb +57 -0
  47. data/lib/cohere/transcribe/gguf_writer.rb +268 -0
  48. data/lib/cohere/transcribe/hub.rb +436 -0
  49. data/lib/cohere/transcribe/input.rb +110 -0
  50. data/lib/cohere/transcribe/licenses/crispasr.txt +21 -0
  51. data/lib/cohere/transcribe/loader.rb +128 -0
  52. data/lib/cohere/transcribe/model_identity.rb +440 -0
  53. data/lib/cohere/transcribe/output/publication.rb +1118 -0
  54. data/lib/cohere/transcribe/output/rendering.rb +105 -0
  55. data/lib/cohere/transcribe/output/timing.rb +86 -0
  56. data/lib/cohere/transcribe/python_text.rb +70 -0
  57. data/lib/cohere/transcribe/pytorch_checkpoint.rb +1180 -0
  58. data/lib/cohere/transcribe/runtime/engine.rb +1676 -0
  59. data/lib/cohere/transcribe/runtime/model_provider.rb +390 -0
  60. data/lib/cohere/transcribe/runtime/precision.rb +57 -0
  61. data/lib/cohere/transcribe/runtime/preparation.rb +215 -0
  62. data/lib/cohere/transcribe/runtime/resources.rb +165 -0
  63. data/lib/cohere/transcribe/runtime/word_pipeline.rb +364 -0
  64. data/lib/cohere/transcribe/safetensors.rb +579 -0
  65. data/lib/cohere/transcribe/state/checkpoint.rb +224 -0
  66. data/lib/cohere/transcribe/state/contracts.rb +141 -0
  67. data/lib/cohere/transcribe/state/io.rb +727 -0
  68. data/lib/cohere/transcribe/state/locking.rb +211 -0
  69. data/lib/cohere/transcribe/state/manifest.rb +155 -0
  70. data/lib/cohere/transcribe/state.rb +7 -0
  71. data/lib/cohere/transcribe/types.rb +535 -0
  72. data/lib/cohere/transcribe/vad/ATTRIBUTION.md +14 -0
  73. data/lib/cohere/transcribe/vad/LICENSE.faster-whisper +21 -0
  74. data/lib/cohere/transcribe/vad/LICENSE.silero-vad +21 -0
  75. data/lib/cohere/transcribe/vad/silero.rb +344 -0
  76. data/lib/cohere/transcribe/vad/silero_vad_v6.onnx +0 -0
  77. data/lib/cohere/transcribe/vad/timestamps.rb +219 -0
  78. data/lib/cohere/transcribe/version.rb +7 -0
  79. data/lib/cohere/transcribe.rb +26 -0
  80. data/sig/cohere/transcribe.rbs +250 -0
  81. data/vendor/crispasr/AUTHORS +510 -0
  82. data/vendor/crispasr/LICENSE +21 -0
  83. data/vendor/crispasr/UPSTREAM.md +9 -0
  84. data/vendor/crispasr/VERSION +1 -0
  85. data/vendor/crispasr/ggml/AUTHORS +335 -0
  86. data/vendor/crispasr/ggml/CMakeLists.txt +512 -0
  87. data/vendor/crispasr/ggml/LICENSE +21 -0
  88. data/vendor/crispasr/ggml/README.md +49 -0
  89. data/vendor/crispasr/ggml/cmake/FindNCCL.cmake +36 -0
  90. data/vendor/crispasr/ggml/cmake/GitVars.cmake +22 -0
  91. data/vendor/crispasr/ggml/cmake/common.cmake +50 -0
  92. data/vendor/crispasr/ggml/cmake/ggml-config.cmake.in +191 -0
  93. data/vendor/crispasr/ggml/ggml.pc.in +10 -0
  94. data/vendor/crispasr/ggml/include/ggml-alloc.h +85 -0
  95. data/vendor/crispasr/ggml/include/ggml-backend.h +431 -0
  96. data/vendor/crispasr/ggml/include/ggml-blas.h +25 -0
  97. data/vendor/crispasr/ggml/include/ggml-cann.h +123 -0
  98. data/vendor/crispasr/ggml/include/ggml-cpp.h +39 -0
  99. data/vendor/crispasr/ggml/include/ggml-cpu.h +151 -0
  100. data/vendor/crispasr/ggml/include/ggml-cuda.h +50 -0
  101. data/vendor/crispasr/ggml/include/ggml-hexagon.h +19 -0
  102. data/vendor/crispasr/ggml/include/ggml-metal.h +61 -0
  103. data/vendor/crispasr/ggml/include/ggml-opencl.h +26 -0
  104. data/vendor/crispasr/ggml/include/ggml-openvino.h +37 -0
  105. data/vendor/crispasr/ggml/include/ggml-opt.h +256 -0
  106. data/vendor/crispasr/ggml/include/ggml-rpc.h +35 -0
  107. data/vendor/crispasr/ggml/include/ggml-sycl.h +49 -0
  108. data/vendor/crispasr/ggml/include/ggml-virtgpu.h +14 -0
  109. data/vendor/crispasr/ggml/include/ggml-vulkan.h +29 -0
  110. data/vendor/crispasr/ggml/include/ggml-webgpu.h +19 -0
  111. data/vendor/crispasr/ggml/include/ggml-zdnn.h +17 -0
  112. data/vendor/crispasr/ggml/include/ggml-zendnn.h +22 -0
  113. data/vendor/crispasr/ggml/include/ggml.h +2887 -0
  114. data/vendor/crispasr/ggml/include/gguf.h +204 -0
  115. data/vendor/crispasr/ggml/src/CMakeLists.txt +493 -0
  116. data/vendor/crispasr/ggml/src/ggml-alloc.c +1323 -0
  117. data/vendor/crispasr/ggml/src/ggml-backend-dl.cpp +48 -0
  118. data/vendor/crispasr/ggml/src/ggml-backend-dl.h +44 -0
  119. data/vendor/crispasr/ggml/src/ggml-backend-impl.h +275 -0
  120. data/vendor/crispasr/ggml/src/ggml-backend-meta.cpp +2145 -0
  121. data/vendor/crispasr/ggml/src/ggml-backend-reg.cpp +586 -0
  122. data/vendor/crispasr/ggml/src/ggml-backend.cpp +2437 -0
  123. data/vendor/crispasr/ggml/src/ggml-common.h +1900 -0
  124. data/vendor/crispasr/ggml/src/ggml-cpu/CMakeLists.txt +718 -0
  125. data/vendor/crispasr/ggml/src/ggml-cpu/amx/amx.cpp +249 -0
  126. data/vendor/crispasr/ggml/src/ggml-cpu/amx/amx.h +8 -0
  127. data/vendor/crispasr/ggml/src/ggml-cpu/amx/common.h +115 -0
  128. data/vendor/crispasr/ggml/src/ggml-cpu/amx/mmq.cpp +2512 -0
  129. data/vendor/crispasr/ggml/src/ggml-cpu/amx/mmq.h +10 -0
  130. data/vendor/crispasr/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +98 -0
  131. data/vendor/crispasr/ggml/src/ggml-cpu/arch/arm/quants.c +4244 -0
  132. data/vendor/crispasr/ggml/src/ggml-cpu/arch/arm/repack.cpp +5156 -0
  133. data/vendor/crispasr/ggml/src/ggml-cpu/arch/loongarch/quants.c +2158 -0
  134. data/vendor/crispasr/ggml/src/ggml-cpu/arch/powerpc/cpu-feats.cpp +82 -0
  135. data/vendor/crispasr/ggml/src/ggml-cpu/arch/powerpc/quants.c +2304 -0
  136. data/vendor/crispasr/ggml/src/ggml-cpu/arch/riscv/cpu-feats.cpp +38 -0
  137. data/vendor/crispasr/ggml/src/ggml-cpu/arch/riscv/quants.c +4455 -0
  138. data/vendor/crispasr/ggml/src/ggml-cpu/arch/riscv/repack.cpp +1703 -0
  139. data/vendor/crispasr/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
  140. data/vendor/crispasr/ggml/src/ggml-cpu/arch/s390/quants.c +1465 -0
  141. data/vendor/crispasr/ggml/src/ggml-cpu/arch/wasm/quants.c +1220 -0
  142. data/vendor/crispasr/ggml/src/ggml-cpu/arch/x86/cpu-feats.cpp +327 -0
  143. data/vendor/crispasr/ggml/src/ggml-cpu/arch/x86/quants.c +3970 -0
  144. data/vendor/crispasr/ggml/src/ggml-cpu/arch/x86/repack.cpp +6407 -0
  145. data/vendor/crispasr/ggml/src/ggml-cpu/arch-fallback.h +349 -0
  146. data/vendor/crispasr/ggml/src/ggml-cpu/binary-ops.cpp +154 -0
  147. data/vendor/crispasr/ggml/src/ggml-cpu/binary-ops.h +16 -0
  148. data/vendor/crispasr/ggml/src/ggml-cpu/cmake/FindSIMD.cmake +100 -0
  149. data/vendor/crispasr/ggml/src/ggml-cpu/common.h +95 -0
  150. data/vendor/crispasr/ggml/src/ggml-cpu/ggml-cpu-impl.h +539 -0
  151. data/vendor/crispasr/ggml/src/ggml-cpu/ggml-cpu.c +3791 -0
  152. data/vendor/crispasr/ggml/src/ggml-cpu/ggml-cpu.cpp +703 -0
  153. data/vendor/crispasr/ggml/src/ggml-cpu/hbm.cpp +55 -0
  154. data/vendor/crispasr/ggml/src/ggml-cpu/hbm.h +8 -0
  155. data/vendor/crispasr/ggml/src/ggml-cpu/kleidiai/kernels.cpp +939 -0
  156. data/vendor/crispasr/ggml/src/ggml-cpu/kleidiai/kernels.h +90 -0
  157. data/vendor/crispasr/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +1513 -0
  158. data/vendor/crispasr/ggml/src/ggml-cpu/kleidiai/kleidiai.h +17 -0
  159. data/vendor/crispasr/ggml/src/ggml-cpu/llamafile/sgemm.cpp +4051 -0
  160. data/vendor/crispasr/ggml/src/ggml-cpu/llamafile/sgemm.h +25 -0
  161. data/vendor/crispasr/ggml/src/ggml-cpu/ops.cpp +11662 -0
  162. data/vendor/crispasr/ggml/src/ggml-cpu/ops.h +121 -0
  163. data/vendor/crispasr/ggml/src/ggml-cpu/quants.c +1288 -0
  164. data/vendor/crispasr/ggml/src/ggml-cpu/quants.h +103 -0
  165. data/vendor/crispasr/ggml/src/ggml-cpu/repack.cpp +4836 -0
  166. data/vendor/crispasr/ggml/src/ggml-cpu/repack.h +245 -0
  167. data/vendor/crispasr/ggml/src/ggml-cpu/simd-gemm.h +226 -0
  168. data/vendor/crispasr/ggml/src/ggml-cpu/simd-mappings.h +1329 -0
  169. data/vendor/crispasr/ggml/src/ggml-cpu/spacemit/ime.cpp +1025 -0
  170. data/vendor/crispasr/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
  171. data/vendor/crispasr/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
  172. data/vendor/crispasr/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
  173. data/vendor/crispasr/ggml/src/ggml-cpu/traits.cpp +36 -0
  174. data/vendor/crispasr/ggml/src/ggml-cpu/traits.h +38 -0
  175. data/vendor/crispasr/ggml/src/ggml-cpu/unary-ops.cpp +336 -0
  176. data/vendor/crispasr/ggml/src/ggml-cpu/unary-ops.h +35 -0
  177. data/vendor/crispasr/ggml/src/ggml-cpu/vec.cpp +681 -0
  178. data/vendor/crispasr/ggml/src/ggml-cpu/vec.h +1606 -0
  179. data/vendor/crispasr/ggml/src/ggml-cuda/CMakeLists.txt +272 -0
  180. data/vendor/crispasr/ggml/src/ggml-cuda/acc.cu +61 -0
  181. data/vendor/crispasr/ggml/src/ggml-cuda/acc.cuh +5 -0
  182. data/vendor/crispasr/ggml/src/ggml-cuda/add-id.cu +58 -0
  183. data/vendor/crispasr/ggml/src/ggml-cuda/add-id.cuh +3 -0
  184. data/vendor/crispasr/ggml/src/ggml-cuda/arange.cu +34 -0
  185. data/vendor/crispasr/ggml/src/ggml-cuda/arange.cuh +5 -0
  186. data/vendor/crispasr/ggml/src/ggml-cuda/argmax.cu +91 -0
  187. data/vendor/crispasr/ggml/src/ggml-cuda/argmax.cuh +3 -0
  188. data/vendor/crispasr/ggml/src/ggml-cuda/argsort.cu +265 -0
  189. data/vendor/crispasr/ggml/src/ggml-cuda/argsort.cuh +19 -0
  190. data/vendor/crispasr/ggml/src/ggml-cuda/binbcast.cu +534 -0
  191. data/vendor/crispasr/ggml/src/ggml-cuda/binbcast.cuh +12 -0
  192. data/vendor/crispasr/ggml/src/ggml-cuda/clamp.cu +45 -0
  193. data/vendor/crispasr/ggml/src/ggml-cuda/clamp.cuh +5 -0
  194. data/vendor/crispasr/ggml/src/ggml-cuda/col2im-1d.cu +81 -0
  195. data/vendor/crispasr/ggml/src/ggml-cuda/col2im-1d.cuh +3 -0
  196. data/vendor/crispasr/ggml/src/ggml-cuda/common.cuh +1489 -0
  197. data/vendor/crispasr/ggml/src/ggml-cuda/concat.cu +204 -0
  198. data/vendor/crispasr/ggml/src/ggml-cuda/concat.cuh +5 -0
  199. data/vendor/crispasr/ggml/src/ggml-cuda/conv-transpose-1d.cu +97 -0
  200. data/vendor/crispasr/ggml/src/ggml-cuda/conv-transpose-1d.cuh +5 -0
  201. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d-dw.cu +161 -0
  202. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d-dw.cuh +5 -0
  203. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d-transpose.cu +115 -0
  204. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d-transpose.cuh +5 -0
  205. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d.cu +166 -0
  206. data/vendor/crispasr/ggml/src/ggml-cuda/conv2d.cuh +5 -0
  207. data/vendor/crispasr/ggml/src/ggml-cuda/convert.cu +892 -0
  208. data/vendor/crispasr/ggml/src/ggml-cuda/convert.cuh +66 -0
  209. data/vendor/crispasr/ggml/src/ggml-cuda/count-equal.cu +64 -0
  210. data/vendor/crispasr/ggml/src/ggml-cuda/count-equal.cuh +5 -0
  211. data/vendor/crispasr/ggml/src/ggml-cuda/cp-async.cuh +57 -0
  212. data/vendor/crispasr/ggml/src/ggml-cuda/cpy-utils.cuh +217 -0
  213. data/vendor/crispasr/ggml/src/ggml-cuda/cpy.cu +581 -0
  214. data/vendor/crispasr/ggml/src/ggml-cuda/cpy.cuh +7 -0
  215. data/vendor/crispasr/ggml/src/ggml-cuda/cross-entropy-loss.cu +177 -0
  216. data/vendor/crispasr/ggml/src/ggml-cuda/cross-entropy-loss.cuh +7 -0
  217. data/vendor/crispasr/ggml/src/ggml-cuda/cumsum.cu +307 -0
  218. data/vendor/crispasr/ggml/src/ggml-cuda/cumsum.cuh +5 -0
  219. data/vendor/crispasr/ggml/src/ggml-cuda/dequantize.cuh +99 -0
  220. data/vendor/crispasr/ggml/src/ggml-cuda/diag.cu +77 -0
  221. data/vendor/crispasr/ggml/src/ggml-cuda/diag.cuh +5 -0
  222. data/vendor/crispasr/ggml/src/ggml-cuda/diagmask.cu +40 -0
  223. data/vendor/crispasr/ggml/src/ggml-cuda/diagmask.cuh +5 -0
  224. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-common.cuh +1212 -0
  225. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-mma-f16.cuh +1860 -0
  226. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-tile.cu +57 -0
  227. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-tile.cuh +1309 -0
  228. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-vec.cuh +600 -0
  229. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-wmma-f16.cu +696 -0
  230. data/vendor/crispasr/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +51 -0
  231. data/vendor/crispasr/ggml/src/ggml-cuda/fattn.cu +620 -0
  232. data/vendor/crispasr/ggml/src/ggml-cuda/fattn.cuh +5 -0
  233. data/vendor/crispasr/ggml/src/ggml-cuda/fill.cu +37 -0
  234. data/vendor/crispasr/ggml/src/ggml-cuda/fill.cuh +3 -0
  235. data/vendor/crispasr/ggml/src/ggml-cuda/gated_delta_net.cu +273 -0
  236. data/vendor/crispasr/ggml/src/ggml-cuda/gated_delta_net.cuh +4 -0
  237. data/vendor/crispasr/ggml/src/ggml-cuda/getrows.cu +332 -0
  238. data/vendor/crispasr/ggml/src/ggml-cuda/getrows.cuh +15 -0
  239. data/vendor/crispasr/ggml/src/ggml-cuda/ggml-cuda.cu +5580 -0
  240. data/vendor/crispasr/ggml/src/ggml-cuda/gla.cu +93 -0
  241. data/vendor/crispasr/ggml/src/ggml-cuda/gla.cuh +3 -0
  242. data/vendor/crispasr/ggml/src/ggml-cuda/im2col.cu +274 -0
  243. data/vendor/crispasr/ggml/src/ggml-cuda/im2col.cuh +6 -0
  244. data/vendor/crispasr/ggml/src/ggml-cuda/mean.cu +75 -0
  245. data/vendor/crispasr/ggml/src/ggml-cuda/mean.cuh +3 -0
  246. data/vendor/crispasr/ggml/src/ggml-cuda/mma.cuh +1333 -0
  247. data/vendor/crispasr/ggml/src/ggml-cuda/mmf.cu +191 -0
  248. data/vendor/crispasr/ggml/src/ggml-cuda/mmf.cuh +908 -0
  249. data/vendor/crispasr/ggml/src/ggml-cuda/mmid.cu +164 -0
  250. data/vendor/crispasr/ggml/src/ggml-cuda/mmid.cuh +5 -0
  251. data/vendor/crispasr/ggml/src/ggml-cuda/mmq.cu +372 -0
  252. data/vendor/crispasr/ggml/src/ggml-cuda/mmq.cuh +4175 -0
  253. data/vendor/crispasr/ggml/src/ggml-cuda/mmvf.cu +862 -0
  254. data/vendor/crispasr/ggml/src/ggml-cuda/mmvf.cuh +14 -0
  255. data/vendor/crispasr/ggml/src/ggml-cuda/mmvq.cu +1161 -0
  256. data/vendor/crispasr/ggml/src/ggml-cuda/mmvq.cuh +16 -0
  257. data/vendor/crispasr/ggml/src/ggml-cuda/norm.cu +756 -0
  258. data/vendor/crispasr/ggml/src/ggml-cuda/norm.cuh +20 -0
  259. data/vendor/crispasr/ggml/src/ggml-cuda/opt-step-adamw.cu +78 -0
  260. data/vendor/crispasr/ggml/src/ggml-cuda/opt-step-adamw.cuh +5 -0
  261. data/vendor/crispasr/ggml/src/ggml-cuda/opt-step-sgd.cu +49 -0
  262. data/vendor/crispasr/ggml/src/ggml-cuda/opt-step-sgd.cuh +5 -0
  263. data/vendor/crispasr/ggml/src/ggml-cuda/out-prod.cu +68 -0
  264. data/vendor/crispasr/ggml/src/ggml-cuda/out-prod.cuh +3 -0
  265. data/vendor/crispasr/ggml/src/ggml-cuda/pad.cu +106 -0
  266. data/vendor/crispasr/ggml/src/ggml-cuda/pad.cuh +5 -0
  267. data/vendor/crispasr/ggml/src/ggml-cuda/pad_reflect_1d.cu +91 -0
  268. data/vendor/crispasr/ggml/src/ggml-cuda/pad_reflect_1d.cuh +5 -0
  269. data/vendor/crispasr/ggml/src/ggml-cuda/pool2d.cu +94 -0
  270. data/vendor/crispasr/ggml/src/ggml-cuda/pool2d.cuh +5 -0
  271. data/vendor/crispasr/ggml/src/ggml-cuda/quantize.cu +443 -0
  272. data/vendor/crispasr/ggml/src/ggml-cuda/quantize.cuh +41 -0
  273. data/vendor/crispasr/ggml/src/ggml-cuda/reduce_rows.cuh +39 -0
  274. data/vendor/crispasr/ggml/src/ggml-cuda/roll.cu +67 -0
  275. data/vendor/crispasr/ggml/src/ggml-cuda/roll.cuh +5 -0
  276. data/vendor/crispasr/ggml/src/ggml-cuda/rope.cu +665 -0
  277. data/vendor/crispasr/ggml/src/ggml-cuda/rope.cuh +9 -0
  278. data/vendor/crispasr/ggml/src/ggml-cuda/scale.cu +34 -0
  279. data/vendor/crispasr/ggml/src/ggml-cuda/scale.cuh +5 -0
  280. data/vendor/crispasr/ggml/src/ggml-cuda/set-rows.cu +330 -0
  281. data/vendor/crispasr/ggml/src/ggml-cuda/set-rows.cuh +7 -0
  282. data/vendor/crispasr/ggml/src/ggml-cuda/set.cu +39 -0
  283. data/vendor/crispasr/ggml/src/ggml-cuda/set.cuh +7 -0
  284. data/vendor/crispasr/ggml/src/ggml-cuda/softcap.cu +34 -0
  285. data/vendor/crispasr/ggml/src/ggml-cuda/softcap.cuh +5 -0
  286. data/vendor/crispasr/ggml/src/ggml-cuda/softmax.cu +472 -0
  287. data/vendor/crispasr/ggml/src/ggml-cuda/softmax.cuh +7 -0
  288. data/vendor/crispasr/ggml/src/ggml-cuda/solve_tri.cu +275 -0
  289. data/vendor/crispasr/ggml/src/ggml-cuda/solve_tri.cuh +3 -0
  290. data/vendor/crispasr/ggml/src/ggml-cuda/ssm-conv.cu +197 -0
  291. data/vendor/crispasr/ggml/src/ggml-cuda/ssm-conv.cuh +3 -0
  292. data/vendor/crispasr/ggml/src/ggml-cuda/ssm-scan.cu +342 -0
  293. data/vendor/crispasr/ggml/src/ggml-cuda/ssm-scan.cuh +3 -0
  294. data/vendor/crispasr/ggml/src/ggml-cuda/sum.cu +41 -0
  295. data/vendor/crispasr/ggml/src/ggml-cuda/sum.cuh +5 -0
  296. data/vendor/crispasr/ggml/src/ggml-cuda/sumrows.cu +43 -0
  297. data/vendor/crispasr/ggml/src/ggml-cuda/sumrows.cuh +4 -0
  298. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +5 -0
  299. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +6 -0
  300. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +11 -0
  301. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_1.cu +10 -0
  302. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_2.cu +10 -0
  303. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +12 -0
  304. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +5 -0
  305. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +6 -0
  306. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +12 -0
  307. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +11 -0
  308. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_1.cu +10 -0
  309. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_2.cu +10 -0
  310. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +5 -0
  311. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_2.cu +10 -0
  312. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +12 -0
  313. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +11 -0
  314. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_64-ncols2_1.cu +10 -0
  315. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_1.cu +10 -0
  316. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_2.cu +10 -0
  317. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +12 -0
  318. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +11 -0
  319. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
  320. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
  321. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
  322. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq320-dv256.cu +5 -0
  323. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
  324. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq512-dv512.cu +5 -0
  325. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
  326. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
  327. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
  328. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
  329. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
  330. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu +7 -0
  331. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-f16.cu +7 -0
  332. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_0.cu +7 -0
  333. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_1.cu +7 -0
  334. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_0.cu +7 -0
  335. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_1.cu +7 -0
  336. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q8_0.cu +7 -0
  337. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-bf16.cu +7 -0
  338. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  339. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  340. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  341. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  342. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  343. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  344. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-bf16.cu +7 -0
  345. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  346. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  347. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  348. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  349. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  350. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  351. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-bf16.cu +7 -0
  352. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  353. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  354. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  355. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  356. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  357. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  358. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-bf16.cu +7 -0
  359. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  360. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  361. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  362. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  363. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  364. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  365. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-bf16.cu +7 -0
  366. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  367. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  368. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  369. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  370. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  371. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  372. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-bf16.cu +7 -0
  373. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  374. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  375. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  376. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  377. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  378. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  379. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  380. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  381. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  382. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  383. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  384. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  385. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  386. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  387. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  388. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  389. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  390. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  391. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  392. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  393. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  394. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  395. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_s.cu +5 -0
  396. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_s.cu +5 -0
  397. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xs.cu +5 -0
  398. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xxs.cu +5 -0
  399. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_s.cu +5 -0
  400. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_xxs.cu +5 -0
  401. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_nl.cu +5 -0
  402. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_xs.cu +5 -0
  403. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-mxfp4.cu +5 -0
  404. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-nvfp4.cu +5 -0
  405. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q1_0.cu +5 -0
  406. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k.cu +5 -0
  407. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k.cu +5 -0
  408. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_0.cu +5 -0
  409. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_1.cu +5 -0
  410. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_k.cu +5 -0
  411. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_0.cu +5 -0
  412. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1.cu +5 -0
  413. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_k.cu +5 -0
  414. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q6_k.cu +5 -0
  415. data/vendor/crispasr/ggml/src/ggml-cuda/template-instances/mmq-instance-q8_0.cu +5 -0
  416. data/vendor/crispasr/ggml/src/ggml-cuda/top-k.cu +95 -0
  417. data/vendor/crispasr/ggml/src/ggml-cuda/top-k.cuh +3 -0
  418. data/vendor/crispasr/ggml/src/ggml-cuda/topk-moe.cu +415 -0
  419. data/vendor/crispasr/ggml/src/ggml-cuda/topk-moe.cuh +27 -0
  420. data/vendor/crispasr/ggml/src/ggml-cuda/tri.cu +136 -0
  421. data/vendor/crispasr/ggml/src/ggml-cuda/tri.cuh +5 -0
  422. data/vendor/crispasr/ggml/src/ggml-cuda/tsembd.cu +47 -0
  423. data/vendor/crispasr/ggml/src/ggml-cuda/tsembd.cuh +5 -0
  424. data/vendor/crispasr/ggml/src/ggml-cuda/unary.cu +644 -0
  425. data/vendor/crispasr/ggml/src/ggml-cuda/unary.cuh +116 -0
  426. data/vendor/crispasr/ggml/src/ggml-cuda/upscale.cu +293 -0
  427. data/vendor/crispasr/ggml/src/ggml-cuda/upscale.cuh +5 -0
  428. data/vendor/crispasr/ggml/src/ggml-cuda/vecdotq.cuh +1317 -0
  429. data/vendor/crispasr/ggml/src/ggml-cuda/vendors/cuda.h +28 -0
  430. data/vendor/crispasr/ggml/src/ggml-cuda/vendors/hip.h +306 -0
  431. data/vendor/crispasr/ggml/src/ggml-cuda/vendors/musa.h +148 -0
  432. data/vendor/crispasr/ggml/src/ggml-cuda/wkv.cu +199 -0
  433. data/vendor/crispasr/ggml/src/ggml-cuda/wkv.cuh +7 -0
  434. data/vendor/crispasr/ggml/src/ggml-impl.h +783 -0
  435. data/vendor/crispasr/ggml/src/ggml-metal/CMakeLists.txt +124 -0
  436. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-common.cpp +457 -0
  437. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  438. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-context.h +41 -0
  439. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-context.m +778 -0
  440. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-device.cpp +2117 -0
  441. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-device.h +299 -0
  442. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-device.m +2093 -0
  443. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-impl.h +1267 -0
  444. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-ops.cpp +5023 -0
  445. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal-ops.h +111 -0
  446. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal.cpp +954 -0
  447. data/vendor/crispasr/ggml/src/ggml-metal/ggml-metal.metal +11756 -0
  448. data/vendor/crispasr/ggml/src/ggml-opt.cpp +1094 -0
  449. data/vendor/crispasr/ggml/src/ggml-quants.c +5491 -0
  450. data/vendor/crispasr/ggml/src/ggml-quants.h +112 -0
  451. data/vendor/crispasr/ggml/src/ggml-threading.cpp +12 -0
  452. data/vendor/crispasr/ggml/src/ggml-threading.h +14 -0
  453. data/vendor/crispasr/ggml/src/ggml.c +7925 -0
  454. data/vendor/crispasr/ggml/src/ggml.cpp +26 -0
  455. data/vendor/crispasr/ggml/src/gguf.cpp +1556 -0
  456. data/vendor/crispasr/src/cohere-arch.h +137 -0
  457. data/vendor/crispasr/src/cohere.cpp +5642 -0
  458. data/vendor/crispasr/src/cohere.h +327 -0
  459. data/vendor/crispasr/src/cohere_batch_planner.h +82 -0
  460. data/vendor/crispasr/src/cohere_chunking.h +64 -0
  461. data/vendor/crispasr/src/cohere_decoder_batch_layout.h +60 -0
  462. data/vendor/crispasr/src/cohere_encoder_padded_layout.h +27 -0
  463. data/vendor/crispasr/src/cohere_frontend.cpp +189 -0
  464. data/vendor/crispasr/src/cohere_frontend.h +31 -0
  465. data/vendor/crispasr/src/cohere_ragged_controller.h +149 -0
  466. data/vendor/crispasr/src/cohere_token_renderer.h +181 -0
  467. data/vendor/crispasr/src/core/attention.h +924 -0
  468. data/vendor/crispasr/src/core/audio_chunking.h +97 -0
  469. data/vendor/crispasr/src/core/beam_decode.h +486 -0
  470. data/vendor/crispasr/src/core/cpu_ops.h +135 -0
  471. data/vendor/crispasr/src/core/gguf_loader.cpp +1021 -0
  472. data/vendor/crispasr/src/core/gguf_loader.h +216 -0
  473. data/vendor/crispasr/src/core/gpu_backend_pref.h +119 -0
  474. data/vendor/crispasr/src/core/mel.cpp +519 -0
  475. data/vendor/crispasr/src/core/mel.h +265 -0
  476. data/vendor/crispasr/src/core/ngram_loop_fix.h +173 -0
  477. data/vendor/crispasr/src/core/repetition_loop_guard.h +54 -0
  478. data/vendor/crispasr/src/crispasr_imatrix.cpp +255 -0
  479. data/vendor/crispasr/src/crispasr_imatrix.h +38 -0
  480. metadata +596 -0
@@ -0,0 +1,165 @@
1
+ # frozen_string_literal: true
2
+
3
+ require "monitor"
4
+ require "weakref"
5
+
6
+ module Cohere
7
+ module Transcribe
8
+ module Runtime
9
+ # Owns one reusable native ASR session. Across all ModelResources
10
+ # instances, only one may retain ASR at a time; this bounds process memory
11
+ # when applications create multiple reusable Transcribers.
12
+ class ModelResources
13
+ OWNER_GUARD = Monitor.new
14
+
15
+ class << self
16
+ def evict_current_asr_owner
17
+ OWNER_GUARD.synchronize do
18
+ owner = current_owner_locked
19
+ owner&.send(:evict_asr_locked)
20
+ end
21
+ nil
22
+ end
23
+
24
+ def current_asr_owner
25
+ OWNER_GUARD.synchronize { current_owner_locked }
26
+ end
27
+
28
+ private
29
+
30
+ def current_owner_locked
31
+ reference = @owner_reference
32
+ return nil unless reference
33
+
34
+ reference.__getobj__
35
+ rescue WeakRef::RefError
36
+ @owner_reference = nil
37
+ nil
38
+ end
39
+
40
+ def claim_locked(owner)
41
+ @owner_reference = WeakRef.new(owner)
42
+ end
43
+
44
+ def release_locked(owner)
45
+ @owner_reference = nil if current_owner_locked.equal?(owner)
46
+ end
47
+ end
48
+
49
+ attr_reader :asr_key, :batch_controller
50
+
51
+ def initialize
52
+ @asr_key = nil
53
+ @asr_session = nil
54
+ @batch_controller = nil
55
+ @closed = false
56
+ end
57
+
58
+ # Returns [session, loaded]. A key change evicts this instance's prior
59
+ # session, while acquisition also evicts a different process owner.
60
+ def acquire_asr(key)
61
+ raise ArgumentError, "an ASR loader block is required" unless block_given?
62
+
63
+ self.class::OWNER_GUARD.synchronize do
64
+ ensure_open!
65
+ evict_asr_locked if @asr_session && @asr_key != key
66
+
67
+ owner = self.class.send(:current_owner_locked)
68
+ owner.send(:evict_asr_locked) if owner && !owner.equal?(self)
69
+ self.class.send(:claim_locked, self)
70
+
71
+ loaded = false
72
+ unless @asr_session
73
+ installed = false
74
+ begin
75
+ session = yield
76
+ raise TranscriptionRuntimeError, "ASR loader returned no native session" if session.nil?
77
+
78
+ @asr_session = session
79
+ @asr_key = immutable_key(key)
80
+ @batch_controller = nil
81
+ loaded = true
82
+ installed = true
83
+ ensure
84
+ self.class.send(:release_locked, self) unless installed
85
+ end
86
+ end
87
+ [@asr_session, loaded].freeze
88
+ end
89
+ end
90
+
91
+ def install_batch_controller(controller)
92
+ self.class::OWNER_GUARD.synchronize do
93
+ ensure_open!
94
+ raise TranscriptionRuntimeError, "Cannot install an ASR controller before acquiring ASR" unless @asr_session
95
+
96
+ @batch_controller ||= controller
97
+ end
98
+ end
99
+
100
+ def asr_session
101
+ self.class::OWNER_GUARD.synchronize { @asr_session }
102
+ end
103
+
104
+ def asr?
105
+ self.class::OWNER_GUARD.synchronize { !@asr_session.nil? }
106
+ end
107
+ alias has_asr? asr?
108
+
109
+ def asr_circuit_broken?
110
+ self.class::OWNER_GUARD.synchronize do
111
+ @batch_controller&.circuit_open?
112
+ end || false
113
+ end
114
+
115
+ def evict_asr
116
+ self.class::OWNER_GUARD.synchronize { evict_asr_locked }
117
+ nil
118
+ end
119
+
120
+ def close
121
+ self.class::OWNER_GUARD.synchronize do
122
+ return if @closed
123
+
124
+ evict_asr_locked
125
+ @closed = true
126
+ end
127
+ nil
128
+ end
129
+
130
+ def closed?
131
+ self.class::OWNER_GUARD.synchronize { @closed }
132
+ end
133
+
134
+ private
135
+
136
+ def ensure_open!
137
+ raise TranscriberClosedError, "Model resources have been closed" if @closed
138
+ end
139
+
140
+ def evict_asr_locked
141
+ self.class.send(:release_locked, self)
142
+ session = @asr_session
143
+ @asr_session = nil
144
+ @asr_key = nil
145
+ @batch_controller = nil
146
+ session&.close
147
+ nil
148
+ end
149
+
150
+ def immutable_key(value)
151
+ case value
152
+ when Array
153
+ value.map { |item| immutable_key(item) }.freeze
154
+ when Hash
155
+ value.to_h { |key, item| [immutable_key(key), immutable_key(item)] }.freeze
156
+ when String
157
+ value.dup.freeze
158
+ else
159
+ value
160
+ end
161
+ end
162
+ end
163
+ end
164
+ end
165
+ end
@@ -0,0 +1,364 @@
1
+ # frozen_string_literal: true
2
+
3
+ module Cohere
4
+ module Transcribe
5
+ module Runtime
6
+ # Scheduling primitives for word-aligned runs. The coordinator creates a
7
+ # hard phase barrier: every ASR outcome is collected before Dense is
8
+ # evicted, and no alignment callback can run before that eviction.
9
+ module WordPipeline
10
+ UNSET = Object.new.freeze
11
+ private_constant :UNSET
12
+
13
+ # The complete state needed after ASR. Deliberately absent are decoded
14
+ # samples (or a decoder result): multi-file word mode must re-decode one
15
+ # source at a time instead of retaining every waveform until alignment.
16
+ AlignmentWork = Data.define(
17
+ :index,
18
+ :entry,
19
+ :plan,
20
+ :result,
21
+ :segment_times,
22
+ :speech_spans,
23
+ :generation_id,
24
+ :decode_backend,
25
+ :expected_sample_count,
26
+ :audio_required,
27
+ :source_snapshot
28
+ ) do
29
+ def initialize(index:, entry:, plan:, result:, segment_times:, speech_spans:,
30
+ generation_id:, decode_backend:, expected_sample_count:, audio_required:,
31
+ source_snapshot:)
32
+ index = Integer(index)
33
+ expected_sample_count = Integer(expected_sample_count)
34
+ raise ArgumentError, "word-pipeline index must be nonnegative" if index.negative?
35
+ raise ArgumentError, "expected alignment sample count must be nonnegative" if expected_sample_count.negative?
36
+
37
+ super(
38
+ index: index,
39
+ entry: entry,
40
+ plan: plan,
41
+ result: result,
42
+ segment_times: immutable_spans(segment_times, "segment_times"),
43
+ speech_spans: immutable_spans(speech_spans, "speech_spans"),
44
+ generation_id: immutable_optional_string(generation_id),
45
+ decode_backend: immutable_optional_string(decode_backend),
46
+ expected_sample_count: expected_sample_count,
47
+ audio_required: !!audio_required,
48
+ source_snapshot: immutable_snapshot(source_snapshot)
49
+ )
50
+ end
51
+
52
+ private
53
+
54
+ def immutable_spans(value, label)
55
+ raise ArgumentError, "#{label} must be an array" unless value.is_a?(Array)
56
+
57
+ value.map do |span|
58
+ raise ArgumentError, "#{label} contains an invalid span" unless
59
+ span.is_a?(Array) && span.length == 2 && span.all?(Numeric)
60
+
61
+ [span.fetch(0).to_f, span.fetch(1).to_f].freeze
62
+ end.freeze
63
+ end
64
+
65
+ def immutable_optional_string(value)
66
+ return nil if value.nil?
67
+ raise ArgumentError, "word-pipeline metadata strings must be strings" unless value.is_a?(String)
68
+
69
+ value.dup.freeze
70
+ end
71
+
72
+ def immutable_snapshot(value)
73
+ value.is_a?(Array) ? value.dup.freeze : value
74
+ end
75
+ end
76
+
77
+ Final = Data.define(:index, :result) do
78
+ def initialize(index:, result:)
79
+ index = Integer(index)
80
+ raise ArgumentError, "word-pipeline index must be nonnegative" if index.negative?
81
+
82
+ super
83
+ end
84
+ end
85
+
86
+ ReloadResult = Data.define(:decoded, :error) do
87
+ def ok?
88
+ error.nil?
89
+ end
90
+ end
91
+
92
+ # Executes an ordered two-phase batch. The ASR callback returns either
93
+ # AlignmentWork or Final. Alignment errors intended to be per-file must
94
+ # likewise be converted to a final result by the alignment callback;
95
+ # fatal/cancellation exceptions intentionally cross this boundary.
96
+ class Coordinator
97
+ def initialize(total:, evict_asr:, align:, fixed_results: {}, completed: nil, gc_interval: 8,
98
+ reload: nil, memory_byte_limit: nil, start_alignment: nil)
99
+ @total = Integer(total)
100
+ raise ArgumentError, "word-pipeline total must be nonnegative" if @total.negative?
101
+ raise ArgumentError, "evict_asr must respond to call" unless evict_asr.respond_to?(:call)
102
+ raise ArgumentError, "align must respond to call" unless align.respond_to?(:call)
103
+ raise ArgumentError, "completed must respond to call" if completed && !completed.respond_to?(:call)
104
+ raise ArgumentError, "start_alignment must respond to call" if
105
+ start_alignment && !start_alignment.respond_to?(:call)
106
+
107
+ @fixed_results = fixed_results.to_h.freeze
108
+ @evict_asr = evict_asr
109
+ @align = align
110
+ @completed = completed
111
+ @gc_interval = Integer(gc_interval)
112
+ raise ArgumentError, "gc_interval must be positive" unless @gc_interval.positive?
113
+ raise ArgumentError, "reload and memory_byte_limit must be provided together" unless
114
+ reload.nil? == memory_byte_limit.nil?
115
+ raise ArgumentError, "reload must respond to call" if reload && !reload.respond_to?(:call)
116
+
117
+ @reload = reload
118
+ @start_alignment = start_alignment
119
+ @memory_byte_limit = Integer(memory_byte_limit) if memory_byte_limit
120
+ raise ArgumentError, "memory_byte_limit must be positive" if
121
+ @memory_byte_limit && !@memory_byte_limit.positive?
122
+ end
123
+
124
+ def run(prepared, &asr)
125
+ raise ArgumentError, "an ASR callback is required" unless asr
126
+
127
+ slots = Array.new(@total, UNSET)
128
+ @fixed_results.each do |index, result|
129
+ outcome = if result.is_a?(AlignmentWork) || result.is_a?(Final)
130
+ result
131
+ else
132
+ Final.new(index: index, result: result)
133
+ end
134
+ install!(slots, outcome)
135
+ end
136
+ transitioned = false
137
+ begin
138
+ collect_asr!(slots, prepared, &asr)
139
+ missing = slots.each_index.select { |index| slots.fetch(index).equal?(UNSET) }
140
+ unless missing.empty?
141
+ raise TranscriptionRuntimeError,
142
+ "Word ASR phase omitted input index(es): #{missing.join(", ")}"
143
+ end
144
+
145
+ # Set the flag before calling so an eviction exception is not
146
+ # followed by a second, potentially destructive close attempt.
147
+ transitioned = true
148
+ alignment_required = slots.grep(AlignmentWork).any?(&:audio_required)
149
+ @evict_asr.call if alignment_required
150
+
151
+ align_slots!(slots)
152
+ completed_count = 0
153
+ slots.each_with_index do |outcome, index|
154
+ result = outcome.is_a?(Final) ? outcome.result : outcome
155
+ raise TranscriptionRuntimeError, "Word pipeline returned no result for input #{index}" if result.nil?
156
+
157
+ slots[index] = result
158
+ completed_count += 1
159
+ @completed&.call(index, completed_count, @total, result)
160
+ GC.start(full_mark: false, immediate_mark: true, immediate_sweep: true) if
161
+ (completed_count % @gc_interval).zero?
162
+ end
163
+ slots.freeze
164
+ ensure
165
+ @evict_asr.call unless transitioned
166
+ end
167
+ end
168
+
169
+ private
170
+
171
+ # Keeping ASR collection in its own frame guarantees that the final
172
+ # PreparedEntry (and its PCM) is unreachable before alignment starts.
173
+ def collect_asr!(slots, prepared)
174
+ prepared.each { |prepared_entry| install!(slots, yield(prepared_entry)) }
175
+ end
176
+
177
+ def align_slots!(slots)
178
+ works = slots.grep(AlignmentWork)
179
+ if @reload
180
+ PairBudgetReload.new(
181
+ works,
182
+ memory_byte_limit: @memory_byte_limit,
183
+ reload: @reload,
184
+ started: -> { @start_alignment&.call(works) }
185
+ ).each do |work, reload_result|
186
+ slots[work.index] = alignment_result(work, reload_result)
187
+ end
188
+ else
189
+ works.each { |work| slots[work.index] = alignment_result(work) }
190
+ end
191
+ end
192
+
193
+ def alignment_result(work, reload_result = UNSET)
194
+ aligned = if reload_result.equal?(UNSET)
195
+ @align.call(work)
196
+ else
197
+ @align.call(work, reload_result)
198
+ end
199
+ if aligned.is_a?(AlignmentWork) || aligned.is_a?(Final)
200
+ raise TranscriptionRuntimeError,
201
+ "Word alignment returned a scheduler outcome instead of a result"
202
+ end
203
+ aligned
204
+ end
205
+
206
+ def install!(slots, outcome)
207
+ unless outcome.is_a?(AlignmentWork) || outcome.is_a?(Final)
208
+ raise TranscriptionRuntimeError,
209
+ "Word ASR phase must return AlignmentWork or Final, got #{outcome.class}"
210
+ end
211
+ index = outcome.index
212
+ unless index.between?(0, @total - 1)
213
+ raise TranscriptionRuntimeError,
214
+ "Word ASR phase returned out-of-range input index #{index}"
215
+ end
216
+ raise TranscriptionRuntimeError, "Word ASR phase returned input #{index} more than once" unless
217
+ slots.fetch(index).equal?(UNSET)
218
+
219
+ slots[index] = outcome
220
+ end
221
+ end
222
+
223
+ # Reloads alignment PCM with one worker and at most one lookahead. The
224
+ # next file may overlap MMS computation only when the current/next PCM
225
+ # pair fits the configured budget; otherwise the current reference is
226
+ # dropped before the next decode starts.
227
+ class PairBudgetReload
228
+ SAMPLE_BYTES = 4
229
+
230
+ def initialize(works, memory_byte_limit:, reload:, started: nil)
231
+ @works = works.to_a.freeze
232
+ @memory_byte_limit = Integer(memory_byte_limit)
233
+ raise ArgumentError, "memory_byte_limit must be positive" unless @memory_byte_limit.positive?
234
+ raise ArgumentError, "reload must respond to call" unless reload.respond_to?(:call)
235
+ raise ArgumentError, "started must respond to call" if started && !started.respond_to?(:call)
236
+
237
+ @reload = reload
238
+ @started = started
239
+ end
240
+
241
+ def each
242
+ return enum_for(__method__) unless block_given?
243
+ return if @works.empty?
244
+
245
+ pending = submit(@works.first, 0)
246
+ @started&.call
247
+ @works.each_with_index do |work, index|
248
+ current = pending
249
+ pending = nil
250
+ reload_result = current.value
251
+ current = nil # rubocop:disable Lint/UselessAssignment -- release Thread#value before a serial decode
252
+
253
+ next_work = @works[index + 1]
254
+ pending = submit(next_work, index + 1) if next_work && pair_fits?(work, next_work)
255
+ yield [work, reload_result].freeze
256
+ reload_result = nil # rubocop:disable Lint/UselessAssignment -- release PCM before a serial decode
257
+ unless pending || next_work.nil?
258
+ GC.start(full_mark: false, immediate_mark: true, immediate_sweep: true)
259
+ pending = submit(next_work, index + 1)
260
+ end
261
+ end
262
+ completed = true
263
+ ensure
264
+ cancel(pending) if defined?(pending) && pending
265
+ cancel(current) if defined?(current) && current
266
+ GC.start(full_mark: false, immediate_mark: true, immediate_sweep: true) if defined?(completed) && completed
267
+ end
268
+
269
+ private
270
+
271
+ def pair_fits?(left, right)
272
+ pcm_bytes(left) + pcm_bytes(right) <= @memory_byte_limit
273
+ end
274
+
275
+ def pcm_bytes(work)
276
+ work.audio_required ? work.expected_sample_count * SAMPLE_BYTES : 0
277
+ end
278
+
279
+ def submit(work, index)
280
+ Thread.new do
281
+ Thread.current.name = "cohere-alignment-reload-#{index}" if Thread.current.respond_to?(:name=)
282
+ begin
283
+ ReloadResult.new(decoded: @reload.call(work), error: nil)
284
+ rescue StandardError => e
285
+ ReloadResult.new(decoded: nil, error: e)
286
+ end
287
+ end.tap { |thread| thread.report_on_exception = false }
288
+ end
289
+
290
+ def cancel(thread)
291
+ begin
292
+ @reload.cancel if @reload.respond_to?(:cancel)
293
+ rescue StandardError
294
+ nil
295
+ end
296
+ thread.kill if thread.alive?
297
+ thread.join
298
+ rescue Exception # rubocop:disable Lint/RescueException -- preserve caller cancellation/failure
299
+ nil
300
+ end
301
+ end
302
+
303
+ # Re-decodes one alignment candidate with the concrete backend recorded
304
+ # during ASR/checkpointing. It rejects automatic fallback and sample
305
+ # drift, which makes the second decode reproducible and source-consistent.
306
+ class ConcreteRedecoder
307
+ def initialize(decoder:, sample_rate:, memory_byte_limit:, validate: nil, skip: nil)
308
+ @decoder = decoder
309
+ @sample_rate = Integer(sample_rate)
310
+ @memory_byte_limit = Integer(memory_byte_limit)
311
+ raise ArgumentError, "sample_rate must be positive" unless @sample_rate.positive?
312
+ raise ArgumentError, "memory_byte_limit must be positive" unless @memory_byte_limit.positive?
313
+ raise ArgumentError, "validate must respond to call" if validate && !validate.respond_to?(:call)
314
+ raise ArgumentError, "skip must respond to call" if skip && !skip.respond_to?(:call)
315
+
316
+ @validate = validate
317
+ @skip = skip
318
+ end
319
+
320
+ def call(work)
321
+ return nil if @skip&.call(work)
322
+
323
+ @validate&.call(work)
324
+ return nil unless work.audio_required
325
+
326
+ backend = work.decode_backend
327
+ if backend.nil? || backend.empty? || backend == "auto"
328
+ raise TranscriptionRuntimeError,
329
+ "Cannot re-decode #{work.entry.path} for word alignment without a concrete ASR backend"
330
+ end
331
+
332
+ decoded = @decoder.decode(
333
+ work.entry.path,
334
+ backend: backend,
335
+ sample_rate: @sample_rate,
336
+ max_decoded_bytes: @memory_byte_limit
337
+ )
338
+ if decoded.backend != backend
339
+ raise TranscriptionRuntimeError,
340
+ "Alignment decoder backend changed for #{work.entry.path}: " \
341
+ "#{backend} -> #{decoded.backend}"
342
+ end
343
+ unless decoded.samples.length == work.expected_sample_count
344
+ raise TranscriptionRuntimeError,
345
+ "Decoded sample count changed between ASR and alignment for #{work.entry.path}: " \
346
+ "#{decoded.samples.length} != #{work.expected_sample_count}"
347
+ end
348
+
349
+ @validate&.call(work)
350
+ decoded
351
+ end
352
+
353
+ def cancel
354
+ Audio::FFmpegNative.cancel_active! if
355
+ defined?(Audio::FFmpegNative) && Audio::FFmpegNative.respond_to?(:cancel_active!)
356
+ nil
357
+ rescue StandardError
358
+ nil
359
+ end
360
+ end
361
+ end
362
+ end
363
+ end
364
+ end